Tutorial - Sentiment.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Movie Reviews Dataset\n",
    "* Input patterns: movie reviews in natural language.\n",
    "* Labels: sentiment analysis of the text\n",
    "    * Classes: *negative, positive*\n",
    "* Ground truth: What the person thought when writing the review\n",
    "* Annotations: re-collected from Amazon Mechanical Turk (AMT).\n",
    "    * Number of annotators (T) = 203  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import keras, time, sys, os, gc\n",
    "\n",
    "DTYPE_OP = 'float32'\n",
    "keras.backend.set_floatx(DTYPE_OP)\n",
    "\n",
    "if DTYPE_OP == 'float64':\n",
    "    keras.backend.set_epsilon(np.finfo(np.float64).eps)\n",
    "elif DTYPE_OP == 'float32':\n",
    "    keras.backend.set_epsilon(np.finfo(np.float32).eps)\n",
    "    \n",
    "np.set_printoptions(precision=4, suppress=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "#GLOBAL Variables\n",
    "BATCH_SIZE = 128 #rodrigues = BATCH_SIZE = 64\n",
    "EPOCHS_BASE = 50\n",
    "OPT = 'adam' #optimizer for neural network \n",
    "TOL = 3e-2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "folder = \"./data/Sentiment/\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load Data and preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num. train texts: 4999\n",
      "Num. test texts:  3999\n"
     ]
    }
   ],
   "source": [
    "from codeE.utils import read_texts\n",
    "texts_train = read_texts(folder+\"Sentiment_text_train.txt\")\n",
    "texts_test  = read_texts(folder+\"Sentiment_text_test.txt\")\n",
    "\n",
    "Z_train = np.loadtxt(folder+\"Sentiment_Z_train.txt\", dtype='int')\n",
    "Z_test  = np.loadtxt(folder+\"Sentiment_Z_test.txt\", dtype='int')\n",
    "\n",
    "print(\"Num. train texts: %d\" % len(texts_train))\n",
    "print(\"Num. test texts:  %d\" % len(texts_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 14096 unique tokens.\n"
     ]
    }
   ],
   "source": [
    "from keras import preprocessing\n",
    "MAX_NB_WORDS = 14000\n",
    "tokenizer = preprocessing.text.Tokenizer(num_words=MAX_NB_WORDS)\n",
    "tokenizer.fit_on_texts(texts_train+texts_test)\n",
    "sequences_train = tokenizer.texts_to_sequences(texts_train)\n",
    "sequences_test = tokenizer.texts_to_sequences(texts_test)\n",
    "#MAX_NB_WORDS = len(tokenizer.word_index)\n",
    "print('Found %s unique tokens.' % len(tokenizer.word_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Real max:  51\n",
      "Used max:  50\n",
      "Shape of train tensor: (4999, 50)\n"
     ]
    }
   ],
   "source": [
    "lens = list(map(len,sequences_train))\n",
    "max_L = np.max(lens)\n",
    "print(\"Real max: \",max_L)\n",
    "max_L = 50 \n",
    "print(\"Used max: \",max_L)\n",
    "\n",
    "X_train = preprocessing.sequence.pad_sequences(sequences_train, maxlen=max_L,dtype='int32', value=0,padding='pre')\n",
    "X_test = preprocessing.sequence.pad_sequences(sequences_test, maxlen=max_L,dtype='int32', value=0,padding='pre')\n",
    "print('Shape of train tensor:', X_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Preparing embedding matrix.\n",
      "Words found on glove:  12935\n"
     ]
    }
   ],
   "source": [
    "EMBEDDING_DIM = 300\n",
    "GLOVE_FILE = \"../GLOVE/glove.6B.%dd.txt\"%(EMBEDDING_DIM)\n",
    "\n",
    "embeddings_index = {}\n",
    "with open(GLOVE_FILE) as file:\n",
    "    for line in file:\n",
    "        values = line.split()\n",
    "        coefs = np.asarray(values[1:], dtype='float32')\n",
    "        embeddings_index[values[0]] = coefs\n",
    "print('Preparing embedding matrix.')\n",
    "sorted_x = sorted(tokenizer.word_counts.items(), key=lambda kv: kv[1], reverse=True)\n",
    "vocab = {value[0]:tokenizer.word_index[value[0]] for i, value in enumerate(sorted_x) if i < MAX_NB_WORDS}\n",
    "embedding_matrix = np.zeros((len(vocab)+1, EMBEDDING_DIM))\n",
    "v=0\n",
    "for word, i in vocab.items():\n",
    "    embedding_vector = embeddings_index.get(word)\n",
    "    if embedding_vector is not None:\n",
    "        embedding_matrix[i] = embedding_vector # words not found in embedding index will be all-zeros.\n",
    "        v+=1\n",
    "del embeddings_index, sorted_x, tokenizer\n",
    "gc.collect()\n",
    "print(\"Words found on glove: \",v)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.countplot(Z_train)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load annotations "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading AMT data...\n",
      "Shape (data,annotators):  (4999, 203)\n",
      "Classes:  2\n"
     ]
    }
   ],
   "source": [
    "print(\"Loading AMT data...\")\n",
    "y_obs = np.loadtxt(folder+\"answers.txt\",dtype='int16') #not annotation symbol ==-1\n",
    "\n",
    "N,T = y_obs.shape\n",
    "print(\"Shape (data,annotators): \",(N,T))\n",
    "print(\"Classes: \", np.max(y_obs)+1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Minimum annotations in a example:  4\n",
      "Maximum annotations in a example:  10\n",
      "T data (expected value):  5.550310062012403\n",
      "Graphics of T weights\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Minimum annotations by an annotator:  5\n",
      "Maximum annotations by an annotator:  3993\n",
      "Mean annotations by an annotator:  136.67980295566502\n"
     ]
    }
   ],
   "source": [
    "N_ann = np.sum(y_obs != -1,axis=1) \n",
    "\n",
    "sns.countplot(N_ann)\n",
    "plt.show()\n",
    "print(\"Minimum annotations in a example: \",N_ann.min())\n",
    "print(\"Maximum annotations in a example: \",N_ann.max())\n",
    "print(\"T data (expected value): \",N_ann.mean())\n",
    "\n",
    "T_weights = np.sum(y_obs != -1,axis=0) #number of annotations per annotator\n",
    "print(\"Graphics of T weights\")\n",
    "plt.hist(T_weights)\n",
    "plt.show()\n",
    "\n",
    "plt.boxplot(T_weights)\n",
    "plt.show()\n",
    "print(\"Minimum annotations by an annotator: \",T_weights.min())\n",
    "print(\"Maximum annotations by an annotator: \",T_weights.max())\n",
    "print(\"Mean annotations by an annotator: \",T_weights.mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Upper Bound Model -- ideal"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "#### Delta Convergence criteria\n",
    "from codeE.learning_models import default_RNN_text, Clonable_Model\n",
    "from codeE.utils import EarlyStopRelative\n",
    "ourCallback = EarlyStopRelative(monitor='loss',patience=1,min_delta=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:190: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:197: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:3733: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/casapanshop/anaconda2/envs/newpy3_tf1/lib/python3.6/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
      "Trained IDeal Model, Epochs to converge = 40\n"
     ]
    }
   ],
   "source": [
    "Z_train_onehot = keras.utils.to_categorical(Z_train)\n",
    "\n",
    "model_UB = default_RNN_text(max_L, Z_train_onehot.shape[1], embed_M=embedding_matrix)\n",
    "model_UB.compile(loss='categorical_crossentropy',optimizer=OPT)\n",
    "hist = model_UB.fit(X_train,Z_train_onehot,epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])\n",
    "print(\"Trained IDeal Model, Epochs to converge =\",len(hist.epoch))\n",
    "clone_UB = Clonable_Model(model_UB)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fe67e0ce588>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(hist.history[\"loss\"],label=\"train\")\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.991998399679936\n",
      "Test accuracy of model = 0.7271817954488622\n"
     ]
    }
   ],
   "source": [
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(model_UB, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(model_UB, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate Matrix representation of both scenarios"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual representation shape (N,T,K)=  (4999, 203, 2)\n",
      "Global representation shape (N,K)=  (4999, 2)\n"
     ]
    }
   ],
   "source": [
    "from codeE.representation import set_representation\n",
    "\n",
    "y_obs_categorical = set_representation(y_obs,'onehot') \n",
    "print(\"Individual representation shape (N,T,K)= \",y_obs_categorical.shape)\n",
    "\n",
    "r_obs = set_representation(y_obs,\"global\")\n",
    "print(\"Global representation shape (N,K)= \",r_obs.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0, 5],\n",
       "       [5, 0],\n",
       "       [4, 2],\n",
       "       ...,\n",
       "       [2, 4],\n",
       "       [1, 4],\n",
       "       [3, 2]], dtype=int32)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "r_obs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### Real Confusion matrices\n",
    "Generate the individual and global confusion matrices based on the sampled annotations."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual matrix shape (T,K,K)=  (203, 2, 2)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[[0.7059, 0.2941],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.9483, 0.0517],\n",
       "        [0.1373, 0.8627]],\n",
       "\n",
       "       [[0.9756, 0.0244],\n",
       "        [0.1053, 0.8947]],\n",
       "\n",
       "       [[0.9306, 0.0694],\n",
       "        [0.0763, 0.9237]],\n",
       "\n",
       "       [[0.9401, 0.0599],\n",
       "        [0.2284, 0.7716]],\n",
       "\n",
       "       [[0.6469, 0.3531],\n",
       "        [0.1608, 0.8392]],\n",
       "\n",
       "       [[0.9344, 0.0656],\n",
       "        [0.0792, 0.9208]],\n",
       "\n",
       "       [[0.8919, 0.1081],\n",
       "        [0.1127, 0.8873]],\n",
       "\n",
       "       [[0.625 , 0.375 ],\n",
       "        [0.381 , 0.619 ]],\n",
       "\n",
       "       [[0.8139, 0.1861],\n",
       "        [0.1332, 0.8668]],\n",
       "\n",
       "       [[0.8372, 0.1628],\n",
       "        [0.2289, 0.7711]],\n",
       "\n",
       "       [[0.8983, 0.1017],\n",
       "        [0.06  , 0.94  ]],\n",
       "\n",
       "       [[0.9123, 0.0877],\n",
       "        [0.1466, 0.8534]],\n",
       "\n",
       "       [[0.9301, 0.0699],\n",
       "        [0.1252, 0.8748]],\n",
       "\n",
       "       [[0.7703, 0.2297],\n",
       "        [0.1429, 0.8571]],\n",
       "\n",
       "       [[0.9327, 0.0673],\n",
       "        [0.1178, 0.8822]],\n",
       "\n",
       "       [[0.9286, 0.0714],\n",
       "        [0.0909, 0.9091]],\n",
       "\n",
       "       [[0.9383, 0.0617],\n",
       "        [0.0678, 0.9322]],\n",
       "\n",
       "       [[0.4615, 0.5385],\n",
       "        [0.25  , 0.75  ]],\n",
       "\n",
       "       [[0.7164, 0.2836],\n",
       "        [0.1176, 0.8824]],\n",
       "\n",
       "       [[0.7436, 0.2564],\n",
       "        [0.0779, 0.9221]],\n",
       "\n",
       "       [[0.7917, 0.2083],\n",
       "        [0.3048, 0.6952]],\n",
       "\n",
       "       [[0.7347, 0.2653],\n",
       "        [0.3084, 0.6916]],\n",
       "\n",
       "       [[0.375 , 0.625 ],\n",
       "        [0.2857, 0.7143]],\n",
       "\n",
       "       [[0.8393, 0.1607],\n",
       "        [0.0847, 0.9153]],\n",
       "\n",
       "       [[0.7292, 0.2708],\n",
       "        [0.2097, 0.7903]],\n",
       "\n",
       "       [[0.9481, 0.0519],\n",
       "        [0.1029, 0.8971]],\n",
       "\n",
       "       [[0.754 , 0.246 ],\n",
       "        [0.112 , 0.888 ]],\n",
       "\n",
       "       [[0.9167, 0.0833],\n",
       "        [0.0769, 0.9231]],\n",
       "\n",
       "       [[0.5714, 0.4286],\n",
       "        [0.1875, 0.8125]],\n",
       "\n",
       "       [[0.7742, 0.2258],\n",
       "        [0.069 , 0.931 ]],\n",
       "\n",
       "       [[0.7444, 0.2556],\n",
       "        [0.2095, 0.7905]],\n",
       "\n",
       "       [[0.92  , 0.08  ],\n",
       "        [0.25  , 0.75  ]],\n",
       "\n",
       "       [[0.8158, 0.1842],\n",
       "        [0.1622, 0.8378]],\n",
       "\n",
       "       [[0.3864, 0.6136],\n",
       "        [0.1839, 0.8161]],\n",
       "\n",
       "       [[0.4882, 0.5118],\n",
       "        [0.1429, 0.8571]],\n",
       "\n",
       "       [[0.6944, 0.3056],\n",
       "        [0.1471, 0.8529]],\n",
       "\n",
       "       [[0.913 , 0.087 ],\n",
       "        [0.0909, 0.9091]],\n",
       "\n",
       "       [[0.9478, 0.0522],\n",
       "        [0.0574, 0.9426]],\n",
       "\n",
       "       [[0.9032, 0.0968],\n",
       "        [0.1034, 0.8966]],\n",
       "\n",
       "       [[0.9394, 0.0606],\n",
       "        [0.0909, 0.9091]],\n",
       "\n",
       "       [[0.9091, 0.0909],\n",
       "        [0.2105, 0.7895]],\n",
       "\n",
       "       [[0.375 , 0.625 ],\n",
       "        [0.1429, 0.8571]],\n",
       "\n",
       "       [[0.8571, 0.1429],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.7222, 0.2778],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.9125, 0.0875],\n",
       "        [0.1286, 0.8714]],\n",
       "\n",
       "       [[0.6471, 0.3529],\n",
       "        [0.2222, 0.7778]],\n",
       "\n",
       "       [[0.9091, 0.0909],\n",
       "        [0.2857, 0.7143]],\n",
       "\n",
       "       [[0.8889, 0.1111],\n",
       "        [0.1084, 0.8916]],\n",
       "\n",
       "       [[0.8875, 0.1125],\n",
       "        [0.15  , 0.85  ]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.28  , 0.72  ]],\n",
       "\n",
       "       [[0.9032, 0.0968],\n",
       "        [0.2647, 0.7353]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.8571, 0.1429],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.7143, 0.2857],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.7333, 0.2667],\n",
       "        [0.25  , 0.75  ]],\n",
       "\n",
       "       [[0.3636, 0.6364],\n",
       "        [0.375 , 0.625 ]],\n",
       "\n",
       "       [[0.6154, 0.3846],\n",
       "        [0.1364, 0.8636]],\n",
       "\n",
       "       [[0.9231, 0.0769],\n",
       "        [0.2727, 0.7273]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.6   , 0.4   ],\n",
       "        [0.4   , 0.6   ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.1579, 0.8421]],\n",
       "\n",
       "       [[0.72  , 0.28  ],\n",
       "        [0.2667, 0.7333]],\n",
       "\n",
       "       [[0.2222, 0.7778],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.0714, 0.9286]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [0.5   , 0.5   ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.6667, 0.3333]],\n",
       "\n",
       "       [[0.7143, 0.2857],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.25  , 0.75  ],\n",
       "        [0.1429, 0.8571]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.25  , 0.75  ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.674 , 0.326 ],\n",
       "        [0.1571, 0.8429]],\n",
       "\n",
       "       [[0.8529, 0.1471],\n",
       "        [0.2836, 0.7164]],\n",
       "\n",
       "       [[0.7647, 0.2353],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.75  , 0.25  ],\n",
       "        [0.1667, 0.8333]],\n",
       "\n",
       "       [[0.963 , 0.037 ],\n",
       "        [0.1429, 0.8571]],\n",
       "\n",
       "       [[0.4524, 0.5476],\n",
       "        [0.1772, 0.8228]],\n",
       "\n",
       "       [[0.8777, 0.1223],\n",
       "        [0.0541, 0.9459]],\n",
       "\n",
       "       [[0.471 , 0.529 ],\n",
       "        [0.2368, 0.7632]],\n",
       "\n",
       "       [[0.8841, 0.1159],\n",
       "        [0.1207, 0.8793]],\n",
       "\n",
       "       [[0.9122, 0.0878],\n",
       "        [0.1338, 0.8662]],\n",
       "\n",
       "       [[0.8276, 0.1724],\n",
       "        [0.0769, 0.9231]],\n",
       "\n",
       "       [[0.8878, 0.1122],\n",
       "        [0.1609, 0.8391]],\n",
       "\n",
       "       [[0.3192, 0.6808],\n",
       "        [0.2083, 0.7917]],\n",
       "\n",
       "       [[0.877 , 0.123 ],\n",
       "        [0.1525, 0.8475]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.0667, 0.9333]],\n",
       "\n",
       "       [[0.7439, 0.2561],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.9181, 0.0819],\n",
       "        [0.0798, 0.9202]],\n",
       "\n",
       "       [[0.8987, 0.1013],\n",
       "        [0.1   , 0.9   ]],\n",
       "\n",
       "       [[0.7887, 0.2113],\n",
       "        [0.1014, 0.8986]],\n",
       "\n",
       "       [[0.5102, 0.4898],\n",
       "        [0.2292, 0.7708]],\n",
       "\n",
       "       [[0.8667, 0.1333],\n",
       "        [0.1   , 0.9   ]],\n",
       "\n",
       "       [[0.9062, 0.0938],\n",
       "        [0.0741, 0.9259]],\n",
       "\n",
       "       [[0.9308, 0.0692],\n",
       "        [0.1187, 0.8813]],\n",
       "\n",
       "       [[0.9739, 0.0261],\n",
       "        [0.0837, 0.9163]],\n",
       "\n",
       "       [[0.633 , 0.367 ],\n",
       "        [0.1354, 0.8646]],\n",
       "\n",
       "       [[0.9574, 0.0426],\n",
       "        [0.0407, 0.9593]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.9444, 0.0556],\n",
       "        [0.3529, 0.6471]],\n",
       "\n",
       "       [[0.875 , 0.125 ],\n",
       "        [0.087 , 0.913 ]],\n",
       "\n",
       "       [[0.7015, 0.2985],\n",
       "        [0.1528, 0.8472]],\n",
       "\n",
       "       [[0.9643, 0.0357],\n",
       "        [0.1875, 0.8125]],\n",
       "\n",
       "       [[0.8254, 0.1746],\n",
       "        [0.0351, 0.9649]],\n",
       "\n",
       "       [[0.96  , 0.04  ],\n",
       "        [0.1143, 0.8857]],\n",
       "\n",
       "       [[0.9383, 0.0617],\n",
       "        [0.1186, 0.8814]],\n",
       "\n",
       "       [[0.8367, 0.1633],\n",
       "        [0.0714, 0.9286]],\n",
       "\n",
       "       [[0.8913, 0.1087],\n",
       "        [0.0794, 0.9206]],\n",
       "\n",
       "       [[0.3766, 0.6234],\n",
       "        [0.3864, 0.6136]],\n",
       "\n",
       "       [[0.7969, 0.2031],\n",
       "        [0.0893, 0.9107]],\n",
       "\n",
       "       [[0.9167, 0.0833],\n",
       "        [0.2308, 0.7692]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.7941, 0.2059],\n",
       "        [0.4286, 0.5714]],\n",
       "\n",
       "       [[0.9   , 0.1   ],\n",
       "        [0.0678, 0.9322]],\n",
       "\n",
       "       [[0.5263, 0.4737],\n",
       "        [0.4375, 0.5625]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.125 , 0.875 ]],\n",
       "\n",
       "       [[0.9038, 0.0962],\n",
       "        [0.1235, 0.8765]],\n",
       "\n",
       "       [[0.9483, 0.0517],\n",
       "        [0.2131, 0.7869]],\n",
       "\n",
       "       [[0.8571, 0.1429],\n",
       "        [0.125 , 0.875 ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[0.6579, 0.3421],\n",
       "        [0.1622, 0.8378]],\n",
       "\n",
       "       [[0.8889, 0.1111],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.875 , 0.125 ],\n",
       "        [0.2195, 0.7805]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.9767, 0.0233],\n",
       "        [0.1081, 0.8919]],\n",
       "\n",
       "       [[0.9474, 0.0526],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.8636, 0.1364],\n",
       "        [0.0714, 0.9286]],\n",
       "\n",
       "       [[0.8788, 0.1212],\n",
       "        [0.2857, 0.7143]],\n",
       "\n",
       "       [[0.75  , 0.25  ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.0714, 0.9286]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.1667, 0.8333]],\n",
       "\n",
       "       [[0.7692, 0.2308],\n",
       "        [0.0909, 0.9091]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.8333, 0.1667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.8333, 0.1667],\n",
       "        [0.1875, 0.8125]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.7885, 0.2115],\n",
       "        [0.1061, 0.8939]],\n",
       "\n",
       "       [[0.8857, 0.1143],\n",
       "        [0.12  , 0.88  ]],\n",
       "\n",
       "       [[0.375 , 0.625 ],\n",
       "        [0.2857, 0.7143]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.4   , 0.6   ]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.2222, 0.7778]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.1905, 0.8095]],\n",
       "\n",
       "       [[0.7568, 0.2432],\n",
       "        [0.3478, 0.6522]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.2857, 0.7143]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[0.8   , 0.2   ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.5   , 0.5   ]],\n",
       "\n",
       "       [[0.5714, 0.4286],\n",
       "        [0.5   , 0.5   ]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.1111, 0.8889]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.5   , 0.5   ]],\n",
       "\n",
       "       [[0.4   , 0.6   ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.8333, 0.1667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.25  , 0.75  ],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.5   , 0.5   ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.6   , 0.4   ],\n",
       "        [0.2   , 0.8   ]],\n",
       "\n",
       "       [[0.25  , 0.75  ],\n",
       "        [1.    , 0.    ]],\n",
       "\n",
       "       [[0.6   , 0.4   ],\n",
       "        [0.5   , 0.5   ]],\n",
       "\n",
       "       [[1.    , 0.    ],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.25  , 0.75  ]],\n",
       "\n",
       "       [[0.3333, 0.6667],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.    , 1.    ],\n",
       "        [0.3333, 0.6667]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.6667, 0.3333],\n",
       "        [0.    , 1.    ]],\n",
       "\n",
       "       [[0.4913, 0.5087],\n",
       "        [0.5183, 0.4817]],\n",
       "\n",
       "       [[0.5942, 0.4058],\n",
       "        [0.3934, 0.6066]],\n",
       "\n",
       "       [[0.5417, 0.4583],\n",
       "        [0.6265, 0.3735]],\n",
       "\n",
       "       [[0.506 , 0.494 ],\n",
       "        [0.4098, 0.5902]],\n",
       "\n",
       "       [[0.5873, 0.4127],\n",
       "        [0.4464, 0.5536]],\n",
       "\n",
       "       [[0.5424, 0.4576],\n",
       "        [0.4937, 0.5063]],\n",
       "\n",
       "       [[0.6533, 0.3467],\n",
       "        [0.3571, 0.6429]],\n",
       "\n",
       "       [[0.4194, 0.5806],\n",
       "        [0.6133, 0.3867]],\n",
       "\n",
       "       [[0.8837, 0.1163],\n",
       "        [0.0845, 0.9155]],\n",
       "\n",
       "       [[0.9367, 0.0633],\n",
       "        [0.0521, 0.9479]]], dtype=float32)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.utils import generate_Individual_conf\n",
    "\n",
    "conf_matrix_I = generate_Individual_conf(Z_train, y_obs_categorical)\n",
    "print(\"Individual matrix shape (T,K,K)= \",conf_matrix_I.shape)\n",
    "conf_matrix_I"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global matrix shape (K,K)=  (2, 2)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from codeE.utils import generate_Global_conf\n",
    "\n",
    "#confe_matrix_G = generate_Global_conf(Z_train, y_obs_categorical)\n",
    "confe_matrix_G = generate_Global_conf(Z_train, r_obs)\n",
    "print(\"Global matrix shape (K,K)= \",confe_matrix_G.shape)\n",
    "\n",
    "from codeE.utils import plot_confusion_matrix\n",
    "plot_confusion_matrix(confe_matrix_G, [\"negative\", \"positive\"], title= \"Global Matrix\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Aggregation methods (without predictive model)\n",
    "* Majority Voting (MV)\n",
    "* Weighted Majority Voting\n",
    "* Dawid and Skene (DS): with inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1, 0, 0, ..., 1, 1, 0])"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import LabelAgg\n",
    "label_A = LabelAgg(scenario=\"global\")\n",
    "\n",
    "mv_soft = label_A.infer(  r_obs, 'softMV')\n",
    "mv_hard = label_A.predict(r_obs, 'hardMV')\n",
    "mv_hard"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.    , 1.    ],\n",
       "       [1.    , 0.    ],\n",
       "       [0.6667, 0.3333],\n",
       "       ...,\n",
       "       [0.3333, 0.6667],\n",
       "       [0.2   , 0.8   ],\n",
       "       [0.6   , 0.4   ]], dtype=float32)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import LabelAgg\n",
    "label_A = LabelAgg(scenario=\"individual\")\n",
    "\n",
    "mv_soft = label_A.infer(  y_obs_categorical, 'softMV')\n",
    "mv_hard = label_A.predict(y_obs_categorical, 'hardMV')\n",
    "mv_soft"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ACC MV on train: 0.8895779155831166\n"
     ]
    }
   ],
   "source": [
    "print(\"ACC MV on train:\",np.mean(mv_hard==Z_train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### generate confusion matrix to understand MV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from codeE.utils import generate_confusionM\n",
    "prob_Yz = generate_confusionM(mv_hard, r_obs) #confusion matrix of all annotators\n",
    "plot_confusion_matrix(prob_Yz, title= \"hard-MV confusion matrix\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Metrics to compare the estimation of confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from codeE.utils import compare_conf_ma\n",
    "compare_conf_ma(prob_Yz, true_conf_mat=confe_matrix_G, text=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.0018885730640643085\n",
      "Global D (based on normalized Frobenius) = 0.03987046331167221\n"
     ]
    }
   ],
   "source": [
    "from codeE.evaluation import D_JS, D_KL, D_NormF\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Weighted"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.    , 1.    ],\n",
       "       [1.    , 0.    ],\n",
       "       [0.9575, 0.0425],\n",
       "       ...,\n",
       "       [0.4652, 0.5348],\n",
       "       [0.0954, 0.9046],\n",
       "       [0.4188, 0.5812]], dtype=float32)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import LabelAgg\n",
    "label_A = LabelAgg(scenario=\"individual\")\n",
    "\n",
    "Wmv_soft = label_A.infer(y_obs_categorical, 'softMV', weights=T_weights)\n",
    "Wmv_soft"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ACC MV on train: 0.7631526305261053\n"
     ]
    }
   ],
   "source": [
    "print(\"ACC MV on train:\",np.mean(Wmv_soft.argmax(axis=-1)==Z_train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Dawid and Skene model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Z marginal shape (2,)\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -14653.156\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13799.342\tTol1: 0.05827\tTol2: 0.25907\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13743.929\tTol1: 0.00402\tTol2: 0.13517\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13734.473\tTol1: 0.00069\tTol2: 0.09663\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13731.723\tTol1: 0.00020\tTol2: 0.06555\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13730.729\tTol1: 0.00007\tTol2: 0.04730\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13730.295\tTol1: 0.00003\tTol2: 0.03600\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.01 sec)\tlogL: -13730.066\tTol1: 0.00002\tTol2: 0.02774\t\n",
      "Finished training\n"
     ]
    }
   ],
   "source": [
    "from codeE.methods import LabelInf_EM as DS\n",
    "\n",
    "DS_model = DS(init_Z='softmv') \n",
    "#DS_model = DS(init_Z='softmv', priors=\"None\", fast=False) \n",
    "    \n",
    "# if you want you can initialize step E forcing some method\n",
    "#DS_model.init_E(y_obs_categorical)\n",
    "#DS_model.get_qestimation()\n",
    "\n",
    "hist = DS_model.fit(y_obs_categorical, max_iter=EPOCHS_BASE, tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "p(z) =  [0.5038 0.4962]\n",
      "ACC D&S on train: 0.9151830366073215\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.    , 1.    ],\n",
       "       [1.    , 0.    ],\n",
       "       [0.9578, 0.0422],\n",
       "       ...,\n",
       "       [0.    , 1.    ],\n",
       "       [0.    , 1.    ],\n",
       "       [0.991 , 0.009 ]], dtype=float32)"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(\"p(z) = \",DS_model.get_marginalZ())\n",
    "\n",
    "ds_labels = DS_model.infer() #could infer/predict on the trainint set only\n",
    "print(\"ACC D&S on train:\",np.mean(ds_labels.argmax(axis=1)==Z_train))\n",
    "\n",
    "ds_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Global Label Noise - Without predictive model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 0.00 sec)\tlogL: -16078.105\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.00 sec)\tlogL: -15815.982\tTol1: 0.01630\tTol2: 0.14435\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.00 sec)\tlogL: -15803.926\tTol1: 0.00076\tTol2: 0.02254\t\n",
      "Finished training!\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([-16078.1055, -15815.9824, -15803.9258])"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import LabelInf_EM_G as LabG_Noise\n",
    "\n",
    "LabG_Noise_model = LabG_Noise(init_Z='softmv') \n",
    "LabG_Noise_model.fit(r_obs, max_iter=EPOCHS_BASE,tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "p(z) =  [0.4881 0.5119]\n",
      "ACC on train: 0.8895779155831166\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.0008, 0.9992],\n",
       "       [0.9997, 0.0003],\n",
       "       [0.9775, 0.0225],\n",
       "       ...,\n",
       "       [0.0842, 0.9158],\n",
       "       [0.0171, 0.9829],\n",
       "       [0.8917, 0.1083]], dtype=float32)"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(\"p(z) = \", LabG_Noise_model.get_marginalZ())\n",
    "\n",
    "LabG_labels = LabG_Noise_model.infer() \n",
    "print(\"ACC on train:\",np.mean(LabG_labels.argmax(axis=1)==Z_train))\n",
    "\n",
    "LabG_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Second Phase\n",
    "train a predictive model over the inference/aggregation label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "from codeE.evaluation import accuracy_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trained model over soft-MV, Epochs to converge = 4\n",
      "Train accuracy of model = 0.7919583916783357\n",
      "Test accuracy of model = 0.7481870467616905\n"
     ]
    }
   ],
   "source": [
    "model_mvsoft = clone_UB.get_model()\n",
    "model_mvsoft.compile(loss='categorical_crossentropy',optimizer=OPT)\n",
    "hist=model_mvsoft.fit(X_train, mv_soft, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])\n",
    "print(\"Trained model over soft-MV, Epochs to converge =\",len(hist.epoch))\n",
    "\n",
    "print(\"Train accuracy of model =\", accuracy_model(model_mvsoft, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(model_mvsoft, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 1.],\n",
       "       [1., 0.],\n",
       "       [1., 0.],\n",
       "       ...,\n",
       "       [0., 1.],\n",
       "       [0., 1.],\n",
       "       [1., 0.]], dtype=float32)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mv_hard_onehot = keras.utils.to_categorical(mv_hard)\n",
    "mv_hard_onehot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trained model over hard-MV, Epochs to converge = 8\n",
      "Train accuracy of model = 0.8387677535507101\n",
      "Test accuracy of model = 0.7309327331832958\n"
     ]
    }
   ],
   "source": [
    "model_mvhard = clone_UB.get_model()\n",
    "model_mvhard.compile(loss='categorical_crossentropy',optimizer=OPT)\n",
    "hist=model_mvhard.fit(X_train, mv_hard_onehot, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])\n",
    "print(\"Trained model over hard-MV, Epochs to converge =\",len(hist.epoch))\n",
    "\n",
    "print(\"Train accuracy of model =\", accuracy_model(model_mvhard, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(model_mvhard, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trained model over D&S, Epochs to converge = 10\n",
      "Train accuracy of model = 0.884376875375075\n",
      "Test accuracy of model = 0.746436609152288\n"
     ]
    }
   ],
   "source": [
    "model_ds = clone_UB.get_model()\n",
    "model_ds.compile(loss='categorical_crossentropy',optimizer=OPT)\n",
    "hist=model_ds.fit(X_train, ds_labels, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])\n",
    "print(\"Trained model over D&S, Epochs to converge =\",len(hist.epoch))\n",
    "\n",
    "print(\"Train accuracy of model =\", accuracy_model(model_ds, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(model_ds, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trained model over Global LabelNoise, Epochs to converge = 19\n",
      "Train accuracy of model = 0.8867773554710943\n",
      "Test accuracy of model = 0.746686671667917\n"
     ]
    }
   ],
   "source": [
    "model_lg = clone_UB.get_model()\n",
    "model_lg.compile(loss='categorical_crossentropy',optimizer=OPT)\n",
    "hist=model_lg.fit(X_train, LabG_labels, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])\n",
    "print(\"Trained model over Global LabelNoise, Epochs to converge =\",len(hist.epoch))\n",
    "\n",
    "print(\"Train accuracy of model =\", accuracy_model(model_lg, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(model_lg, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Raykar Model (joint predict model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 300)           4200300   \n",
      "_________________________________________________________________\n",
      "cu_dnngru_1 (CuDNNGRU)       (None, 50, 128)           165120    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 50, 128)           0         \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (None, 64)                37248     \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 4,402,798\n",
      "Trainable params: 202,498\n",
      "Non-trainable params: 4,200,300\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_R = clone_UB.get_model()\n",
    "model_R.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.64 sec)\tlogL: -13937.993\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12598.732\tTol1: 0.09609\tTol2: 0.26592\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12296.549\tTol1: 0.02399\tTol2: 0.13869\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12243.105\tTol1: 0.00435\tTol2: 0.09371\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12006.040\tTol1: 0.01936\tTol2: 0.08064\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12080.297\tTol1: 0.00618\tTol2: 0.04411\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11911.111\tTol1: 0.01401\tTol2: 0.06805\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11528.218\tTol1: 0.03215\tTol2: 0.05102\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11436.942\tTol1: 0.00792\tTol2: 0.03409\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11167.068\tTol1: 0.02360\tTol2: 0.03307\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11049.994\tTol1: 0.01048\tTol2: 0.02697\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -13833.984\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12585.893\tTol1: 0.09022\tTol2: 0.26623\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12838.598\tTol1: 0.02008\tTol2: 0.13462\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12223.263\tTol1: 0.04793\tTol2: 0.12145\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12118.959\tTol1: 0.00853\tTol2: 0.07565\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11870.133\tTol1: 0.02053\tTol2: 0.06761\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11719.893\tTol1: 0.01266\tTol2: 0.03492\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11624.826\tTol1: 0.00811\tTol2: 0.03098\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11721.676\tTol1: 0.00833\tTol2: 0.03168\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11242.373\tTol1: 0.04089\tTol2: 0.05219\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11047.060\tTol1: 0.01737\tTol2: 0.02565\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.45 sec)\tlogL: -13906.287\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12563.078\tTol1: 0.09659\tTol2: 0.26504\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12333.841\tTol1: 0.01825\tTol2: 0.14302\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12222.180\tTol1: 0.00905\tTol2: 0.09574\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11995.277\tTol1: 0.01856\tTol2: 0.06264\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11886.434\tTol1: 0.00907\tTol2: 0.04780\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11798.132\tTol1: 0.00743\tTol2: 0.03466\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.60 sec)\tlogL: -11751.567\tTol1: 0.00395\tTol2: 0.04363\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11269.717\tTol1: 0.04100\tTol2: 0.07221\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11023.748\tTol1: 0.02183\tTol2: 0.03046\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -10948.387\tTol1: 0.00684\tTol2: 0.03082\t\n",
      "Iter 12/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11060.787\tTol1: 0.01027\tTol2: 0.02933\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.38 sec)\tlogL: -14029.521\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12703.245\tTol1: 0.09453\tTol2: 0.26562\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12646.689\tTol1: 0.00445\tTol2: 0.13547\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12556.879\tTol1: 0.00710\tTol2: 0.09505\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12214.330\tTol1: 0.02728\tTol2: 0.12186\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11875.930\tTol1: 0.02771\tTol2: 0.07138\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11668.882\tTol1: 0.01743\tTol2: 0.04175\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11466.957\tTol1: 0.01730\tTol2: 0.02955\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.35 sec)\tlogL: -14027.131\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12621.278\tTol1: 0.10022\tTol2: 0.26624\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12477.352\tTol1: 0.01140\tTol2: 0.14358\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12186.584\tTol1: 0.02330\tTol2: 0.11609\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12078.432\tTol1: 0.00887\tTol2: 0.06540\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11959.352\tTol1: 0.00986\tTol2: 0.04829\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11713.769\tTol1: 0.02053\tTol2: 0.04652\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11610.142\tTol1: 0.00885\tTol2: 0.04598\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11518.139\tTol1: 0.00792\tTol2: 0.04595\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11369.127\tTol1: 0.01294\tTol2: 0.02949\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -14125.691\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12969.260\tTol1: 0.08187\tTol2: 0.26287\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12542.572\tTol1: 0.03290\tTol2: 0.13515\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12233.311\tTol1: 0.02466\tTol2: 0.14749\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12071.045\tTol1: 0.01326\tTol2: 0.06780\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12050.284\tTol1: 0.00172\tTol2: 0.04496\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11841.688\tTol1: 0.01731\tTol2: 0.03910\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11514.968\tTol1: 0.02759\tTol2: 0.03253\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11259.082\tTol1: 0.02222\tTol2: 0.03054\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11163.841\tTol1: 0.00846\tTol2: 0.03587\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11882.378\tTol1: 0.06436\tTol2: 0.03647\t\n",
      "Iter 12/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11062.805\tTol1: 0.06897\tTol2: 0.06130\t\n",
      "Iter 13/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -10692.686\tTol1: 0.03346\tTol2: 0.02961\t\n",
      "Iter 14/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -10578.305\tTol1: 0.01070\tTol2: 0.02419\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.43 sec)\tlogL: -13957.662\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12569.775\tTol1: 0.09944\tTol2: 0.26649\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12342.212\tTol1: 0.01810\tTol2: 0.13439\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12282.590\tTol1: 0.00483\tTol2: 0.09463\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12036.199\tTol1: 0.02006\tTol2: 0.06840\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11894.108\tTol1: 0.01181\tTol2: 0.04873\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11725.053\tTol1: 0.01421\tTol2: 0.03218\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11540.064\tTol1: 0.01578\tTol2: 0.03427\t\n",
      "Iter 9/50 \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11414.887\tTol1: 0.01085\tTol2: 0.02436\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.48 sec)\tlogL: -13944.529\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12712.770\tTol1: 0.08833\tTol2: 0.26550\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12339.699\tTol1: 0.02935\tTol2: 0.13457\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12201.787\tTol1: 0.01118\tTol2: 0.10202\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12026.624\tTol1: 0.01436\tTol2: 0.06268\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11878.887\tTol1: 0.01228\tTol2: 0.04537\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11663.879\tTol1: 0.01810\tTol2: 0.03694\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11478.645\tTol1: 0.01588\tTol2: 0.03033\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11291.145\tTol1: 0.01633\tTol2: 0.02882\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.34 sec)\tlogL: -13854.299\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12585.760\tTol1: 0.09156\tTol2: 0.26643\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12352.056\tTol1: 0.01857\tTol2: 0.13944\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12207.962\tTol1: 0.01167\tTol2: 0.10173\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12147.316\tTol1: 0.00497\tTol2: 0.06874\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11901.396\tTol1: 0.02024\tTol2: 0.07131\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11776.210\tTol1: 0.01052\tTol2: 0.04627\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11493.474\tTol1: 0.02401\tTol2: 0.04008\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11369.876\tTol1: 0.01075\tTol2: 0.02938\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.43 sec)\tlogL: -14000.500\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12790.982\tTol1: 0.08639\tTol2: 0.26701\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12356.504\tTol1: 0.03397\tTol2: 0.13958\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12330.610\tTol1: 0.00210\tTol2: 0.09595\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12087.148\tTol1: 0.01974\tTol2: 0.07265\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11951.326\tTol1: 0.01124\tTol2: 0.06421\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11743.594\tTol1: 0.01738\tTol2: 0.04850\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11521.221\tTol1: 0.01894\tTol2: 0.03307\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11661.050\tTol1: 0.01214\tTol2: 0.02841\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.36 sec)\tlogL: -14058.616\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13401.913\tTol1: 0.04671\tTol2: 0.26573\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12393.304\tTol1: 0.07526\tTol2: 0.13843\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12336.979\tTol1: 0.00454\tTol2: 0.13122\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12074.123\tTol1: 0.02131\tTol2: 0.06484\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12001.345\tTol1: 0.00603\tTol2: 0.05814\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11753.787\tTol1: 0.02063\tTol2: 0.04157\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11562.498\tTol1: 0.01627\tTol2: 0.03358\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11544.598\tTol1: 0.00155\tTol2: 0.04805\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11066.629\tTol1: 0.04140\tTol2: 0.04211\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11542.529\tTol1: 0.04300\tTol2: 0.03314\t\n",
      "Iter 12/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -10805.885\tTol1: 0.06382\tTol2: 0.06869\t\n",
      "Iter 13/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -10641.676\tTol1: 0.01520\tTol2: 0.03695\t\n",
      "Iter 14/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -10617.692\tTol1: 0.00225\tTol2: 0.02113\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.36 sec)\tlogL: -13870.132\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12530.112\tTol1: 0.09661\tTol2: 0.26547\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12366.806\tTol1: 0.01303\tTol2: 0.13467\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12272.364\tTol1: 0.00764\tTol2: 0.10495\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12029.021\tTol1: 0.01983\tTol2: 0.06566\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11892.523\tTol1: 0.01135\tTol2: 0.05305\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11717.654\tTol1: 0.01470\tTol2: 0.04738\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -11700.412\tTol1: 0.00147\tTol2: 0.04206\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11296.711\tTol1: 0.03450\tTol2: 0.03749\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11182.775\tTol1: 0.01009\tTol2: 0.03398\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -10961.270\tTol1: 0.01981\tTol2: 0.02982\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -13938.742\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12601.179\tTol1: 0.09596\tTol2: 0.26546\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12321.030\tTol1: 0.02223\tTol2: 0.14037\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12348.360\tTol1: 0.00222\tTol2: 0.09023\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12086.818\tTol1: 0.02118\tTol2: 0.07707\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11946.284\tTol1: 0.01163\tTol2: 0.06917\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11596.975\tTol1: 0.02924\tTol2: 0.05330\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11663.447\tTol1: 0.00573\tTol2: 0.03541\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11158.418\tTol1: 0.04330\tTol2: 0.04381\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11042.935\tTol1: 0.01035\tTol2: 0.03755\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -10880.125\tTol1: 0.01474\tTol2: 0.02777\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.33 sec)\tlogL: -13924.914\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12573.835\tTol1: 0.09703\tTol2: 0.26502\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12532.320\tTol1: 0.00330\tTol2: 0.13285\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12340.283\tTol1: 0.01532\tTol2: 0.10076\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12035.366\tTol1: 0.02471\tTol2: 0.08431\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11906.582\tTol1: 0.01070\tTol2: 0.04719\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11822.256\tTol1: 0.00708\tTol2: 0.03761\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11492.748\tTol1: 0.02787\tTol2: 0.05272\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11351.280\tTol1: 0.01231\tTol2: 0.03128\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11116.881\tTol1: 0.02065\tTol2: 0.03988\t\n",
      "Iter 11/50 \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -10949.771\tTol1: 0.01503\tTol2: 0.03890\t\n",
      "Iter 12/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11272.643\tTol1: 0.02949\tTol2: 0.02659\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.33 sec)\tlogL: -13866.527\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12591.152\tTol1: 0.09198\tTol2: 0.26560\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12340.788\tTol1: 0.01988\tTol2: 0.14372\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12589.909\tTol1: 0.02019\tTol2: 0.10071\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12041.184\tTol1: 0.04358\tTol2: 0.09517\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11839.508\tTol1: 0.01675\tTol2: 0.04463\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11621.826\tTol1: 0.01839\tTol2: 0.03876\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11847.121\tTol1: 0.01939\tTol2: 0.03318\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11311.118\tTol1: 0.04524\tTol2: 0.04971\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11261.424\tTol1: 0.00439\tTol2: 0.02952\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -14224.232\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12663.908\tTol1: 0.10969\tTol2: 0.26355\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12427.135\tTol1: 0.01870\tTol2: 0.14228\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12281.618\tTol1: 0.01171\tTol2: 0.09631\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12081.744\tTol1: 0.01627\tTol2: 0.06312\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12164.770\tTol1: 0.00687\tTol2: 0.05051\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11756.605\tTol1: 0.03355\tTol2: 0.08788\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11561.012\tTol1: 0.01664\tTol2: 0.03732\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11376.314\tTol1: 0.01598\tTol2: 0.02593\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.34 sec)\tlogL: -13926.379\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12527.782\tTol1: 0.10043\tTol2: 0.26682\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12448.257\tTol1: 0.00635\tTol2: 0.13801\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12175.426\tTol1: 0.02192\tTol2: 0.09783\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12004.826\tTol1: 0.01401\tTol2: 0.06622\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11959.981\tTol1: 0.00374\tTol2: 0.05321\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11821.602\tTol1: 0.01157\tTol2: 0.04694\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11925.992\tTol1: 0.00883\tTol2: 0.03807\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11351.043\tTol1: 0.04821\tTol2: 0.04019\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11307.430\tTol1: 0.00384\tTol2: 0.04082\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11015.139\tTol1: 0.02585\tTol2: 0.02591\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.32 sec)\tlogL: -13970.890\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12850.611\tTol1: 0.08019\tTol2: 0.26571\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12349.773\tTol1: 0.03897\tTol2: 0.14330\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12167.621\tTol1: 0.01475\tTol2: 0.09984\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11985.701\tTol1: 0.01495\tTol2: 0.06378\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11836.595\tTol1: 0.01244\tTol2: 0.04343\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11921.209\tTol1: 0.00715\tTol2: 0.03367\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11470.960\tTol1: 0.03777\tTol2: 0.05841\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11290.148\tTol1: 0.01576\tTol2: 0.04064\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11162.891\tTol1: 0.01127\tTol2: 0.02796\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.34 sec)\tlogL: -14006.801\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12743.197\tTol1: 0.09021\tTol2: 0.26505\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12391.648\tTol1: 0.02759\tTol2: 0.13739\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12172.718\tTol1: 0.01767\tTol2: 0.09081\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12109.485\tTol1: 0.00519\tTol2: 0.06271\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11817.745\tTol1: 0.02409\tTol2: 0.04802\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11885.197\tTol1: 0.00571\tTol2: 0.04305\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11592.438\tTol1: 0.02463\tTol2: 0.06005\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11390.719\tTol1: 0.01740\tTol2: 0.03283\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11220.505\tTol1: 0.01494\tTol2: 0.02751\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "Betas shape:  (203, 2, 2)\n",
      "Q estimate shape:  (4999, 2)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.37 sec)\tlogL: -13948.646\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12623.918\tTol1: 0.09497\tTol2: 0.26588\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12474.752\tTol1: 0.01182\tTol2: 0.13775\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12211.127\tTol1: 0.02113\tTol2: 0.09614\t\n",
      "Iter 5/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12183.912\tTol1: 0.00223\tTol2: 0.05886\t\n",
      "Iter 6/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11848.329\tTol1: 0.02754\tTol2: 0.06621\t\n",
      "Iter 7/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11676.106\tTol1: 0.01454\tTol2: 0.03877\t\n",
      "Iter 8/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11425.599\tTol1: 0.02145\tTol2: 0.03087\t\n",
      "Iter 9/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11256.678\tTol1: 0.01478\tTol2: 0.03171\t\n",
      "Iter 10/50 \n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11098.922\tTol1: 0.01401\tTol2: 0.03047\t\n",
      "Iter 11/50 \n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11087.029\tTol1: 0.00107\tTol2: 0.02692\t\n",
      "Finished training\n",
      "20 runs over Raykar, Epochs to converge=  10.55\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "([[-13937.9931640625,\n",
       "   -12598.732421875,\n",
       "   -12296.548828125,\n",
       "   -12243.10546875,\n",
       "   -12006.0400390625,\n",
       "   -12080.296875,\n",
       "   -11911.111328125,\n",
       "   -11528.2177734375,\n",
       "   -11436.9423828125,\n",
       "   -11167.068359375,\n",
       "   -11049.994140625],\n",
       "  [-13833.984375,\n",
       "   -12585.892578125,\n",
       "   -12838.59765625,\n",
       "   -12223.2626953125,\n",
       "   -12118.958984375,\n",
       "   -11870.1328125,\n",
       "   -11719.892578125,\n",
       "   -11624.826171875,\n",
       "   -11721.67578125,\n",
       "   -11242.373046875,\n",
       "   -11047.0595703125],\n",
       "  [-13906.287109375,\n",
       "   -12563.078125,\n",
       "   -12333.8408203125,\n",
       "   -12222.1796875,\n",
       "   -11995.27734375,\n",
       "   -11886.43359375,\n",
       "   -11798.1318359375,\n",
       "   -11751.5673828125,\n",
       "   -11269.716796875,\n",
       "   -11023.748046875,\n",
       "   -10948.38671875,\n",
       "   -11060.787109375],\n",
       "  [-14029.521484375,\n",
       "   -12703.2451171875,\n",
       "   -12646.689453125,\n",
       "   -12556.87890625,\n",
       "   -12214.330078125,\n",
       "   -11875.9296875,\n",
       "   -11668.8818359375,\n",
       "   -11466.95703125],\n",
       "  [-14027.130859375,\n",
       "   -12621.2783203125,\n",
       "   -12477.3515625,\n",
       "   -12186.583984375,\n",
       "   -12078.431640625,\n",
       "   -11959.3515625,\n",
       "   -11713.7685546875,\n",
       "   -11610.1416015625,\n",
       "   -11518.138671875,\n",
       "   -11369.126953125],\n",
       "  [-14125.69140625,\n",
       "   -12969.259765625,\n",
       "   -12542.572265625,\n",
       "   -12233.310546875,\n",
       "   -12071.044921875,\n",
       "   -12050.2841796875,\n",
       "   -11841.6884765625,\n",
       "   -11514.9677734375,\n",
       "   -11259.08203125,\n",
       "   -11163.8408203125,\n",
       "   -11882.3779296875,\n",
       "   -11062.8046875,\n",
       "   -10692.685546875,\n",
       "   -10578.3046875],\n",
       "  [-13957.662109375,\n",
       "   -12569.775390625,\n",
       "   -12342.2119140625,\n",
       "   -12282.58984375,\n",
       "   -12036.19921875,\n",
       "   -11894.1083984375,\n",
       "   -11725.052734375,\n",
       "   -11540.064453125,\n",
       "   -11414.88671875],\n",
       "  [-13944.529296875,\n",
       "   -12712.76953125,\n",
       "   -12339.69921875,\n",
       "   -12201.787109375,\n",
       "   -12026.6240234375,\n",
       "   -11878.88671875,\n",
       "   -11663.87890625,\n",
       "   -11478.64453125,\n",
       "   -11291.14453125],\n",
       "  [-13854.298828125,\n",
       "   -12585.759765625,\n",
       "   -12352.0556640625,\n",
       "   -12207.9619140625,\n",
       "   -12147.31640625,\n",
       "   -11901.396484375,\n",
       "   -11776.2099609375,\n",
       "   -11493.4736328125,\n",
       "   -11369.8759765625],\n",
       "  [-14000.5,\n",
       "   -12790.982421875,\n",
       "   -12356.50390625,\n",
       "   -12330.6103515625,\n",
       "   -12087.1484375,\n",
       "   -11951.326171875,\n",
       "   -11743.59375,\n",
       "   -11521.220703125,\n",
       "   -11661.0498046875],\n",
       "  [-14058.6162109375,\n",
       "   -13401.9130859375,\n",
       "   -12393.3037109375,\n",
       "   -12336.978515625,\n",
       "   -12074.123046875,\n",
       "   -12001.3447265625,\n",
       "   -11753.787109375,\n",
       "   -11562.498046875,\n",
       "   -11544.59765625,\n",
       "   -11066.62890625,\n",
       "   -11542.529296875,\n",
       "   -10805.884765625,\n",
       "   -10641.67578125,\n",
       "   -10617.6923828125],\n",
       "  [-13870.1318359375,\n",
       "   -12530.1123046875,\n",
       "   -12366.8056640625,\n",
       "   -12272.3642578125,\n",
       "   -12029.021484375,\n",
       "   -11892.5234375,\n",
       "   -11717.654296875,\n",
       "   -11700.412109375,\n",
       "   -11296.7109375,\n",
       "   -11182.775390625,\n",
       "   -10961.26953125],\n",
       "  [-13938.7421875,\n",
       "   -12601.1787109375,\n",
       "   -12321.0302734375,\n",
       "   -12348.3603515625,\n",
       "   -12086.818359375,\n",
       "   -11946.2841796875,\n",
       "   -11596.974609375,\n",
       "   -11663.447265625,\n",
       "   -11158.41796875,\n",
       "   -11042.9345703125,\n",
       "   -10880.125],\n",
       "  [-13924.9140625,\n",
       "   -12573.8349609375,\n",
       "   -12532.3203125,\n",
       "   -12340.283203125,\n",
       "   -12035.3662109375,\n",
       "   -11906.58203125,\n",
       "   -11822.255859375,\n",
       "   -11492.748046875,\n",
       "   -11351.2802734375,\n",
       "   -11116.880859375,\n",
       "   -10949.771484375,\n",
       "   -11272.642578125],\n",
       "  [-13866.52734375,\n",
       "   -12591.15234375,\n",
       "   -12340.7880859375,\n",
       "   -12589.9091796875,\n",
       "   -12041.18359375,\n",
       "   -11839.5078125,\n",
       "   -11621.826171875,\n",
       "   -11847.12109375,\n",
       "   -11311.1181640625,\n",
       "   -11261.423828125],\n",
       "  [-14224.232421875,\n",
       "   -12663.908203125,\n",
       "   -12427.134765625,\n",
       "   -12281.6181640625,\n",
       "   -12081.744140625,\n",
       "   -12164.76953125,\n",
       "   -11756.60546875,\n",
       "   -11561.01171875,\n",
       "   -11376.314453125],\n",
       "  [-13926.37890625,\n",
       "   -12527.7822265625,\n",
       "   -12448.2568359375,\n",
       "   -12175.42578125,\n",
       "   -12004.826171875,\n",
       "   -11959.9814453125,\n",
       "   -11821.6015625,\n",
       "   -11925.9921875,\n",
       "   -11351.04296875,\n",
       "   -11307.4296875,\n",
       "   -11015.138671875],\n",
       "  [-13970.8896484375,\n",
       "   -12850.611328125,\n",
       "   -12349.7734375,\n",
       "   -12167.62109375,\n",
       "   -11985.701171875,\n",
       "   -11836.5947265625,\n",
       "   -11921.208984375,\n",
       "   -11470.9599609375,\n",
       "   -11290.1484375,\n",
       "   -11162.890625],\n",
       "  [-14006.80078125,\n",
       "   -12743.197265625,\n",
       "   -12391.6484375,\n",
       "   -12172.7177734375,\n",
       "   -12109.4853515625,\n",
       "   -11817.7451171875,\n",
       "   -11885.197265625,\n",
       "   -11592.4384765625,\n",
       "   -11390.71875,\n",
       "   -11220.5048828125],\n",
       "  [-13948.646484375,\n",
       "   -12623.91796875,\n",
       "   -12474.751953125,\n",
       "   -12211.126953125,\n",
       "   -12183.912109375,\n",
       "   -11848.3291015625,\n",
       "   -11676.1064453125,\n",
       "   -11425.5986328125,\n",
       "   -11256.677734375,\n",
       "   -11098.921875,\n",
       "   -11087.029296875]],\n",
       " 5)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import ModelInf_EM as Raykar\n",
    "R_model = Raykar(init_Z=\"softmv\")\n",
    "#R_model = Raykar(init_Z=\"model\", priors='laplace', n_init_Z=5)\n",
    "\n",
    "args = {'epochs':1, 'batch_size':BATCH_SIZE, 'optimizer':OPT}\n",
    "R_model.set_model(model_R, **args)\n",
    "\n",
    "R_model.fit(X_train, y_obs_categorical, runs=20, max_iter=EPOCHS_BASE, tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(15,7))\n",
    "for a, value in enumerate(logL_hists): \n",
    "    if a != i_r:\n",
    "        plt.plot(range(len(value)),value,'.-')\n",
    "plt.plot(range(len(logL_hists[i_r])),logL_hists[i_r],'o-',markersize=10,label=\"Selected run\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.9117823564712942\n",
      "Test accuracy of model = 0.7334333583395849\n"
     ]
    }
   ],
   "source": [
    "raykar_fx = R_model.get_basemodel()\n",
    "\n",
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(raykar_fx, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(raykar_fx, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reliability Annotator Model "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 300)           4200300   \n",
      "_________________________________________________________________\n",
      "cu_dnngru_1 (CuDNNGRU)       (None, 50, 128)           165120    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 50, 128)           0         \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (None, 64)                37248     \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 4,402,798\n",
      "Trainable params: 202,498\n",
      "Non-trainable params: 4,200,300\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "aux_model = clone_UB.get_model()\n",
    "aux_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.72 sec)\tlogL: -17058.905\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15980.445\tTol1: 0.06322\tTol2: 0.00109\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -15451.859\tTol1: 0.03308\tTol2: 0.00165\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15051.144\tTol1: 0.02593\tTol2: 0.00190\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.53 sec)\tlogL: -17034.645\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -15982.595\tTol1: 0.06176\tTol2: 0.00116\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -15564.066\tTol1: 0.02619\tTol2: 0.00168\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.55 sec)\tlogL: -18515.214\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15976.216\tTol1: 0.13713\tTol2: 0.00153\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15890.855\tTol1: 0.00534\tTol2: 0.00192\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.50 sec)\tlogL: -16708.305\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -16046.473\tTol1: 0.03961\tTol2: 0.00130\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15442.901\tTol1: 0.03761\tTol2: 0.00192\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -15008.666\tTol1: 0.02812\tTol2: 0.00302\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.41 sec)\tlogL: -16749.402\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -16378.183\tTol1: 0.02216\tTol2: 0.00117\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.42 sec)\tlogL: -17050.589\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15870.575\tTol1: 0.06921\tTol2: 0.00090\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15417.755\tTol1: 0.02853\tTol2: 0.00198\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.42 sec)\tlogL: -16854.894\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -16286.595\tTol1: 0.03372\tTol2: 0.00112\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15443.355\tTol1: 0.05178\tTol2: 0.00195\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -15392.254\tTol1: 0.00331\tTol2: 0.00216\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.51 sec)\tlogL: -16712.316\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -16632.088\tTol1: 0.00480\tTol2: 0.00115\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.46 sec)\tlogL: -16740.527\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -16041.754\tTol1: 0.04174\tTol2: 0.00105\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15842.383\tTol1: 0.01243\tTol2: 0.00151\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.45 sec)\tlogL: -16821.772\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -15910.392\tTol1: 0.05418\tTol2: 0.00110\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15407.943\tTol1: 0.03158\tTol2: 0.00140\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15125.800\tTol1: 0.01831\tTol2: 0.00184\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.41 sec)\tlogL: -16911.175\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15959.719\tTol1: 0.05626\tTol2: 0.00118\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15388.847\tTol1: 0.03577\tTol2: 0.00154\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15233.399\tTol1: 0.01010\tTol2: 0.00205\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.43 sec)\tlogL: -16834.195\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -16071.779\tTol1: 0.04529\tTol2: 0.00106\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15418.989\tTol1: 0.04062\tTol2: 0.00125\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15081.622\tTol1: 0.02188\tTol2: 0.00193\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.47 sec)\tlogL: -16971.555\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15796.322\tTol1: 0.06925\tTol2: 0.00107\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15378.347\tTol1: 0.02646\tTol2: 0.00179\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.43 sec)\tlogL: -16920.836\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -16109.020\tTol1: 0.04798\tTol2: 0.00103\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15535.295\tTol1: 0.03562\tTol2: 0.00148\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15151.570\tTol1: 0.02470\tTol2: 0.00207\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.49 sec)\tlogL: -16649.014\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -16419.855\tTol1: 0.01376\tTol2: 0.00128\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.48 sec)\tlogL: -17319.555\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15819.293\tTol1: 0.08662\tTol2: 0.00154\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -15422.052\tTol1: 0.02511\tTol2: 0.00181\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.42 sec)\tlogL: -16950.321\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15882.211\tTol1: 0.06301\tTol2: 0.00131\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -15436.570\tTol1: 0.02806\tTol2: 0.00145\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.49 sec)\tlogL: -16687.359\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15716.687\tTol1: 0.05817\tTol2: 0.00179\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15376.194\tTol1: 0.02166\tTol2: 0.00199\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.47 sec)\tlogL: -17837.651\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -16197.262\tTol1: 0.09196\tTol2: 0.00158\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -15483.671\tTol1: 0.04406\tTol2: 0.00212\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -15153.012\tTol1: 0.02136\tTol2: 0.00251\t\n",
      "Finished training\n",
      "Initializing new EM...\n",
      "b shape:  (203, 1)\n",
      "R estimate shape:  (4999, 203, 1)\n",
      "Iter 1/50 \n",
      "M step: done,  E step: done //  (in 1.54 sec)\tlogL: -16955.365\t\n",
      "Iter 2/50 \n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -15880.681\tTol1: 0.06338\tTol2: 0.00116\t\n",
      "Iter 3/50 \n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15326.555\tTol1: 0.03489\tTol2: 0.00165\t\n",
      "Iter 4/50 \n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -15221.559\tTol1: 0.00685\tTol2: 0.00223\t\n",
      "Finished training\n",
      "20 runs, Epochs to converge=  3.3\n"
     ]
    }
   ],
   "source": [
    "from codeE.methods import ModelInf_EM_R\n",
    "LogReg_crowds = ModelInf_EM_R()\n",
    "\n",
    "args = {'epochs':1, 'batch_size':BATCH_SIZE, 'optimizer':OPT}\n",
    "LogReg_crowds.set_model(aux_model, **args)\n",
    "\n",
    "#LogReg_crowds.init_E(X_train, y_obs_categorical, method=\"softmv\")\n",
    "#LogReg_crowds.get_restimation()\n",
    "\n",
    "logL_hists,i_r = LogReg_crowds.fit(X_train, y_obs_categorical, runs = 20, max_iter=EPOCHS_BASE, tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(15,7))\n",
    "for a, value in enumerate(logL_hists): \n",
    "    if a != i_r:\n",
    "        plt.plot(range(len(value)),value,'.-')\n",
    "plt.plot(range(len(logL_hists[i_r])),logL_hists[i_r],'o-',markersize=10,label=\"Selected run\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.795759151830366\n",
      "Test accuracy of model = 0.7434358589647412\n"
     ]
    }
   ],
   "source": [
    "LogReg_fx = LogReg_crowds.get_basemodel()\n",
    "\n",
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(LogReg_fx, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(LogReg_fx, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## C-MoA\n",
    "Groups over individual scenario"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Auxiliar representation with variable number of annotations per annotator and identity"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual sparse representation, variable shape (N,)=  (4999,)\n",
      "one-hot vectors of K-dimensions, K= 2\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([array([[0, 1],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [0, 1]], dtype=int8),\n",
       "       array([[1, 0],\n",
       "       [1, 0],\n",
       "       [1, 0],\n",
       "       [1, 0],\n",
       "       [1, 0]], dtype=int8),\n",
       "       array([[0, 1],\n",
       "       [1, 0],\n",
       "       [0, 1],\n",
       "       [1, 0],\n",
       "       [1, 0],\n",
       "       [1, 0]], dtype=int8),\n",
       "       ...,\n",
       "       array([[0, 1],\n",
       "       [1, 0],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [1, 0]], dtype=int8),\n",
       "       array([[0, 1],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [1, 0],\n",
       "       [0, 1]], dtype=int8),\n",
       "       array([[1, 0],\n",
       "       [0, 1],\n",
       "       [0, 1],\n",
       "       [1, 0],\n",
       "       [1, 0]], dtype=int8)], dtype=object)"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_cat_var, A_idx_var = set_representation(y_obs,\"onehotvar\")\n",
    "print(\"Individual sparse representation, variable shape (N,)= \",y_cat_var.shape)\n",
    "K = y_cat_var[0].shape[1]\n",
    "print(\"one-hot vectors of K-dimensions, K=\",K)\n",
    "y_cat_var"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 300)           4200300   \n",
      "_________________________________________________________________\n",
      "cu_dnngru_1 (CuDNNGRU)       (None, 50, 128)           165120    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 50, 128)           0         \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (None, 64)                37248     \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 4,402,798\n",
      "Trainable params: 202,498\n",
      "Non-trainable params: 4,200,300\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "M_seted = 3\n",
    "\n",
    "aux_model = clone_UB.get_model()\n",
    "aux_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Fixed annotators representation (non trainable embedding)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shape of annotator representation (T, R_t)= (203, 2)\n"
     ]
    }
   ],
   "source": [
    "T= np.concatenate(A_idx_var).max() +1\n",
    "A_rep = np.zeros((T, K))\n",
    "for i in range(N):\n",
    "    for l, t_idx in enumerate(A_idx_var[i]):\n",
    "        obs_t = y_cat_var[i][l].argmax(axis=-1)\n",
    "        A_rep[t_idx, obs_t] += 1\n",
    "print(\"shape of annotator representation (T, R_t)=\", A_rep.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 1, 2)              406       \n",
      "_________________________________________________________________\n",
      "reshape_1 (Reshape)          (None, 2)                 0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 6)                 18        \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 3)                 21        \n",
      "=================================================================\n",
      "Total params: 445\n",
      "Trainable params: 39\n",
      "Non-trainable params: 406\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import *\n",
    "group_model = Sequential()\n",
    "group_model.add(Embedding(T, K,input_length=1, \n",
    "                         trainable=False,weights=[A_rep]))\n",
    "#                         trainable=True))\n",
    "group_model.add(Reshape([K]))\n",
    "group_model.add(Dense(K*M_seted, activation='relu'))\n",
    "group_model.add(Dense(M_seted, activation='softmax'))\n",
    "group_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.24 sec)\tlogL: -17625.454\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16747.986\tTol1: 0.04978\tTol2: 0.09618\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -15980.623\tTol1: 0.04582\tTol2: 0.08676\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -15282.435\tTol1: 0.04369\tTol2: 0.09088\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14951.506\tTol1: 0.02165\tTol2: 0.07884\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14578.729\tTol1: 0.02493\tTol2: 0.10473\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14734.526\tTol1: 0.01069\tTol2: 0.08248\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14090.047\tTol1: 0.04374\tTol2: 0.19222\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13746.264\tTol1: 0.02440\tTol2: 0.10038\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -13756.858\tTol1: 0.00077\tTol2: 0.06114\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13291.597\tTol1: 0.03382\tTol2: 0.16161\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12863.333\tTol1: 0.03222\tTol2: 0.11923\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12594.473\tTol1: 0.02090\tTol2: 0.07468\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12305.974\tTol1: 0.02291\tTol2: 0.07677\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12189.431\tTol1: 0.00947\tTol2: 0.08682\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11902.671\tTol1: 0.02353\tTol2: 0.12150\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11762.321\tTol1: 0.01179\tTol2: 0.06871\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11613.758\tTol1: 0.01263\tTol2: 0.06957\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11469.350\tTol1: 0.01243\tTol2: 0.08620\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11359.851\tTol1: 0.00955\tTol2: 0.09371\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11422.550\tTol1: 0.00552\tTol2: 0.08455\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11418.367\tTol1: 0.00037\tTol2: 0.16976\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11262.334\tTol1: 0.01367\tTol2: 0.14958\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11080.099\tTol1: 0.01618\tTol2: 0.12125\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10988.245\tTol1: 0.00829\tTol2: 0.08326\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10938.625\tTol1: 0.00452\tTol2: 0.07016\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10888.891\tTol1: 0.00455\tTol2: 0.05274\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10872.995\tTol1: 0.00146\tTol2: 0.02697\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.93 sec)\tlogL: -17661.371\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -16756.018\tTol1: 0.05126\tTol2: 0.21426\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -16253.682\tTol1: 0.02998\tTol2: 0.15625\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15578.627\tTol1: 0.04153\tTol2: 0.19081\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15100.758\tTol1: 0.03067\tTol2: 0.04392\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14729.052\tTol1: 0.02462\tTol2: 0.05980\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14373.769\tTol1: 0.02412\tTol2: 0.05177\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14061.627\tTol1: 0.02172\tTol2: 0.06069\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13983.763\tTol1: 0.00554\tTol2: 0.06456\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13521.813\tTol1: 0.03303\tTol2: 0.12145\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13481.413\tTol1: 0.00299\tTol2: 0.02552\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.00 sec)\tlogL: -17840.810\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -16855.941\tTol1: 0.05520\tTol2: 0.19638\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16297.658\tTol1: 0.03312\tTol2: 0.18914\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15764.671\tTol1: 0.03270\tTol2: 0.16386\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15267.771\tTol1: 0.03152\tTol2: 0.15101\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14862.488\tTol1: 0.02655\tTol2: 0.15155\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14571.509\tTol1: 0.01958\tTol2: 0.13371\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14321.356\tTol1: 0.01717\tTol2: 0.12388\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14120.821\tTol1: 0.01400\tTol2: 0.11458\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13831.936\tTol1: 0.02046\tTol2: 0.12970\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13397.645\tTol1: 0.03140\tTol2: 0.12486\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13082.690\tTol1: 0.02351\tTol2: 0.12816\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12767.237\tTol1: 0.02411\tTol2: 0.13256\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13207.255\tTol1: 0.03446\tTol2: 0.11157\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12405.870\tTol1: 0.06068\tTol2: 0.17359\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -12172.477\tTol1: 0.01881\tTol2: 0.17153\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11932.818\tTol1: 0.01969\tTol2: 0.12738\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11986.184\tTol1: 0.00447\tTol2: 0.10944\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11697.050\tTol1: 0.02412\tTol2: 0.13829\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11562.190\tTol1: 0.01153\tTol2: 0.12352\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11854.139\tTol1: 0.02525\tTol2: 0.09814\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11542.022\tTol1: 0.02633\tTol2: 0.14933\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11628.736\tTol1: 0.00751\tTol2: 0.14123\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11298.107\tTol1: 0.02843\tTol2: 0.15888\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11231.334\tTol1: 0.00591\tTol2: 0.08089\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11111.634\tTol1: 0.01066\tTol2: 0.11499\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11036.880\tTol1: 0.00673\tTol2: 0.09496\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11379.808\tTol1: 0.03107\tTol2: 0.08924\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11039.235\tTol1: 0.02993\tTol2: 0.13718\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11104.559\tTol1: 0.00592\tTol2: 0.05723\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10946.041\tTol1: 0.01428\tTol2: 0.11607\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10826.122\tTol1: 0.01096\tTol2: 0.04847\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10852.811\tTol1: 0.00247\tTol2: 0.04669\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10745.433\tTol1: 0.00989\tTol2: 0.07065\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10710.884\tTol1: 0.00322\tTol2: 0.05322\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11064.366\tTol1: 0.03300\tTol2: 0.03214\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10952.254\tTol1: 0.01013\tTol2: 0.12931\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10633.716\tTol1: 0.02908\tTol2: 0.07073\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10592.028\tTol1: 0.00392\tTol2: 0.03791\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10540.154\tTol1: 0.00490\tTol2: 0.03164\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10638.900\tTol1: 0.00937\tTol2: 0.05574\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10502.326\tTol1: 0.01284\tTol2: 0.09158\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10475.729\tTol1: 0.00253\tTol2: 0.04196\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10405.798\tTol1: 0.00668\tTol2: 0.04693\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10485.184\tTol1: 0.00763\tTol2: 0.03635\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10368.355\tTol1: 0.01114\tTol2: 0.05184\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10358.737\tTol1: 0.00093\tTol2: 0.02876\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.91 sec)\tlogL: -17250.543\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16446.129\tTol1: 0.04663\tTol2: 0.10115\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15863.884\tTol1: 0.03540\tTol2: 0.09152\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15510.416\tTol1: 0.02228\tTol2: 0.09001\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15275.244\tTol1: 0.01516\tTol2: 0.09474\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14791.835\tTol1: 0.03165\tTol2: 0.12042\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14461.270\tTol1: 0.02235\tTol2: 0.09325\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14173.893\tTol1: 0.01987\tTol2: 0.08394\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13942.248\tTol1: 0.01634\tTol2: 0.09041\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13852.650\tTol1: 0.00643\tTol2: 0.10147\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13312.836\tTol1: 0.03897\tTol2: 0.13716\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13019.522\tTol1: 0.02203\tTol2: 0.11251\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13286.556\tTol1: 0.02051\tTol2: 0.11899\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12468.018\tTol1: 0.06161\tTol2: 0.19587\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12627.557\tTol1: 0.01280\tTol2: 0.10743\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12248.613\tTol1: 0.03001\tTol2: 0.18450\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11970.068\tTol1: 0.02274\tTol2: 0.14773\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11726.681\tTol1: 0.02033\tTol2: 0.16435\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11621.043\tTol1: 0.00901\tTol2: 0.14914\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11787.482\tTol1: 0.01432\tTol2: 0.15342\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11368.521\tTol1: 0.03554\tTol2: 0.18078\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11270.547\tTol1: 0.00862\tTol2: 0.11875\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11258.400\tTol1: 0.00108\tTol2: 0.09712\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11245.286\tTol1: 0.00116\tTol2: 0.11727\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11330.756\tTol1: 0.00760\tTol2: 0.11751\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11205.917\tTol1: 0.01102\tTol2: 0.11554\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11062.738\tTol1: 0.01278\tTol2: 0.09839\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11142.339\tTol1: 0.00720\tTol2: 0.07365\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10952.501\tTol1: 0.01704\tTol2: 0.14486\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10954.107\tTol1: 0.00015\tTol2: 0.03966\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10823.962\tTol1: 0.01188\tTol2: 0.05684\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10807.506\tTol1: 0.00152\tTol2: 0.06734\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10755.807\tTol1: 0.00478\tTol2: 0.06149\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10700.946\tTol1: 0.00510\tTol2: 0.05447\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10660.856\tTol1: 0.00375\tTol2: 0.05274\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10607.409\tTol1: 0.00501\tTol2: 0.06505\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10586.243\tTol1: 0.00200\tTol2: 0.07812\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10644.637\tTol1: 0.00552\tTol2: 0.07286\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10488.934\tTol1: 0.01463\tTol2: 0.08630\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10480.850\tTol1: 0.00077\tTol2: 0.07762\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10447.721\tTol1: 0.00316\tTol2: 0.09564\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10382.672\tTol1: 0.00623\tTol2: 0.05493\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10343.955\tTol1: 0.00373\tTol2: 0.05849\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10401.846\tTol1: 0.00560\tTol2: 0.04496\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10363.007\tTol1: 0.00373\tTol2: 0.03867\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10296.362\tTol1: 0.00643\tTol2: 0.03916\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10332.569\tTol1: 0.00352\tTol2: 0.05008\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10268.739\tTol1: 0.00618\tTol2: 0.07429\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10239.145\tTol1: 0.00288\tTol2: 0.05829\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10244.688\tTol1: 0.00054\tTol2: 0.06663\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.98 sec)\tlogL: -17310.950\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -16534.274\tTol1: 0.04487\tTol2: 0.09520\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15895.848\tTol1: 0.03861\tTol2: 0.09399\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15453.169\tTol1: 0.02785\tTol2: 0.12350\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15020.908\tTol1: 0.02797\tTol2: 0.09194\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15057.090\tTol1: 0.00241\tTol2: 0.08256\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -14485.197\tTol1: 0.03798\tTol2: 0.17103\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14124.909\tTol1: 0.02487\tTol2: 0.07999\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13840.936\tTol1: 0.02010\tTol2: 0.06097\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13524.049\tTol1: 0.02289\tTol2: 0.04848\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14501.944\tTol1: 0.07231\tTol2: 0.08927\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12999.237\tTol1: 0.10362\tTol2: 0.29597\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12774.038\tTol1: 0.01732\tTol2: 0.06811\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12475.951\tTol1: 0.02334\tTol2: 0.09296\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12472.298\tTol1: 0.00029\tTol2: 0.08012\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12082.730\tTol1: 0.03123\tTol2: 0.11597\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11977.242\tTol1: 0.00873\tTol2: 0.06728\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11861.593\tTol1: 0.00966\tTol2: 0.07654\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11782.266\tTol1: 0.00669\tTol2: 0.07494\t\n",
      "Iter 20/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11654.436\tTol1: 0.01085\tTol2: 0.13350\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11553.435\tTol1: 0.00867\tTol2: 0.08282\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11912.737\tTol1: 0.03110\tTol2: 0.09541\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11405.271\tTol1: 0.04260\tTol2: 0.18199\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11405.509\tTol1: 0.00002\tTol2: 0.07088\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11256.247\tTol1: 0.01309\tTol2: 0.10086\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11180.766\tTol1: 0.00671\tTol2: 0.10009\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11061.641\tTol1: 0.01065\tTol2: 0.07095\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11105.093\tTol1: 0.00393\tTol2: 0.08614\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11166.838\tTol1: 0.00556\tTol2: 0.09934\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10923.064\tTol1: 0.02183\tTol2: 0.11811\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10845.998\tTol1: 0.00706\tTol2: 0.06899\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10796.240\tTol1: 0.00459\tTol2: 0.06696\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10764.989\tTol1: 0.00289\tTol2: 0.06506\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10706.160\tTol1: 0.00546\tTol2: 0.04531\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10815.500\tTol1: 0.01021\tTol2: 0.05412\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10676.616\tTol1: 0.01284\tTol2: 0.09395\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10599.053\tTol1: 0.00726\tTol2: 0.06833\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10555.205\tTol1: 0.00414\tTol2: 0.04119\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10513.806\tTol1: 0.00392\tTol2: 0.05611\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10855.690\tTol1: 0.03252\tTol2: 0.03585\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10462.623\tTol1: 0.03621\tTol2: 0.14834\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10408.588\tTol1: 0.00516\tTol2: 0.05744\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10372.088\tTol1: 0.00351\tTol2: 0.05401\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10391.706\tTol1: 0.00189\tTol2: 0.04403\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10623.439\tTol1: 0.02230\tTol2: 0.05467\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10407.258\tTol1: 0.02035\tTol2: 0.09478\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10289.728\tTol1: 0.01129\tTol2: 0.05854\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10222.913\tTol1: 0.00649\tTol2: 0.06304\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10316.506\tTol1: 0.00916\tTol2: 0.05663\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10188.440\tTol1: 0.01241\tTol2: 0.09229\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.06 sec)\tlogL: -17665.313\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16808.942\tTol1: 0.04848\tTol2: 0.10687\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16085.083\tTol1: 0.04306\tTol2: 0.08193\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -15488.490\tTol1: 0.03709\tTol2: 0.07115\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15112.455\tTol1: 0.02428\tTol2: 0.06054\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14826.189\tTol1: 0.01894\tTol2: 0.06705\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14621.583\tTol1: 0.01380\tTol2: 0.06318\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14211.766\tTol1: 0.02803\tTol2: 0.05924\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13938.436\tTol1: 0.01923\tTol2: 0.06295\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13621.925\tTol1: 0.02271\tTol2: 0.07728\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13682.027\tTol1: 0.00441\tTol2: 0.07881\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13028.174\tTol1: 0.04779\tTol2: 0.15154\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13448.422\tTol1: 0.03226\tTol2: 0.10791\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -12503.666\tTol1: 0.07025\tTol2: 0.16836\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12271.854\tTol1: 0.01854\tTol2: 0.09104\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12024.638\tTol1: 0.02014\tTol2: 0.09016\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12270.192\tTol1: 0.02042\tTol2: 0.08672\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11907.666\tTol1: 0.02955\tTol2: 0.11449\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11805.026\tTol1: 0.00862\tTol2: 0.16723\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11632.036\tTol1: 0.01465\tTol2: 0.10341\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11444.400\tTol1: 0.01613\tTol2: 0.09251\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11422.158\tTol1: 0.00194\tTol2: 0.10338\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11268.607\tTol1: 0.01344\tTol2: 0.08686\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11193.507\tTol1: 0.00666\tTol2: 0.06696\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11071.070\tTol1: 0.01094\tTol2: 0.08439\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11171.692\tTol1: 0.00909\tTol2: 0.05996\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11055.856\tTol1: 0.01037\tTol2: 0.09462\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10947.065\tTol1: 0.00984\tTol2: 0.10327\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10927.192\tTol1: 0.00182\tTol2: 0.06924\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10911.110\tTol1: 0.00147\tTol2: 0.08054\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11404.974\tTol1: 0.04526\tTol2: 0.09500\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10918.759\tTol1: 0.04263\tTol2: 0.17702\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10824.561\tTol1: 0.00863\tTol2: 0.09115\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10861.884\tTol1: 0.00345\tTol2: 0.08347\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10957.597\tTol1: 0.00881\tTol2: 0.11724\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10758.993\tTol1: 0.01812\tTol2: 0.14000\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10689.270\tTol1: 0.00648\tTol2: 0.09578\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10655.164\tTol1: 0.00319\tTol2: 0.12683\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10574.966\tTol1: 0.00753\tTol2: 0.12293\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10505.006\tTol1: 0.00662\tTol2: 0.11652\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10467.207\tTol1: 0.00360\tTol2: 0.09542\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10450.121\tTol1: 0.00163\tTol2: 0.11252\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10559.801\tTol1: 0.01050\tTol2: 0.09857\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10482.288\tTol1: 0.00734\tTol2: 0.14044\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10406.979\tTol1: 0.00718\tTol2: 0.09354\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10378.133\tTol1: 0.00277\tTol2: 0.08297\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10458.849\tTol1: 0.00778\tTol2: 0.08500\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10328.254\tTol1: 0.01249\tTol2: 0.08835\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10348.941\tTol1: 0.00200\tTol2: 0.07659\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10349.377\tTol1: 0.00004\tTol2: 0.07119\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.95 sec)\tlogL: -17455.398\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -16532.025\tTol1: 0.05290\tTol2: 0.11696\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -15994.671\tTol1: 0.03250\tTol2: 0.10554\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15364.176\tTol1: 0.03942\tTol2: 0.13267\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14939.654\tTol1: 0.02763\tTol2: 0.09852\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15167.799\tTol1: 0.01527\tTol2: 0.08821\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14391.598\tTol1: 0.05117\tTol2: 0.15481\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14147.806\tTol1: 0.01694\tTol2: 0.06319\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13765.426\tTol1: 0.02703\tTol2: 0.13508\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13413.638\tTol1: 0.02556\tTol2: 0.06633\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13179.664\tTol1: 0.01744\tTol2: 0.09886\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12730.093\tTol1: 0.03411\tTol2: 0.07558\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12654.729\tTol1: 0.00592\tTol2: 0.03817\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12281.291\tTol1: 0.02951\tTol2: 0.09773\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12061.195\tTol1: 0.01792\tTol2: 0.05936\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11809.670\tTol1: 0.02085\tTol2: 0.04093\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11682.562\tTol1: 0.01076\tTol2: 0.03988\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11607.582\tTol1: 0.00642\tTol2: 0.05315\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -11468.062\tTol1: 0.01202\tTol2: 0.07925\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11535.726\tTol1: 0.00590\tTol2: 0.10732\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11277.359\tTol1: 0.02240\tTol2: 0.13741\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11046.307\tTol1: 0.02049\tTol2: 0.04922\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10992.678\tTol1: 0.00485\tTol2: 0.03177\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11062.563\tTol1: 0.00636\tTol2: 0.03807\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11002.841\tTol1: 0.00540\tTol2: 0.08798\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10798.481\tTol1: 0.01857\tTol2: 0.13497\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10788.128\tTol1: 0.00096\tTol2: 0.06093\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10758.184\tTol1: 0.00278\tTol2: 0.06424\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10745.323\tTol1: 0.00120\tTol2: 0.07082\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10653.822\tTol1: 0.00852\tTol2: 0.08743\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10743.638\tTol1: 0.00843\tTol2: 0.07044\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10606.689\tTol1: 0.01275\tTol2: 0.06905\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10675.814\tTol1: 0.00652\tTol2: 0.09130\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10578.936\tTol1: 0.00907\tTol2: 0.07652\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10647.855\tTol1: 0.00651\tTol2: 0.05793\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10546.555\tTol1: 0.00951\tTol2: 0.08952\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10472.041\tTol1: 0.00707\tTol2: 0.06066\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10422.054\tTol1: 0.00477\tTol2: 0.07392\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10448.347\tTol1: 0.00252\tTol2: 0.05853\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10360.986\tTol1: 0.00836\tTol2: 0.06499\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -10435.682\tTol1: 0.00721\tTol2: 0.07049\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -10454.026\tTol1: 0.00176\tTol2: 0.12367\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10524.356\tTol1: 0.00673\tTol2: 0.09570\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10297.450\tTol1: 0.02156\tTol2: 0.15294\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10265.375\tTol1: 0.00311\tTol2: 0.07143\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10210.447\tTol1: 0.00535\tTol2: 0.05872\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10182.728\tTol1: 0.00271\tTol2: 0.07295\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10194.687\tTol1: 0.00117\tTol2: 0.09479\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10146.748\tTol1: 0.00470\tTol2: 0.10528\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10257.564\tTol1: 0.01092\tTol2: 0.10609\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.96 sec)\tlogL: -17880.732\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -16961.159\tTol1: 0.05143\tTol2: 0.14204\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16225.899\tTol1: 0.04335\tTol2: 0.10120\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -15579.874\tTol1: 0.03981\tTol2: 0.11264\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15137.139\tTol1: 0.02842\tTol2: 0.10806\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14816.908\tTol1: 0.02116\tTol2: 0.10192\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14286.187\tTol1: 0.03582\tTol2: 0.09835\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14295.785\tTol1: 0.00067\tTol2: 0.07349\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13577.230\tTol1: 0.05026\tTol2: 0.11915\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13233.496\tTol1: 0.02532\tTol2: 0.05701\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -12897.025\tTol1: 0.02543\tTol2: 0.07922\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12837.312\tTol1: 0.00463\tTol2: 0.09595\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12327.830\tTol1: 0.03969\tTol2: 0.14172\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12294.743\tTol1: 0.00268\tTol2: 0.07661\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12036.715\tTol1: 0.02099\tTol2: 0.11851\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11879.672\tTol1: 0.01305\tTol2: 0.10973\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11564.019\tTol1: 0.02657\tTol2: 0.16779\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11338.882\tTol1: 0.01947\tTol2: 0.13189\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11142.252\tTol1: 0.01734\tTol2: 0.08996\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11095.298\tTol1: 0.00421\tTol2: 0.08733\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10971.478\tTol1: 0.01116\tTol2: 0.08369\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10854.211\tTol1: 0.01069\tTol2: 0.08611\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10840.714\tTol1: 0.00124\tTol2: 0.11302\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10837.780\tTol1: 0.00027\tTol2: 0.13536\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -11324.801\tTol1: 0.04494\tTol2: 0.12593\t\n",
      "Iter 26/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10921.773\tTol1: 0.03559\tTol2: 0.20722\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10748.733\tTol1: 0.01584\tTol2: 0.09608\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10644.052\tTol1: 0.00974\tTol2: 0.07983\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10675.602\tTol1: 0.00296\tTol2: 0.06825\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10555.720\tTol1: 0.01123\tTol2: 0.07263\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10472.851\tTol1: 0.00785\tTol2: 0.09419\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10488.570\tTol1: 0.00150\tTol2: 0.09551\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10541.656\tTol1: 0.00506\tTol2: 0.06545\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10465.070\tTol1: 0.00727\tTol2: 0.07510\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10410.690\tTol1: 0.00520\tTol2: 0.11000\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10376.810\tTol1: 0.00325\tTol2: 0.09258\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10376.667\tTol1: 0.00001\tTol2: 0.08058\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10364.787\tTol1: 0.00114\tTol2: 0.06713\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10364.541\tTol1: 0.00002\tTol2: 0.06156\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10336.991\tTol1: 0.00266\tTol2: 0.04061\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10415.736\tTol1: 0.00762\tTol2: 0.05454\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10380.051\tTol1: 0.00343\tTol2: 0.09180\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10391.612\tTol1: 0.00111\tTol2: 0.05043\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10310.592\tTol1: 0.00780\tTol2: 0.06473\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10260.698\tTol1: 0.00484\tTol2: 0.03802\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10264.582\tTol1: 0.00038\tTol2: 0.02730\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.89 sec)\tlogL: -17517.033\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16505.296\tTol1: 0.05776\tTol2: 0.12391\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15809.127\tTol1: 0.04218\tTol2: 0.09807\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15326.218\tTol1: 0.03055\tTol2: 0.08781\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14941.386\tTol1: 0.02511\tTol2: 0.08486\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -14555.076\tTol1: 0.02586\tTol2: 0.07957\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14252.310\tTol1: 0.02080\tTol2: 0.06133\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14160.580\tTol1: 0.00644\tTol2: 0.05338\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13628.799\tTol1: 0.03755\tTol2: 0.11058\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13344.125\tTol1: 0.02089\tTol2: 0.06668\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12957.464\tTol1: 0.02898\tTol2: 0.06541\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12665.856\tTol1: 0.02251\tTol2: 0.06150\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12980.213\tTol1: 0.02482\tTol2: 0.06429\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12278.531\tTol1: 0.05406\tTol2: 0.20080\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12047.749\tTol1: 0.01880\tTol2: 0.05405\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11990.101\tTol1: 0.00478\tTol2: 0.08561\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11660.366\tTol1: 0.02750\tTol2: 0.12741\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11506.910\tTol1: 0.01316\tTol2: 0.07591\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11388.262\tTol1: 0.01031\tTol2: 0.04260\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11462.891\tTol1: 0.00655\tTol2: 0.05761\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11245.935\tTol1: 0.01893\tTol2: 0.12264\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11131.842\tTol1: 0.01015\tTol2: 0.04652\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11050.507\tTol1: 0.00731\tTol2: 0.06055\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11005.996\tTol1: 0.00403\tTol2: 0.04676\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11020.625\tTol1: 0.00133\tTol2: 0.07008\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11129.186\tTol1: 0.00985\tTol2: 0.08708\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10852.190\tTol1: 0.02489\tTol2: 0.11573\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10797.678\tTol1: 0.00502\tTol2: 0.07708\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10860.215\tTol1: 0.00579\tTol2: 0.08537\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10925.807\tTol1: 0.00604\tTol2: 0.10195\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10891.767\tTol1: 0.00312\tTol2: 0.10280\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -10724.013\tTol1: 0.01540\tTol2: 0.10424\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10800.292\tTol1: 0.00711\tTol2: 0.06849\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10648.213\tTol1: 0.01408\tTol2: 0.10429\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10781.067\tTol1: 0.01248\tTol2: 0.07974\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10626.620\tTol1: 0.01433\tTol2: 0.18464\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10542.611\tTol1: 0.00791\tTol2: 0.10197\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10546.979\tTol1: 0.00041\tTol2: 0.10648\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10561.426\tTol1: 0.00137\tTol2: 0.10537\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10442.552\tTol1: 0.01126\tTol2: 0.10854\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10412.840\tTol1: 0.00285\tTol2: 0.07385\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10520.218\tTol1: 0.01031\tTol2: 0.08040\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10355.135\tTol1: 0.01569\tTol2: 0.10601\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10309.642\tTol1: 0.00439\tTol2: 0.07354\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10316.251\tTol1: 0.00064\tTol2: 0.09137\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10357.916\tTol1: 0.00404\tTol2: 0.07930\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10268.437\tTol1: 0.00864\tTol2: 0.07151\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10368.027\tTol1: 0.00970\tTol2: 0.05808\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10290.060\tTol1: 0.00752\tTol2: 0.07227\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10241.087\tTol1: 0.00476\tTol2: 0.08074\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.93 sec)\tlogL: -17710.675\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16823.880\tTol1: 0.05007\tTol2: 0.16229\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -16269.640\tTol1: 0.03294\tTol2: 0.16791\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15755.216\tTol1: 0.03162\tTol2: 0.16433\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15464.572\tTol1: 0.01845\tTol2: 0.12786\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -14887.576\tTol1: 0.03731\tTol2: 0.11878\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14592.971\tTol1: 0.01979\tTol2: 0.09662\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14482.038\tTol1: 0.00760\tTol2: 0.10003\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13976.382\tTol1: 0.03492\tTol2: 0.18944\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13384.321\tTol1: 0.04236\tTol2: 0.13717\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13873.845\tTol1: 0.03657\tTol2: 0.08260\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -12909.264\tTol1: 0.06953\tTol2: 0.16867\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -12669.573\tTol1: 0.01857\tTol2: 0.08041\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12325.210\tTol1: 0.02718\tTol2: 0.06779\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -13187.938\tTol1: 0.07000\tTol2: 0.06993\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -12181.857\tTol1: 0.07629\tTol2: 0.19047\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11942.960\tTol1: 0.01961\tTol2: 0.08100\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11740.570\tTol1: 0.01695\tTol2: 0.05148\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11591.779\tTol1: 0.01267\tTol2: 0.05889\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11502.399\tTol1: 0.00771\tTol2: 0.05867\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -11586.673\tTol1: 0.00733\tTol2: 0.06741\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11499.964\tTol1: 0.00748\tTol2: 0.10779\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -11455.567\tTol1: 0.00386\tTol2: 0.13495\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -11201.421\tTol1: 0.02219\tTol2: 0.10027\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -11082.085\tTol1: 0.01065\tTol2: 0.05936\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.79 sec)\tlogL: -11182.961\tTol1: 0.00910\tTol2: 0.04647\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -11141.233\tTol1: 0.00373\tTol2: 0.07250\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11068.703\tTol1: 0.00651\tTol2: 0.05817\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10999.968\tTol1: 0.00621\tTol2: 0.06034\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10924.613\tTol1: 0.00685\tTol2: 0.14775\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10873.984\tTol1: 0.00463\tTol2: 0.07073\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10820.241\tTol1: 0.00494\tTol2: 0.12227\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10747.559\tTol1: 0.00672\tTol2: 0.11869\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10845.750\tTol1: 0.00914\tTol2: 0.08953\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10797.340\tTol1: 0.00446\tTol2: 0.09967\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10800.191\tTol1: 0.00026\tTol2: 0.16339\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10654.421\tTol1: 0.01350\tTol2: 0.09978\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10603.207\tTol1: 0.00481\tTol2: 0.11003\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10586.884\tTol1: 0.00154\tTol2: 0.10670\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10599.354\tTol1: 0.00118\tTol2: 0.10735\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.78 sec)\tlogL: -10517.112\tTol1: 0.00776\tTol2: 0.10883\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10524.433\tTol1: 0.00070\tTol2: 0.09997\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10526.712\tTol1: 0.00022\tTol2: 0.12978\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10464.143\tTol1: 0.00594\tTol2: 0.13057\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10481.326\tTol1: 0.00164\tTol2: 0.11085\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10527.612\tTol1: 0.00442\tTol2: 0.14408\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -10472.754\tTol1: 0.00521\tTol2: 0.16896\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10405.631\tTol1: 0.00641\tTol2: 0.14832\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10344.783\tTol1: 0.00585\tTol2: 0.12932\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10295.652\tTol1: 0.00475\tTol2: 0.13219\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.02 sec)\tlogL: -17449.417\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16621.661\tTol1: 0.04744\tTol2: 0.11132\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16089.047\tTol1: 0.03204\tTol2: 0.11923\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -15479.816\tTol1: 0.03787\tTol2: 0.08751\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15008.575\tTol1: 0.03044\tTol2: 0.09936\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14707.313\tTol1: 0.02007\tTol2: 0.10721\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14968.993\tTol1: 0.01779\tTol2: 0.07719\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14153.282\tTol1: 0.05449\tTol2: 0.12173\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13810.852\tTol1: 0.02419\tTol2: 0.14072\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13806.385\tTol1: 0.00032\tTol2: 0.09304\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13175.024\tTol1: 0.04573\tTol2: 0.09535\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13128.669\tTol1: 0.00352\tTol2: 0.10634\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12449.908\tTol1: 0.05170\tTol2: 0.20203\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12212.877\tTol1: 0.01904\tTol2: 0.06292\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12345.594\tTol1: 0.01087\tTol2: 0.07792\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12000.638\tTol1: 0.02794\tTol2: 0.10695\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12409.691\tTol1: 0.03409\tTol2: 0.07973\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11948.899\tTol1: 0.03713\tTol2: 0.10841\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11851.060\tTol1: 0.00819\tTol2: 0.09977\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11514.778\tTol1: 0.02838\tTol2: 0.09147\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -11443.977\tTol1: 0.00615\tTol2: 0.14508\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11408.819\tTol1: 0.00307\tTol2: 0.07410\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11259.647\tTol1: 0.01308\tTol2: 0.06560\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11390.543\tTol1: 0.01163\tTol2: 0.09141\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -11065.805\tTol1: 0.02851\tTol2: 0.15722\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10990.053\tTol1: 0.00685\tTol2: 0.07266\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10939.754\tTol1: 0.00458\tTol2: 0.07510\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10993.626\tTol1: 0.00492\tTol2: 0.05859\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10819.000\tTol1: 0.01588\tTol2: 0.10239\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10807.125\tTol1: 0.00110\tTol2: 0.08401\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10737.465\tTol1: 0.00645\tTol2: 0.07793\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10744.721\tTol1: 0.00068\tTol2: 0.07793\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10632.667\tTol1: 0.01043\tTol2: 0.06265\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10747.859\tTol1: 0.01083\tTol2: 0.05691\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10614.205\tTol1: 0.01244\tTol2: 0.09620\t\n",
      "Iter 36/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10514.895\tTol1: 0.00936\tTol2: 0.08922\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10593.756\tTol1: 0.00750\tTol2: 0.05178\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10502.738\tTol1: 0.00859\tTol2: 0.07326\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10508.689\tTol1: 0.00057\tTol2: 0.07118\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10449.473\tTol1: 0.00563\tTol2: 0.05546\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10423.369\tTol1: 0.00250\tTol2: 0.07132\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10442.559\tTol1: 0.00184\tTol2: 0.05270\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10369.557\tTol1: 0.00699\tTol2: 0.04468\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10342.028\tTol1: 0.00265\tTol2: 0.06391\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10304.318\tTol1: 0.00365\tTol2: 0.03665\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10294.520\tTol1: 0.00095\tTol2: 0.04818\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10378.418\tTol1: 0.00815\tTol2: 0.04289\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10305.894\tTol1: 0.00699\tTol2: 0.05183\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10207.499\tTol1: 0.00955\tTol2: 0.07577\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10171.314\tTol1: 0.00354\tTol2: 0.03527\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.95 sec)\tlogL: -17512.302\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16995.749\tTol1: 0.02950\tTol2: 0.08077\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16136.623\tTol1: 0.05055\tTol2: 0.10829\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -15781.084\tTol1: 0.02203\tTol2: 0.08824\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15255.301\tTol1: 0.03332\tTol2: 0.07018\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14846.131\tTol1: 0.02682\tTol2: 0.06324\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14498.372\tTol1: 0.02342\tTol2: 0.04373\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14432.859\tTol1: 0.00452\tTol2: 0.03808\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -13955.048\tTol1: 0.03311\tTol2: 0.12359\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13640.234\tTol1: 0.02256\tTol2: 0.06505\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13428.763\tTol1: 0.01550\tTol2: 0.05658\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13000.118\tTol1: 0.03192\tTol2: 0.08741\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12659.208\tTol1: 0.02622\tTol2: 0.04620\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12928.314\tTol1: 0.02126\tTol2: 0.01581\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.93 sec)\tlogL: -17274.558\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16437.764\tTol1: 0.04844\tTol2: 0.15428\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15991.575\tTol1: 0.02714\tTol2: 0.11582\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15465.090\tTol1: 0.03292\tTol2: 0.12636\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15598.312\tTol1: 0.00861\tTol2: 0.11350\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14976.352\tTol1: 0.03987\tTol2: 0.14627\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14571.897\tTol1: 0.02701\tTol2: 0.10310\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14596.759\tTol1: 0.00171\tTol2: 0.13016\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13997.544\tTol1: 0.04105\tTol2: 0.12255\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13648.362\tTol1: 0.02495\tTol2: 0.11871\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13521.024\tTol1: 0.00933\tTol2: 0.13722\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13692.768\tTol1: 0.01270\tTol2: 0.11753\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13062.044\tTol1: 0.04606\tTol2: 0.16019\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12721.442\tTol1: 0.02608\tTol2: 0.12414\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -12917.555\tTol1: 0.01542\tTol2: 0.20783\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12853.213\tTol1: 0.00498\tTol2: 0.22808\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -12170.915\tTol1: 0.05308\tTol2: 0.21952\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12205.523\tTol1: 0.00284\tTol2: 0.10826\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11716.007\tTol1: 0.04011\tTol2: 0.15865\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -11526.293\tTol1: 0.01619\tTol2: 0.09448\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -11421.985\tTol1: 0.00905\tTol2: 0.07355\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11526.199\tTol1: 0.00912\tTol2: 0.06631\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11336.037\tTol1: 0.01650\tTol2: 0.09364\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11221.885\tTol1: 0.01007\tTol2: 0.06523\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -11132.565\tTol1: 0.00796\tTol2: 0.06276\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11061.557\tTol1: 0.00638\tTol2: 0.06959\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11069.776\tTol1: 0.00074\tTol2: 0.06300\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11145.971\tTol1: 0.00688\tTol2: 0.06868\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10970.257\tTol1: 0.01576\tTol2: 0.12237\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10871.493\tTol1: 0.00900\tTol2: 0.06336\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10845.753\tTol1: 0.00237\tTol2: 0.05826\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10821.359\tTol1: 0.00225\tTol2: 0.04513\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10799.332\tTol1: 0.00204\tTol2: 0.04727\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10746.746\tTol1: 0.00487\tTol2: 0.05704\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10765.775\tTol1: 0.00177\tTol2: 0.04475\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10667.498\tTol1: 0.00913\tTol2: 0.06236\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10677.456\tTol1: 0.00093\tTol2: 0.04419\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10641.823\tTol1: 0.00334\tTol2: 0.06429\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10540.603\tTol1: 0.00951\tTol2: 0.04684\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10720.138\tTol1: 0.01703\tTol2: 0.03962\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10911.319\tTol1: 0.01783\tTol2: 0.06880\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10576.063\tTol1: 0.03073\tTol2: 0.09186\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10485.581\tTol1: 0.00856\tTol2: 0.03358\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10508.043\tTol1: 0.00214\tTol2: 0.03219\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10381.185\tTol1: 0.01207\tTol2: 0.10456\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10275.999\tTol1: 0.01013\tTol2: 0.03144\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10283.902\tTol1: 0.00077\tTol2: 0.02673\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.15 sec)\tlogL: -17100.196\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -16443.999\tTol1: 0.03837\tTol2: 0.30588\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15887.017\tTol1: 0.03387\tTol2: 0.18464\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15473.834\tTol1: 0.02601\tTol2: 0.18075\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -15156.440\tTol1: 0.02051\tTol2: 0.08997\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -14771.109\tTol1: 0.02542\tTol2: 0.08689\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14535.388\tTol1: 0.01596\tTol2: 0.05284\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14299.403\tTol1: 0.01624\tTol2: 0.05492\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -14142.382\tTol1: 0.01098\tTol2: 0.09444\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13647.471\tTol1: 0.03499\tTol2: 0.08223\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -13293.754\tTol1: 0.02592\tTol2: 0.12212\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -13111.497\tTol1: 0.01371\tTol2: 0.10736\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13135.932\tTol1: 0.00186\tTol2: 0.11711\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12628.483\tTol1: 0.03863\tTol2: 0.13946\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12245.937\tTol1: 0.03029\tTol2: 0.07327\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12067.129\tTol1: 0.01460\tTol2: 0.05916\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11846.331\tTol1: 0.01830\tTol2: 0.06153\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11607.425\tTol1: 0.02017\tTol2: 0.04871\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -11440.870\tTol1: 0.01435\tTol2: 0.07117\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11328.561\tTol1: 0.00982\tTol2: 0.05636\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11286.847\tTol1: 0.00368\tTol2: 0.04494\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11314.493\tTol1: 0.00245\tTol2: 0.06122\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11556.745\tTol1: 0.02141\tTol2: 0.07637\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -11479.842\tTol1: 0.00665\tTol2: 0.11889\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11387.623\tTol1: 0.00803\tTol2: 0.10785\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11151.249\tTol1: 0.02076\tTol2: 0.08637\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11051.214\tTol1: 0.00897\tTol2: 0.07962\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11014.038\tTol1: 0.00336\tTol2: 0.10025\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10794.530\tTol1: 0.01993\tTol2: 0.07041\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10850.577\tTol1: 0.00519\tTol2: 0.06021\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10742.377\tTol1: 0.00997\tTol2: 0.09151\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10753.160\tTol1: 0.00100\tTol2: 0.06039\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10669.019\tTol1: 0.00782\tTol2: 0.04764\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10628.988\tTol1: 0.00375\tTol2: 0.06565\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10632.301\tTol1: 0.00031\tTol2: 0.07931\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10544.476\tTol1: 0.00826\tTol2: 0.06715\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10734.405\tTol1: 0.01801\tTol2: 0.05566\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10910.854\tTol1: 0.01644\tTol2: 0.08777\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10525.888\tTol1: 0.03528\tTol2: 0.09335\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10511.773\tTol1: 0.00134\tTol2: 0.08479\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10421.258\tTol1: 0.00861\tTol2: 0.04337\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10337.272\tTol1: 0.00806\tTol2: 0.05680\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10287.347\tTol1: 0.00483\tTol2: 0.08471\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10273.688\tTol1: 0.00133\tTol2: 0.08105\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10211.264\tTol1: 0.00608\tTol2: 0.07095\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10218.336\tTol1: 0.00069\tTol2: 0.07061\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10611.386\tTol1: 0.03847\tTol2: 0.08753\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10192.657\tTol1: 0.03946\tTol2: 0.11974\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10685.750\tTol1: 0.04838\tTol2: 0.05875\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10247.462\tTol1: 0.04102\tTol2: 0.11603\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.95 sec)\tlogL: -17355.103\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -16688.002\tTol1: 0.03844\tTol2: 0.08242\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -16057.670\tTol1: 0.03777\tTol2: 0.11752\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15666.480\tTol1: 0.02436\tTol2: 0.10150\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15377.316\tTol1: 0.01846\tTol2: 0.09391\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14794.581\tTol1: 0.03790\tTol2: 0.16541\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14574.902\tTol1: 0.01485\tTol2: 0.07733\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14184.954\tTol1: 0.02675\tTol2: 0.07891\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14083.615\tTol1: 0.00714\tTol2: 0.04317\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13475.687\tTol1: 0.04317\tTol2: 0.15643\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13248.932\tTol1: 0.01683\tTol2: 0.03632\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13113.287\tTol1: 0.01024\tTol2: 0.05459\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13286.959\tTol1: 0.01324\tTol2: 0.10176\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12536.031\tTol1: 0.05652\tTol2: 0.16781\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12259.611\tTol1: 0.02205\tTol2: 0.07265\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12087.739\tTol1: 0.01402\tTol2: 0.08354\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11931.083\tTol1: 0.01296\tTol2: 0.05853\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11800.142\tTol1: 0.01097\tTol2: 0.04722\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11612.345\tTol1: 0.01591\tTol2: 0.06138\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11592.931\tTol1: 0.00167\tTol2: 0.05288\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11451.464\tTol1: 0.01220\tTol2: 0.08270\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11406.857\tTol1: 0.00390\tTol2: 0.05047\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11459.554\tTol1: 0.00462\tTol2: 0.04630\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11277.419\tTol1: 0.01589\tTol2: 0.13239\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11358.997\tTol1: 0.00723\tTol2: 0.07544\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11292.761\tTol1: 0.00583\tTol2: 0.11603\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11096.075\tTol1: 0.01742\tTol2: 0.08997\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11121.311\tTol1: 0.00227\tTol2: 0.04806\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11001.437\tTol1: 0.01078\tTol2: 0.07667\t\n",
      "Iter 30/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10914.803\tTol1: 0.00787\tTol2: 0.05506\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10994.960\tTol1: 0.00734\tTol2: 0.07503\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11037.262\tTol1: 0.00385\tTol2: 0.09897\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10819.825\tTol1: 0.01970\tTol2: 0.09698\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10783.912\tTol1: 0.00332\tTol2: 0.05872\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10728.285\tTol1: 0.00516\tTol2: 0.04840\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10678.789\tTol1: 0.00461\tTol2: 0.06373\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10685.184\tTol1: 0.00060\tTol2: 0.06413\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -10591.921\tTol1: 0.00873\tTol2: 0.07616\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.78 sec)\tlogL: -10551.089\tTol1: 0.00385\tTol2: 0.05587\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10528.925\tTol1: 0.00210\tTol2: 0.06625\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10655.570\tTol1: 0.01203\tTol2: 0.07941\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10749.862\tTol1: 0.00885\tTol2: 0.12458\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11395.755\tTol1: 0.06008\tTol2: 0.14569\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10620.954\tTol1: 0.06799\tTol2: 0.29702\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10436.186\tTol1: 0.01740\tTol2: 0.09993\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10386.580\tTol1: 0.00475\tTol2: 0.07280\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10342.382\tTol1: 0.00426\tTol2: 0.08807\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10311.099\tTol1: 0.00302\tTol2: 0.07292\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10294.016\tTol1: 0.00166\tTol2: 0.07416\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10351.712\tTol1: 0.00560\tTol2: 0.09034\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.98 sec)\tlogL: -17381.296\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -16464.061\tTol1: 0.05277\tTol2: 0.08991\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15822.257\tTol1: 0.03898\tTol2: 0.07128\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -15999.107\tTol1: 0.01118\tTol2: 0.07045\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -15160.920\tTol1: 0.05239\tTol2: 0.19515\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14843.044\tTol1: 0.02097\tTol2: 0.05326\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14439.855\tTol1: 0.02716\tTol2: 0.06207\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14504.368\tTol1: 0.00447\tTol2: 0.04956\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14041.781\tTol1: 0.03189\tTol2: 0.09822\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -13556.334\tTol1: 0.03457\tTol2: 0.03775\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13191.919\tTol1: 0.02688\tTol2: 0.11454\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -13054.899\tTol1: 0.01039\tTol2: 0.03090\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12738.052\tTol1: 0.02427\tTol2: 0.06181\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12383.438\tTol1: 0.02784\tTol2: 0.03855\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -12297.198\tTol1: 0.00696\tTol2: 0.04136\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12000.260\tTol1: 0.02415\tTol2: 0.11321\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11847.198\tTol1: 0.01275\tTol2: 0.03306\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11730.560\tTol1: 0.00985\tTol2: 0.06931\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11543.490\tTol1: 0.01595\tTol2: 0.07209\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11440.059\tTol1: 0.00896\tTol2: 0.05656\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11301.097\tTol1: 0.01215\tTol2: 0.05016\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11263.341\tTol1: 0.00334\tTol2: 0.03372\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11172.808\tTol1: 0.00804\tTol2: 0.05483\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11182.377\tTol1: 0.00086\tTol2: 0.03065\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11184.443\tTol1: 0.00018\tTol2: 0.02135\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.95 sec)\tlogL: -17322.708\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -16550.620\tTol1: 0.04457\tTol2: 0.17283\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15865.753\tTol1: 0.04138\tTol2: 0.14751\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15456.416\tTol1: 0.02580\tTol2: 0.11620\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15194.398\tTol1: 0.01695\tTol2: 0.11406\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14820.438\tTol1: 0.02461\tTol2: 0.10612\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14811.841\tTol1: 0.00058\tTol2: 0.10744\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14232.634\tTol1: 0.03910\tTol2: 0.16685\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13903.661\tTol1: 0.02311\tTol2: 0.10460\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -13580.815\tTol1: 0.02322\tTol2: 0.09633\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13388.478\tTol1: 0.01416\tTol2: 0.08006\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12909.746\tTol1: 0.03576\tTol2: 0.13444\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -13025.819\tTol1: 0.00899\tTol2: 0.09625\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -12542.285\tTol1: 0.03712\tTol2: 0.13546\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12121.400\tTol1: 0.03356\tTol2: 0.10798\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12079.466\tTol1: 0.00346\tTol2: 0.09295\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11762.643\tTol1: 0.02623\tTol2: 0.10010\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11636.239\tTol1: 0.01075\tTol2: 0.19357\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11585.415\tTol1: 0.00437\tTol2: 0.15886\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11351.722\tTol1: 0.02017\tTol2: 0.12093\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11307.069\tTol1: 0.00393\tTol2: 0.09626\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11245.090\tTol1: 0.00548\tTol2: 0.11220\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11344.790\tTol1: 0.00887\tTol2: 0.12372\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11213.020\tTol1: 0.01162\tTol2: 0.18128\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11045.196\tTol1: 0.01497\tTol2: 0.11548\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10963.526\tTol1: 0.00739\tTol2: 0.10789\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10897.719\tTol1: 0.00600\tTol2: 0.05655\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10934.414\tTol1: 0.00337\tTol2: 0.04592\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10841.353\tTol1: 0.00851\tTol2: 0.06809\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11115.996\tTol1: 0.02533\tTol2: 0.03231\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10786.707\tTol1: 0.02962\tTol2: 0.10276\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10708.873\tTol1: 0.00722\tTol2: 0.04328\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10712.096\tTol1: 0.00030\tTol2: 0.02749\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.93 sec)\tlogL: -17401.280\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16698.436\tTol1: 0.04039\tTol2: 0.07598\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -16061.737\tTol1: 0.03813\tTol2: 0.09845\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15642.754\tTol1: 0.02609\tTol2: 0.09377\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -15373.967\tTol1: 0.01718\tTol2: 0.09635\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -14868.756\tTol1: 0.03286\tTol2: 0.14221\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14534.159\tTol1: 0.02250\tTol2: 0.09941\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -14228.996\tTol1: 0.02100\tTol2: 0.10473\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14020.194\tTol1: 0.01467\tTol2: 0.10849\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -13642.912\tTol1: 0.02691\tTol2: 0.11144\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -13344.709\tTol1: 0.02186\tTol2: 0.10250\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13009.816\tTol1: 0.02510\tTol2: 0.09183\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -12765.211\tTol1: 0.01880\tTol2: 0.09712\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12886.733\tTol1: 0.00952\tTol2: 0.12572\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12866.838\tTol1: 0.00154\tTol2: 0.16338\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12350.911\tTol1: 0.04010\tTol2: 0.16413\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12385.474\tTol1: 0.00280\tTol2: 0.18576\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11909.472\tTol1: 0.03843\tTol2: 0.18572\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11698.143\tTol1: 0.01774\tTol2: 0.09018\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11609.213\tTol1: 0.00760\tTol2: 0.09638\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11568.766\tTol1: 0.00348\tTol2: 0.09292\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11393.500\tTol1: 0.01515\tTol2: 0.11240\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11440.339\tTol1: 0.00411\tTol2: 0.08389\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11239.551\tTol1: 0.01755\tTol2: 0.12638\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11173.669\tTol1: 0.00586\tTol2: 0.08565\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11168.707\tTol1: 0.00044\tTol2: 0.07993\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11268.664\tTol1: 0.00895\tTol2: 0.09021\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11046.442\tTol1: 0.01972\tTol2: 0.10790\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10990.247\tTol1: 0.00509\tTol2: 0.06193\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10960.155\tTol1: 0.00274\tTol2: 0.06706\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10959.614\tTol1: 0.00005\tTol2: 0.07512\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11010.451\tTol1: 0.00464\tTol2: 0.07052\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10830.049\tTol1: 0.01638\tTol2: 0.12542\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10729.115\tTol1: 0.00932\tTol2: 0.04667\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10794.618\tTol1: 0.00611\tTol2: 0.05051\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10930.073\tTol1: 0.01255\tTol2: 0.10103\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10669.206\tTol1: 0.02387\tTol2: 0.12674\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10706.752\tTol1: 0.00352\tTol2: 0.04261\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10627.661\tTol1: 0.00739\tTol2: 0.06080\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10551.288\tTol1: 0.00719\tTol2: 0.05460\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10527.915\tTol1: 0.00222\tTol2: 0.03718\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10513.857\tTol1: 0.00134\tTol2: 0.02919\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.98 sec)\tlogL: -17430.235\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -16688.148\tTol1: 0.04257\tTol2: 0.23246\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -15967.141\tTol1: 0.04320\tTol2: 0.14196\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -15520.936\tTol1: 0.02795\tTol2: 0.12939\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -15265.648\tTol1: 0.01645\tTol2: 0.12877\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -14892.328\tTol1: 0.02445\tTol2: 0.11708\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14650.219\tTol1: 0.01626\tTol2: 0.07778\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -14728.870\tTol1: 0.00537\tTol2: 0.09703\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14196.185\tTol1: 0.03617\tTol2: 0.22531\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13719.075\tTol1: 0.03361\tTol2: 0.14068\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13296.790\tTol1: 0.03078\tTol2: 0.10571\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -13183.826\tTol1: 0.00850\tTol2: 0.08367\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -14178.296\tTol1: 0.07543\tTol2: 0.16664\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -12590.741\tTol1: 0.11197\tTol2: 0.31118\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12541.487\tTol1: 0.00391\tTol2: 0.10439\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -12133.142\tTol1: 0.03256\tTol2: 0.13348\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -12242.785\tTol1: 0.00904\tTol2: 0.12397\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -12081.493\tTol1: 0.01317\tTol2: 0.15334\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11771.298\tTol1: 0.02568\tTol2: 0.13267\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -11600.491\tTol1: 0.01451\tTol2: 0.10942\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11533.786\tTol1: 0.00575\tTol2: 0.09028\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11388.414\tTol1: 0.01260\tTol2: 0.09549\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11333.631\tTol1: 0.00481\tTol2: 0.07473\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11161.795\tTol1: 0.01516\tTol2: 0.09804\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11256.331\tTol1: 0.00847\tTol2: 0.07632\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11105.232\tTol1: 0.01342\tTol2: 0.09842\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11007.907\tTol1: 0.00876\tTol2: 0.08758\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -11046.805\tTol1: 0.00353\tTol2: 0.06000\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10981.096\tTol1: 0.00595\tTol2: 0.10195\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -10929.382\tTol1: 0.00471\tTol2: 0.09669\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10838.046\tTol1: 0.00836\tTol2: 0.09044\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10774.476\tTol1: 0.00587\tTol2: 0.07079\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10934.927\tTol1: 0.01489\tTol2: 0.04970\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10718.058\tTol1: 0.01983\tTol2: 0.11006\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10710.463\tTol1: 0.00071\tTol2: 0.04473\t\n",
      "Iter 36/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10609.795\tTol1: 0.00940\tTol2: 0.08149\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10586.639\tTol1: 0.00218\tTol2: 0.03998\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10612.661\tTol1: 0.00246\tTol2: 0.05109\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10826.019\tTol1: 0.02010\tTol2: 0.06788\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10519.411\tTol1: 0.02832\tTol2: 0.15277\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10438.893\tTol1: 0.00765\tTol2: 0.05857\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10434.419\tTol1: 0.00043\tTol2: 0.05205\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10530.786\tTol1: 0.00924\tTol2: 0.06212\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10352.334\tTol1: 0.01695\tTol2: 0.10740\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10331.699\tTol1: 0.00199\tTol2: 0.04777\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10377.537\tTol1: 0.00444\tTol2: 0.04172\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10193.889\tTol1: 0.01770\tTol2: 0.07978\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10186.120\tTol1: 0.00076\tTol2: 0.05843\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10175.590\tTol1: 0.00103\tTol2: 0.05098\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10475.146\tTol1: 0.02944\tTol2: 0.04348\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (27746, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.10 sec)\tlogL: -17255.865\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -16226.875\tTol1: 0.05963\tTol2: 0.23685\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -15970.986\tTol1: 0.01577\tTol2: 0.09736\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -15216.923\tTol1: 0.04721\tTol2: 0.21068\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -14969.754\tTol1: 0.01624\tTol2: 0.11151\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -14675.382\tTol1: 0.01966\tTol2: 0.11623\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -14478.678\tTol1: 0.01340\tTol2: 0.11687\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -14042.746\tTol1: 0.03011\tTol2: 0.11187\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13755.465\tTol1: 0.02046\tTol2: 0.07779\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13384.646\tTol1: 0.02696\tTol2: 0.09288\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13259.271\tTol1: 0.00937\tTol2: 0.05166\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -12813.025\tTol1: 0.03366\tTol2: 0.10390\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -12592.099\tTol1: 0.01724\tTol2: 0.05707\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -13376.225\tTol1: 0.06227\tTol2: 0.04975\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12639.130\tTol1: 0.05510\tTol2: 0.20624\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -12627.453\tTol1: 0.00092\tTol2: 0.20670\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -12092.173\tTol1: 0.04239\tTol2: 0.14329\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11923.569\tTol1: 0.01394\tTol2: 0.09023\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11773.452\tTol1: 0.01259\tTol2: 0.08100\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11916.342\tTol1: 0.01214\tTol2: 0.07748\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -11643.240\tTol1: 0.02292\tTol2: 0.12993\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11554.845\tTol1: 0.00759\tTol2: 0.11013\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11431.512\tTol1: 0.01067\tTol2: 0.09652\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11331.904\tTol1: 0.00871\tTol2: 0.10864\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11265.032\tTol1: 0.00590\tTol2: 0.08194\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.77 sec)\tlogL: -11215.505\tTol1: 0.00440\tTol2: 0.09143\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -11140.575\tTol1: 0.00668\tTol2: 0.09719\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -11222.203\tTol1: 0.00733\tTol2: 0.07736\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11054.018\tTol1: 0.01499\tTol2: 0.09877\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -11007.637\tTol1: 0.00420\tTol2: 0.09147\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.78 sec)\tlogL: -10852.230\tTol1: 0.01412\tTol2: 0.19338\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10802.598\tTol1: 0.00457\tTol2: 0.12644\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10894.738\tTol1: 0.00853\tTol2: 0.14636\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.81 sec)\tlogL: -10741.677\tTol1: 0.01405\tTol2: 0.08475\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10675.180\tTol1: 0.00619\tTol2: 0.08539\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10608.328\tTol1: 0.00626\tTol2: 0.07366\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.76 sec)\tlogL: -10551.984\tTol1: 0.00531\tTol2: 0.12630\t\n",
      "Iter 38/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10489.072\tTol1: 0.00596\tTol2: 0.06672\t\n",
      "Iter 39/50\n",
      "M step: done,  E step: done //  (in 0.72 sec)\tlogL: -10516.239\tTol1: 0.00259\tTol2: 0.07006\t\n",
      "Iter 40/50\n",
      "M step: done,  E step: done //  (in 0.71 sec)\tlogL: -10415.441\tTol1: 0.00958\tTol2: 0.06559\t\n",
      "Iter 41/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10401.416\tTol1: 0.00135\tTol2: 0.10970\t\n",
      "Iter 42/50\n",
      "M step: done,  E step: done //  (in 0.75 sec)\tlogL: -10381.683\tTol1: 0.00190\tTol2: 0.09324\t\n",
      "Iter 43/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10346.218\tTol1: 0.00342\tTol2: 0.06730\t\n",
      "Iter 44/50\n",
      "M step: done,  E step: done //  (in 0.74 sec)\tlogL: -10329.577\tTol1: 0.00161\tTol2: 0.05135\t\n",
      "Iter 45/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10242.700\tTol1: 0.00841\tTol2: 0.09104\t\n",
      "Iter 46/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10222.875\tTol1: 0.00194\tTol2: 0.05835\t\n",
      "Iter 47/50\n",
      "M step: done,  E step: done //  (in 0.73 sec)\tlogL: -10228.462\tTol1: 0.00055\tTol2: 0.08049\t\n",
      "Iter 48/50\n",
      "M step: done,  E step: done //  (in 0.70 sec)\tlogL: -10156.938\tTol1: 0.00699\tTol2: 0.10720\t\n",
      "Iter 49/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10160.659\tTol1: 0.00037\tTol2: 0.07972\t\n",
      "Iter 50/50\n",
      "M step: done,  E step: done //  (in 0.69 sec)\tlogL: -10186.846\tTol1: 0.00258\tTol2: 0.04999\t\n",
      "Finished training!\n",
      "20 runs over C-MoA, Epochs to converge=  42.15\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "([array([-17625.4541, -16747.9864, -15980.623 , -15282.4352, -14951.5064,\n",
       "         -14578.7287, -14734.5259, -14090.0473, -13746.2639, -13756.8577,\n",
       "         -13291.5975, -12863.333 , -12594.4729, -12305.9739, -12189.4311,\n",
       "         -11902.6715, -11762.3206, -11613.7582, -11469.3497, -11359.8515,\n",
       "         -11422.5503, -11418.3668, -11262.3341, -11080.0989, -10988.2448,\n",
       "         -10938.6245, -10888.8913, -10872.9948]),\n",
       "  array([-17661.371 , -16756.0176, -16253.6821, -15578.6273, -15100.7578,\n",
       "         -14729.0524, -14373.769 , -14061.6267, -13983.7633, -13521.8135,\n",
       "         -13481.4125]),\n",
       "  array([-17840.8097, -16855.9405, -16297.6582, -15764.6711, -15267.7715,\n",
       "         -14862.4884, -14571.509 , -14321.3563, -14120.821 , -13831.9365,\n",
       "         -13397.6454, -13082.6896, -12767.2367, -13207.2546, -12405.8701,\n",
       "         -12172.4767, -11932.8185, -11986.1839, -11697.0495, -11562.1895,\n",
       "         -11854.1394, -11542.0218, -11628.7363, -11298.1075, -11231.3336,\n",
       "         -11111.6337, -11036.8799, -11379.8078, -11039.2349, -11104.5589,\n",
       "         -10946.0408, -10826.1218, -10852.8111, -10745.4332, -10710.8839,\n",
       "         -11064.3659, -10952.254 , -10633.7163, -10592.0279, -10540.1541,\n",
       "         -10638.8999, -10502.3259, -10475.7286, -10405.7982, -10485.1844,\n",
       "         -10368.3554, -10358.7373]),\n",
       "  array([-17250.5433, -16446.1293, -15863.8843, -15510.4162, -15275.2441,\n",
       "         -14791.8346, -14461.2696, -14173.8934, -13942.2484, -13852.6502,\n",
       "         -13312.8355, -13019.5218, -13286.5558, -12468.018 , -12627.5575,\n",
       "         -12248.613 , -11970.0681, -11726.6806, -11621.0434, -11787.4823,\n",
       "         -11368.5213, -11270.5468, -11258.4005, -11245.2859, -11330.7562,\n",
       "         -11205.9171, -11062.7382, -11142.3392, -10952.5005, -10954.107 ,\n",
       "         -10823.9616, -10807.5058, -10755.8074, -10700.9465, -10660.8562,\n",
       "         -10607.4086, -10586.2435, -10644.637 , -10488.9344, -10480.8502,\n",
       "         -10447.7207, -10382.6719, -10343.955 , -10401.8465, -10363.0073,\n",
       "         -10296.3622, -10332.5686, -10268.7392, -10239.1454, -10244.6882]),\n",
       "  array([-17310.9498, -16534.2736, -15895.8477, -15453.1692, -15020.9079,\n",
       "         -15057.0896, -14485.1973, -14124.9094, -13840.9356, -13524.0492,\n",
       "         -14501.9439, -12999.2367, -12774.0379, -12475.9506, -12472.2983,\n",
       "         -12082.7295, -11977.2421, -11861.5927, -11782.2663, -11654.4364,\n",
       "         -11553.4353, -11912.7373, -11405.2712, -11405.509 , -11256.2466,\n",
       "         -11180.7663, -11061.6415, -11105.0929, -11166.8383, -10923.0643,\n",
       "         -10845.9982, -10796.2402, -10764.9895, -10706.1605, -10815.5001,\n",
       "         -10676.6157, -10599.0527, -10555.2045, -10513.8062, -10855.6898,\n",
       "         -10462.6234, -10408.5876, -10372.0883, -10391.7064, -10623.4392,\n",
       "         -10407.2579, -10289.7281, -10222.9126, -10316.5056, -10188.4404]),\n",
       "  array([-17665.3125, -16808.9422, -16085.0833, -15488.49  , -15112.4552,\n",
       "         -14826.1887, -14621.5829, -14211.7659, -13938.4359, -13621.9249,\n",
       "         -13682.027 , -13028.1737, -13448.4221, -12503.6663, -12271.8545,\n",
       "         -12024.6382, -12270.1916, -11907.6657, -11805.0259, -11632.0364,\n",
       "         -11444.3999, -11422.1582, -11268.6072, -11193.5074, -11071.0704,\n",
       "         -11171.6925, -11055.8557, -10947.0645, -10927.1919, -10911.1095,\n",
       "         -11404.9737, -10918.7589, -10824.5609, -10861.8844, -10957.5972,\n",
       "         -10758.9933, -10689.2702, -10655.1637, -10574.9659, -10505.0064,\n",
       "         -10467.2066, -10450.1214, -10559.8009, -10482.288 , -10406.9789,\n",
       "         -10378.1328, -10458.8494, -10328.254 , -10348.9406, -10349.3771]),\n",
       "  array([-17455.398 , -16532.0252, -15994.6705, -15364.1764, -14939.6541,\n",
       "         -15167.7995, -14391.5976, -14147.8064, -13765.4257, -13413.6384,\n",
       "         -13179.6644, -12730.0931, -12654.7286, -12281.2914, -12061.1946,\n",
       "         -11809.6703, -11682.562 , -11607.5822, -11468.0621, -11535.7256,\n",
       "         -11277.3586, -11046.3075, -10992.6781, -11062.5633, -11002.8412,\n",
       "         -10798.4805, -10788.1278, -10758.1839, -10745.3233, -10653.822 ,\n",
       "         -10743.6383, -10606.6893, -10675.8143, -10578.9356, -10647.8548,\n",
       "         -10546.5549, -10472.041 , -10422.0542, -10448.3467, -10360.9856,\n",
       "         -10435.6822, -10454.0261, -10524.3561, -10297.4503, -10265.3752,\n",
       "         -10210.4472, -10182.7279, -10194.6871, -10146.7479, -10257.5642]),\n",
       "  array([-17880.732 , -16961.1585, -16225.8995, -15579.8743, -15137.1387,\n",
       "         -14816.9078, -14286.1873, -14295.7855, -13577.2298, -13233.4963,\n",
       "         -12897.0246, -12837.3121, -12327.8297, -12294.743 , -12036.7152,\n",
       "         -11879.6717, -11564.0192, -11338.8821, -11142.2524, -11095.2979,\n",
       "         -10971.4778, -10854.2112, -10840.7142, -10837.7799, -11324.8009,\n",
       "         -10921.7726, -10748.7326, -10644.0525, -10675.6018, -10555.7197,\n",
       "         -10472.8508, -10488.5696, -10541.6561, -10465.0705, -10410.6898,\n",
       "         -10376.8104, -10376.667 , -10364.7874, -10364.5414, -10336.991 ,\n",
       "         -10415.7364, -10380.0506, -10391.6122, -10310.5917, -10260.6977,\n",
       "         -10264.582 ]),\n",
       "  array([-17517.0328, -16505.2957, -15809.127 , -15326.2182, -14941.3862,\n",
       "         -14555.0763, -14252.3103, -14160.5801, -13628.799 , -13344.125 ,\n",
       "         -12957.4641, -12665.8559, -12980.2133, -12278.5308, -12047.749 ,\n",
       "         -11990.1012, -11660.3657, -11506.9098, -11388.2624, -11462.8911,\n",
       "         -11245.9353, -11131.8417, -11050.5074, -11005.9959, -11020.6248,\n",
       "         -11129.1855, -10852.1896, -10797.6784, -10860.2147, -10925.8068,\n",
       "         -10891.7673, -10724.0134, -10800.292 , -10648.2126, -10781.0665,\n",
       "         -10626.6195, -10542.6114, -10546.9785, -10561.4256, -10442.5516,\n",
       "         -10412.8404, -10520.2177, -10355.135 , -10309.6419, -10316.2509,\n",
       "         -10357.9158, -10268.4369, -10368.0275, -10290.0597, -10241.0865]),\n",
       "  array([-17710.6754, -16823.8801, -16269.64  , -15755.2157, -15464.5723,\n",
       "         -14887.5759, -14592.9711, -14482.0379, -13976.3823, -13384.3208,\n",
       "         -13873.845 , -12909.2638, -12669.5731, -12325.2102, -13187.9376,\n",
       "         -12181.8573, -11942.9601, -11740.5697, -11591.779 , -11502.3986,\n",
       "         -11586.6733, -11499.9637, -11455.5667, -11201.4214, -11082.0847,\n",
       "         -11182.9609, -11141.2329, -11068.703 , -10999.9683, -10924.6125,\n",
       "         -10873.984 , -10820.2414, -10747.5588, -10845.7505, -10797.3397,\n",
       "         -10800.1906, -10654.4214, -10603.2066, -10586.8841, -10599.3539,\n",
       "         -10517.1121, -10524.4327, -10526.7123, -10464.1427, -10481.3257,\n",
       "         -10527.6115, -10472.754 , -10405.6311, -10344.7827, -10295.6521]),\n",
       "  array([-17449.4169, -16621.6614, -16089.047 , -15479.8163, -15008.5747,\n",
       "         -14707.3135, -14968.9931, -14153.2818, -13810.8518, -13806.3851,\n",
       "         -13175.0236, -13128.6688, -12449.908 , -12212.8773, -12345.5936,\n",
       "         -12000.6381, -12409.6911, -11948.8994, -11851.0603, -11514.7777,\n",
       "         -11443.9769, -11408.8189, -11259.6466, -11390.5425, -11065.8053,\n",
       "         -10990.0528, -10939.7538, -10993.6258, -10819.0001, -10807.1252,\n",
       "         -10737.4653, -10744.7206, -10632.6673, -10747.8593, -10614.2048,\n",
       "         -10514.8946, -10593.7558, -10502.7384, -10508.6888, -10449.4727,\n",
       "         -10423.3688, -10442.5586, -10369.5572, -10342.0282, -10304.3176,\n",
       "         -10294.5199, -10378.4182, -10305.8942, -10207.4994, -10171.3139]),\n",
       "  array([-17512.3023, -16995.7492, -16136.623 , -15781.0845, -15255.3005,\n",
       "         -14846.1308, -14498.3718, -14432.8594, -13955.0481, -13640.2344,\n",
       "         -13428.7626, -13000.1179, -12659.2085, -12928.3139]),\n",
       "  array([-17274.558 , -16437.7643, -15991.5747, -15465.0903, -15598.3125,\n",
       "         -14976.3524, -14571.8971, -14596.7589, -13997.5435, -13648.3618,\n",
       "         -13521.0245, -13692.768 , -13062.0436, -12721.4415, -12917.5546,\n",
       "         -12853.2135, -12170.9152, -12205.5228, -11716.0066, -11526.2929,\n",
       "         -11421.9854, -11526.1988, -11336.0373, -11221.8848, -11132.5647,\n",
       "         -11061.5574, -11069.7763, -11145.971 , -10970.2571, -10871.4929,\n",
       "         -10845.7531, -10821.3591, -10799.3318, -10746.7455, -10765.775 ,\n",
       "         -10667.4976, -10677.4562, -10641.8227, -10540.6033, -10720.1383,\n",
       "         -10911.319 , -10576.0631, -10485.5809, -10508.0431, -10381.185 ,\n",
       "         -10275.9994, -10283.9025]),\n",
       "  array([-17100.1963, -16443.9986, -15887.017 , -15473.8343, -15156.4397,\n",
       "         -14771.1086, -14535.3884, -14299.4034, -14142.3817, -13647.471 ,\n",
       "         -13293.7543, -13111.4969, -13135.9321, -12628.4832, -12245.9371,\n",
       "         -12067.1294, -11846.3309, -11607.4249, -11440.8701, -11328.5607,\n",
       "         -11286.8466, -11314.4932, -11556.7448, -11479.8424, -11387.6227,\n",
       "         -11151.2489, -11051.2142, -11014.0379, -10794.5305, -10850.5768,\n",
       "         -10742.3774, -10753.1603, -10669.0186, -10628.9879, -10632.3011,\n",
       "         -10544.4763, -10734.4048, -10910.8543, -10525.8882, -10511.7732,\n",
       "         -10421.258 , -10337.2716, -10287.3467, -10273.6883, -10211.2639,\n",
       "         -10218.336 , -10611.3857, -10192.6573, -10685.7505, -10247.4617]),\n",
       "  array([-17355.1029, -16688.0016, -16057.6697, -15666.4804, -15377.3161,\n",
       "         -14794.5806, -14574.9017, -14184.9542, -14083.6151, -13475.6867,\n",
       "         -13248.932 , -13113.287 , -13286.9588, -12536.0315, -12259.611 ,\n",
       "         -12087.7388, -11931.0825, -11800.1416, -11612.3451, -11592.9313,\n",
       "         -11451.4638, -11406.8569, -11459.5544, -11277.4191, -11358.9974,\n",
       "         -11292.7613, -11096.0752, -11121.3111, -11001.4372, -10914.8028,\n",
       "         -10994.9598, -11037.2615, -10819.8254, -10783.9125, -10728.2847,\n",
       "         -10678.7891, -10685.184 , -10591.9207, -10551.0889, -10528.9252,\n",
       "         -10655.5703, -10749.8618, -11395.7548, -10620.9541, -10436.1856,\n",
       "         -10386.5797, -10342.3824, -10311.0985, -10294.0162, -10351.7119]),\n",
       "  array([-17381.2956, -16464.0607, -15822.257 , -15999.1074, -15160.9198,\n",
       "         -14843.0443, -14439.8546, -14504.3682, -14041.7807, -13556.3342,\n",
       "         -13191.9194, -13054.8988, -12738.0521, -12383.438 , -12297.1977,\n",
       "         -12000.2603, -11847.1982, -11730.5601, -11543.4898, -11440.0594,\n",
       "         -11301.0967, -11263.3408, -11172.808 , -11182.3773, -11184.443 ]),\n",
       "  array([-17322.7083, -16550.6201, -15865.7525, -15456.4163, -15194.3977,\n",
       "         -14820.4384, -14811.8411, -14232.6338, -13903.6608, -13580.8145,\n",
       "         -13388.4777, -12909.7459, -13025.8189, -12542.2855, -12121.4004,\n",
       "         -12079.4663, -11762.6428, -11636.2387, -11585.4148, -11351.7217,\n",
       "         -11307.0687, -11245.09  , -11344.7901, -11213.0197, -11045.1964,\n",
       "         -10963.5264, -10897.7186, -10934.4144, -10841.3532, -11115.9956,\n",
       "         -10786.7069, -10708.8732, -10712.0964]),\n",
       "  array([-17401.2799, -16698.4363, -16061.7374, -15642.7536, -15373.9674,\n",
       "         -14868.7563, -14534.1592, -14228.9959, -14020.1936, -13642.9117,\n",
       "         -13344.7089, -13009.8157, -12765.2105, -12886.7328, -12866.8378,\n",
       "         -12350.9105, -12385.4743, -11909.4716, -11698.1434, -11609.2133,\n",
       "         -11568.7656, -11393.5004, -11440.3393, -11239.5505, -11173.6694,\n",
       "         -11168.7074, -11268.664 , -11046.4421, -10990.2468, -10960.155 ,\n",
       "         -10959.614 , -11010.4506, -10830.0486, -10729.115 , -10794.6176,\n",
       "         -10930.0729, -10669.2058, -10706.752 , -10627.6609, -10551.2884,\n",
       "         -10527.915 , -10513.8571]),\n",
       "  array([-17430.2348, -16688.1477, -15967.141 , -15520.9361, -15265.6484,\n",
       "         -14892.3284, -14650.2188, -14728.8699, -14196.1851, -13719.0749,\n",
       "         -13296.7902, -13183.8255, -14178.2956, -12590.7413, -12541.4869,\n",
       "         -12133.1423, -12242.785 , -12081.4934, -11771.2975, -11600.4908,\n",
       "         -11533.7856, -11388.4137, -11333.6314, -11161.7954, -11256.3306,\n",
       "         -11105.232 , -11007.9075, -11046.8052, -10981.096 , -10929.3819,\n",
       "         -10838.0461, -10774.4762, -10934.9268, -10718.0575, -10710.463 ,\n",
       "         -10609.7946, -10586.6385, -10612.6613, -10826.0187, -10519.4114,\n",
       "         -10438.8931, -10434.4186, -10530.7865, -10352.3341, -10331.6992,\n",
       "         -10377.5369, -10193.8894, -10186.1198, -10175.5897, -10475.1465]),\n",
       "  array([-17255.865 , -16226.8749, -15970.9856, -15216.9227, -14969.7537,\n",
       "         -14675.3815, -14478.6784, -14042.7457, -13755.4654, -13384.646 ,\n",
       "         -13259.2715, -12813.0252, -12592.0995, -13376.2252, -12639.1301,\n",
       "         -12627.4526, -12092.1729, -11923.5692, -11773.4522, -11916.3422,\n",
       "         -11643.2403, -11554.8454, -11431.5118, -11331.9037, -11265.0322,\n",
       "         -11215.505 , -11140.5755, -11222.2031, -11054.0184, -11007.6365,\n",
       "         -10852.2297, -10802.5983, -10894.7384, -10741.6771, -10675.18  ,\n",
       "         -10608.3276, -10551.9837, -10489.072 , -10516.2387, -10415.4408,\n",
       "         -10401.4163, -10381.6833, -10346.2175, -10329.5767, -10242.7001,\n",
       "         -10222.8755, -10228.4619, -10156.9379, -10160.6589, -10186.8456])],\n",
       " 6)"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import ModelInf_EM_CMOA as CMOA\n",
    "CMOA_model = CMOA(M=M_seted) \n",
    "#CMOA_model = CMOA(M=M_seted, init_Z='softmv', n_init_Z=0, n_init_G=0, priors=1) \n",
    "\n",
    "args = {'epochs':1, 'batch_size':BATCH_SIZE, 'optimizer':OPT}\n",
    "CMOA_model.set_model(aux_model, ann_model=group_model, **args)\n",
    "#CMOA_model.set_model(group_model) #you also can set it manually\n",
    "\n",
    "# if you want you can initialize step E forcing some method\n",
    "#CMOA_model.init_E(y_cat_var, A_idx_var, method=\"hardmv\")\n",
    "#CMOA_model.get_qestimation()\n",
    "\n",
    "CMOA_model.fit(X_train, y_cat_var, A_idx_var, runs = 20,max_iter=EPOCHS_BASE,tolerance=TOL) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.8799759951990398\n",
      "Test accuracy of model = 0.7516879219804952\n"
     ]
    }
   ],
   "source": [
    "cmoaK_fx = CMOA_model.get_basemodel()\n",
    "\n",
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(cmoaK_fx, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(cmoaK_fx, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CMM\n",
    "Groups over global scenario"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 300)           4200300   \n",
      "_________________________________________________________________\n",
      "cu_dnngru_1 (CuDNNGRU)       (None, 50, 128)           165120    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 50, 128)           0         \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (None, 64)                37248     \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 4,402,798\n",
      "Trainable params: 202,498\n",
      "Non-trainable params: 4,200,300\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "M_seted = 3\n",
    "\n",
    "aux_model = clone_UB.get_model()\n",
    "aux_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.56 sec)\tlogL: -17139.525\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -16413.043\tTol1: 0.04239\tTol2: 0.36478\tTol3: 0.03615\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15915.286\tTol1: 0.03033\tTol2: 0.05530\tTol3: 0.03928\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15518.041\tTol1: 0.02496\tTol2: 0.04897\tTol3: 0.03836\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15305.557\tTol1: 0.01369\tTol2: 0.04307\tTol3: 0.03388\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14829.171\tTol1: 0.03113\tTol2: 0.06378\tTol3: 0.05460\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14490.487\tTol1: 0.02284\tTol2: 0.03795\tTol3: 0.02270\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14189.173\tTol1: 0.02079\tTol2: 0.02935\tTol3: 0.02299\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -17348.206\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16549.013\tTol1: 0.04607\tTol2: 0.37892\tTol3: 0.06936\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15904.990\tTol1: 0.03892\tTol2: 0.06364\tTol3: 0.03891\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15473.185\tTol1: 0.02715\tTol2: 0.05234\tTol3: 0.03914\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15124.542\tTol1: 0.02253\tTol2: 0.07478\tTol3: 0.06027\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14754.465\tTol1: 0.02447\tTol2: 0.04595\tTol3: 0.04257\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14514.086\tTol1: 0.01629\tTol2: 0.04293\tTol3: 0.03337\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14371.989\tTol1: 0.00979\tTol2: 0.04472\tTol3: 0.04032\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13962.571\tTol1: 0.02849\tTol2: 0.05178\tTol3: 0.04629\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13729.514\tTol1: 0.01669\tTol2: 0.03453\tTol3: 0.02494\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13274.183\tTol1: 0.03316\tTol2: 0.09766\tTol3: 0.08307\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13101.203\tTol1: 0.01303\tTol2: 0.04475\tTol3: 0.03100\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12833.091\tTol1: 0.02046\tTol2: 0.05097\tTol3: 0.03537\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12504.194\tTol1: 0.02563\tTol2: 0.03027\tTol3: 0.02583\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12497.233\tTol1: 0.00056\tTol2: 0.03042\tTol3: 0.02542\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12149.473\tTol1: 0.02783\tTol2: 0.10227\tTol3: 0.08369\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11960.126\tTol1: 0.01558\tTol2: 0.03719\tTol3: 0.03105\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12300.344\tTol1: 0.02845\tTol2: 0.02135\tTol3: 0.01341\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.48 sec)\tlogL: -17164.920\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -16708.766\tTol1: 0.02657\tTol2: 0.36644\tTol3: 0.03702\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15989.816\tTol1: 0.04303\tTol2: 0.09803\tTol3: 0.07783\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15544.093\tTol1: 0.02788\tTol2: 0.06873\tTol3: 0.04285\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15130.689\tTol1: 0.02660\tTol2: 0.04881\tTol3: 0.04164\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14776.545\tTol1: 0.02341\tTol2: 0.04775\tTol3: 0.03855\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14454.859\tTol1: 0.02177\tTol2: 0.03589\tTol3: 0.03925\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14085.440\tTol1: 0.02556\tTol2: 0.03673\tTol3: 0.03549\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13905.694\tTol1: 0.01276\tTol2: 0.04315\tTol3: 0.03426\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13617.396\tTol1: 0.02073\tTol2: 0.06542\tTol3: 0.05516\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13403.693\tTol1: 0.01569\tTol2: 0.05946\tTol3: 0.05553\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13054.986\tTol1: 0.02602\tTol2: 0.11200\tTol3: 0.08405\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12640.920\tTol1: 0.03172\tTol2: 0.07904\tTol3: 0.05462\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12729.686\tTol1: 0.00702\tTol2: 0.02636\tTol3: 0.02284\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.42 sec)\tlogL: -17206.098\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16493.802\tTol1: 0.04140\tTol2: 0.36544\tTol3: 0.03616\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15841.483\tTol1: 0.03955\tTol2: 0.05579\tTol3: 0.03656\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15487.483\tTol1: 0.02235\tTol2: 0.05423\tTol3: 0.04019\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15030.914\tTol1: 0.02948\tTol2: 0.04966\tTol3: 0.03553\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14680.701\tTol1: 0.02330\tTol2: 0.06866\tTol3: 0.05717\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14622.677\tTol1: 0.00395\tTol2: 0.04048\tTol3: 0.02655\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14160.394\tTol1: 0.03161\tTol2: 0.07786\tTol3: 0.06230\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13828.777\tTol1: 0.02342\tTol2: 0.06772\tTol3: 0.05139\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13549.989\tTol1: 0.02016\tTol2: 0.04227\tTol3: 0.02974\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13290.149\tTol1: 0.01918\tTol2: 0.03310\tTol3: 0.02487\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13369.771\tTol1: 0.00599\tTol2: 0.09157\tTol3: 0.07511\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12673.953\tTol1: 0.05204\tTol2: 0.14421\tTol3: 0.09984\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12485.652\tTol1: 0.01486\tTol2: 0.04799\tTol3: 0.02874\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12264.771\tTol1: 0.01769\tTol2: 0.05946\tTol3: 0.05064\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12155.332\tTol1: 0.00892\tTol2: 0.02882\tTol3: 0.02130\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.46 sec)\tlogL: -17533.660\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16701.951\tTol1: 0.04744\tTol2: 0.36106\tTol3: 0.02887\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16087.161\tTol1: 0.03681\tTol2: 0.05157\tTol3: 0.04042\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15607.967\tTol1: 0.02979\tTol2: 0.06465\tTol3: 0.05425\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15193.184\tTol1: 0.02658\tTol2: 0.05729\tTol3: 0.04329\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14982.391\tTol1: 0.01387\tTol2: 0.05137\tTol3: 0.04660\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14526.643\tTol1: 0.03042\tTol2: 0.09949\tTol3: 0.07511\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14346.071\tTol1: 0.01243\tTol2: 0.03568\tTol3: 0.02102\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14235.435\tTol1: 0.00771\tTol2: 0.07003\tTol3: 0.05761\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13691.447\tTol1: 0.03821\tTol2: 0.09447\tTol3: 0.08508\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13459.359\tTol1: 0.01695\tTol2: 0.03938\tTol3: 0.02063\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13213.984\tTol1: 0.01823\tTol2: 0.06600\tTol3: 0.05382\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13161.722\tTol1: 0.00396\tTol2: 0.07156\tTol3: 0.05211\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12474.830\tTol1: 0.05219\tTol2: 0.13365\tTol3: 0.09561\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12286.301\tTol1: 0.01511\tTol2: 0.06115\tTol3: 0.03764\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12093.932\tTol1: 0.01566\tTol2: 0.04836\tTol3: 0.03675\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12118.125\tTol1: 0.00200\tTol2: 0.04436\tTol3: 0.03950\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11895.768\tTol1: 0.01835\tTol2: 0.08178\tTol3: 0.06568\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11645.556\tTol1: 0.02103\tTol2: 0.05991\tTol3: 0.04856\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11489.061\tTol1: 0.01344\tTol2: 0.03571\tTol3: 0.02770\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11389.826\tTol1: 0.00864\tTol2: 0.04623\tTol3: 0.03343\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11309.460\tTol1: 0.00706\tTol2: 0.02479\tTol3: 0.01871\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.35 sec)\tlogL: -17527.615\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16756.334\tTol1: 0.04400\tTol2: 0.36381\tTol3: 0.04000\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16299.254\tTol1: 0.02728\tTol2: 0.05656\tTol3: 0.05069\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15710.852\tTol1: 0.03610\tTol2: 0.05783\tTol3: 0.03746\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15408.836\tTol1: 0.01922\tTol2: 0.06598\tTol3: 0.04361\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14817.825\tTol1: 0.03836\tTol2: 0.10178\tTol3: 0.09866\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14980.928\tTol1: 0.01101\tTol2: 0.06389\tTol3: 0.03212\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14428.285\tTol1: 0.03689\tTol2: 0.12216\tTol3: 0.09580\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13987.198\tTol1: 0.03057\tTol2: 0.05530\tTol3: 0.04110\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13677.219\tTol1: 0.02216\tTol2: 0.06744\tTol3: 0.04791\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13896.295\tTol1: 0.01602\tTol2: 0.05439\tTol3: 0.03652\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13052.147\tTol1: 0.06075\tTol2: 0.13465\tTol3: 0.09964\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12868.036\tTol1: 0.01411\tTol2: 0.08156\tTol3: 0.05021\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12529.796\tTol1: 0.02629\tTol2: 0.05757\tTol3: 0.03896\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12335.036\tTol1: 0.01554\tTol2: 0.05260\tTol3: 0.03868\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12023.516\tTol1: 0.02525\tTol2: 0.05637\tTol3: 0.05776\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -12013.839\tTol1: 0.00080\tTol2: 0.03486\tTol3: 0.01821\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11768.064\tTol1: 0.02046\tTol2: 0.04201\tTol3: 0.03691\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11618.097\tTol1: 0.01274\tTol2: 0.03373\tTol3: 0.02403\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11552.531\tTol1: 0.00564\tTol2: 0.03705\tTol3: 0.02595\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11915.048\tTol1: 0.03138\tTol2: 0.01752\tTol3: 0.01162\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11547.153\tTol1: 0.03088\tTol2: 0.15210\tTol3: 0.11966\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11500.084\tTol1: 0.00408\tTol2: 0.08598\tTol3: 0.05576\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11557.405\tTol1: 0.00498\tTol2: 0.06620\tTol3: 0.04590\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11365.103\tTol1: 0.01664\tTol2: 0.11427\tTol3: 0.09080\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11217.506\tTol1: 0.01299\tTol2: 0.03381\tTol3: 0.02087\t\n",
      "Iter 27/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11178.307\tTol1: 0.00349\tTol2: 0.03279\tTol3: 0.02247\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11133.226\tTol1: 0.00403\tTol2: 0.03609\tTol3: 0.03098\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11052.543\tTol1: 0.00725\tTol2: 0.03277\tTol3: 0.03212\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11048.796\tTol1: 0.00034\tTol2: 0.01929\tTol3: 0.01360\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.49 sec)\tlogL: -17335.299\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -16498.671\tTol1: 0.04826\tTol2: 0.36594\tTol3: 0.04222\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16000.891\tTol1: 0.03017\tTol2: 0.06030\tTol3: 0.04170\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15832.851\tTol1: 0.01050\tTol2: 0.08275\tTol3: 0.07825\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15272.742\tTol1: 0.03538\tTol2: 0.10397\tTol3: 0.09862\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -15011.209\tTol1: 0.01712\tTol2: 0.07701\tTol3: 0.04075\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14616.759\tTol1: 0.02628\tTol2: 0.08111\tTol3: 0.06388\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14271.603\tTol1: 0.02361\tTol2: 0.09967\tTol3: 0.07405\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14007.969\tTol1: 0.01847\tTol2: 0.10305\tTol3: 0.07003\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13877.250\tTol1: 0.00933\tTol2: 0.08225\tTol3: 0.07040\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13207.297\tTol1: 0.04828\tTol2: 0.08618\tTol3: 0.07201\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.60 sec)\tlogL: -13545.452\tTol1: 0.02560\tTol2: 0.05276\tTol3: 0.03248\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12636.742\tTol1: 0.06709\tTol2: 0.13006\tTol3: 0.09536\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13078.303\tTol1: 0.03494\tTol2: 0.06133\tTol3: 0.03223\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.60 sec)\tlogL: -12274.901\tTol1: 0.06143\tTol2: 0.15874\tTol3: 0.11321\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12142.155\tTol1: 0.01081\tTol2: 0.05868\tTol3: 0.03323\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.60 sec)\tlogL: -12191.082\tTol1: 0.00403\tTol2: 0.03720\tTol3: 0.02689\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11840.999\tTol1: 0.02872\tTol2: 0.09445\tTol3: 0.06981\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11806.005\tTol1: 0.00296\tTol2: 0.04213\tTol3: 0.02589\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11837.321\tTol1: 0.00265\tTol2: 0.06216\tTol3: 0.04409\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11613.119\tTol1: 0.01894\tTol2: 0.08089\tTol3: 0.06068\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11623.220\tTol1: 0.00087\tTol2: 0.04800\tTol3: 0.03162\t\n",
      "Iter 23/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11663.834\tTol1: 0.00349\tTol2: 0.03639\tTol3: 0.02934\t\n",
      "Iter 24/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11490.931\tTol1: 0.01482\tTol2: 0.08653\tTol3: 0.06395\t\n",
      "Iter 25/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11412.442\tTol1: 0.00683\tTol2: 0.03514\tTol3: 0.01723\t\n",
      "Iter 26/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11313.596\tTol1: 0.00866\tTol2: 0.03243\tTol3: 0.02459\t\n",
      "Iter 27/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11243.501\tTol1: 0.00620\tTol2: 0.03718\tTol3: 0.02730\t\n",
      "Iter 28/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11146.869\tTol1: 0.00859\tTol2: 0.03091\tTol3: 0.02531\t\n",
      "Iter 29/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -11077.563\tTol1: 0.00622\tTol2: 0.06106\tTol3: 0.03843\t\n",
      "Iter 30/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -11109.573\tTol1: 0.00289\tTol2: 0.04553\tTol3: 0.03432\t\n",
      "Iter 31/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11055.644\tTol1: 0.00485\tTol2: 0.08814\tTol3: 0.05980\t\n",
      "Iter 32/50\n",
      "M step: done,  E step: done //  (in 0.53 sec)\tlogL: -11022.189\tTol1: 0.00303\tTol2: 0.06582\tTol3: 0.04320\t\n",
      "Iter 33/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -10983.236\tTol1: 0.00353\tTol2: 0.09198\tTol3: 0.06197\t\n",
      "Iter 34/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -10883.052\tTol1: 0.00912\tTol2: 0.04522\tTol3: 0.03320\t\n",
      "Iter 35/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -10960.165\tTol1: 0.00709\tTol2: 0.05379\tTol3: 0.04279\t\n",
      "Iter 36/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -10847.410\tTol1: 0.01029\tTol2: 0.05881\tTol3: 0.04631\t\n",
      "Iter 37/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -10770.855\tTol1: 0.00706\tTol2: 0.01853\tTol3: 0.01241\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.44 sec)\tlogL: -17104.613\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16584.966\tTol1: 0.03038\tTol2: 0.37086\tTol3: 0.05581\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15816.730\tTol1: 0.04632\tTol2: 0.05744\tTol3: 0.06087\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -15436.974\tTol1: 0.02401\tTol2: 0.06189\tTol3: 0.04044\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -15134.890\tTol1: 0.01957\tTol2: 0.04647\tTol3: 0.04092\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14869.697\tTol1: 0.01752\tTol2: 0.04625\tTol3: 0.04078\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14547.580\tTol1: 0.02166\tTol2: 0.05247\tTol3: 0.04324\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14249.977\tTol1: 0.02046\tTol2: 0.05282\tTol3: 0.03648\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13836.859\tTol1: 0.02899\tTol2: 0.03591\tTol3: 0.03442\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13513.074\tTol1: 0.02340\tTol2: 0.04542\tTol3: 0.03082\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13297.021\tTol1: 0.01599\tTol2: 0.02931\tTol3: 0.01851\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.47 sec)\tlogL: -17250.975\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16523.871\tTol1: 0.04215\tTol2: 0.36406\tTol3: 0.03130\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -16115.565\tTol1: 0.02471\tTol2: 0.05350\tTol3: 0.04129\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15472.294\tTol1: 0.03992\tTol2: 0.07772\tTol3: 0.06200\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15205.466\tTol1: 0.01725\tTol2: 0.05751\tTol3: 0.03651\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14725.969\tTol1: 0.03153\tTol2: 0.09419\tTol3: 0.08654\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15504.100\tTol1: 0.05284\tTol2: 0.05363\tTol3: 0.03135\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14373.278\tTol1: 0.07294\tTol2: 0.20006\tTol3: 0.16702\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14114.013\tTol1: 0.01804\tTol2: 0.05837\tTol3: 0.03578\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13795.465\tTol1: 0.02257\tTol2: 0.08403\tTol3: 0.07627\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13561.756\tTol1: 0.01694\tTol2: 0.04566\tTol3: 0.03482\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13265.132\tTol1: 0.02187\tTol2: 0.04025\tTol3: 0.03193\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -13156.328\tTol1: 0.00820\tTol2: 0.05908\tTol3: 0.04770\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12909.749\tTol1: 0.01874\tTol2: 0.09319\tTol3: 0.08305\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12552.217\tTol1: 0.02769\tTol2: 0.06583\tTol3: 0.06096\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12439.716\tTol1: 0.00896\tTol2: 0.07514\tTol3: 0.05553\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12885.033\tTol1: 0.03580\tTol2: 0.07955\tTol3: 0.06552\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.59 sec)\tlogL: -12191.811\tTol1: 0.05380\tTol2: 0.13808\tTol3: 0.11266\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12211.173\tTol1: 0.00159\tTol2: 0.09297\tTol3: 0.06839\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -11950.174\tTol1: 0.02137\tTol2: 0.09128\tTol3: 0.07694\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11823.975\tTol1: 0.01056\tTol2: 0.03016\tTol3: 0.02766\t\n",
      "Iter 22/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11752.370\tTol1: 0.00606\tTol2: 0.02412\tTol3: 0.02052\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.35 sec)\tlogL: -17631.540\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -16629.500\tTol1: 0.05683\tTol2: 0.43298\tTol3: 0.12444\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16007.056\tTol1: 0.03743\tTol2: 0.07080\tTol3: 0.04920\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15611.300\tTol1: 0.02472\tTol2: 0.06476\tTol3: 0.04077\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15463.133\tTol1: 0.00949\tTol2: 0.05297\tTol3: 0.04317\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14921.394\tTol1: 0.03503\tTol2: 0.13198\tTol3: 0.10972\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14631.919\tTol1: 0.01940\tTol2: 0.02813\tTol3: 0.01473\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.36 sec)\tlogL: -17477.980\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -16538.213\tTol1: 0.05377\tTol2: 0.39373\tTol3: 0.07819\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16186.417\tTol1: 0.02127\tTol2: 0.07716\tTol3: 0.04363\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15524.106\tTol1: 0.04092\tTol2: 0.07706\tTol3: 0.05638\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15152.344\tTol1: 0.02395\tTol2: 0.06849\tTol3: 0.04161\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14774.275\tTol1: 0.02495\tTol2: 0.04960\tTol3: 0.03514\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14479.025\tTol1: 0.01998\tTol2: 0.04686\tTol3: 0.03701\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14230.663\tTol1: 0.01715\tTol2: 0.04767\tTol3: 0.03854\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13850.915\tTol1: 0.02669\tTol2: 0.05904\tTol3: 0.04089\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13539.801\tTol1: 0.02246\tTol2: 0.05996\tTol3: 0.04489\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13119.963\tTol1: 0.03101\tTol2: 0.05040\tTol3: 0.03482\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13253.063\tTol1: 0.01014\tTol2: 0.04914\tTol3: 0.03849\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12924.039\tTol1: 0.02483\tTol2: 0.12479\tTol3: 0.10991\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12469.976\tTol1: 0.03513\tTol2: 0.09282\tTol3: 0.08244\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12207.474\tTol1: 0.02105\tTol2: 0.07788\tTol3: 0.04691\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11952.078\tTol1: 0.02092\tTol2: 0.02712\tTol3: 0.02037\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.40 sec)\tlogL: -17307.036\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16591.729\tTol1: 0.04133\tTol2: 0.37330\tTol3: 0.06141\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15969.195\tTol1: 0.03752\tTol2: 0.05814\tTol3: 0.06256\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15698.409\tTol1: 0.01696\tTol2: 0.06003\tTol3: 0.04110\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15225.142\tTol1: 0.03015\tTol2: 0.09153\tTol3: 0.07823\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15128.935\tTol1: 0.00632\tTol2: 0.03804\tTol3: 0.02481\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14712.688\tTol1: 0.02751\tTol2: 0.06184\tTol3: 0.05930\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14368.693\tTol1: 0.02338\tTol2: 0.05073\tTol3: 0.03177\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14582.215\tTol1: 0.01486\tTol2: 0.03324\tTol3: 0.02588\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13842.915\tTol1: 0.05070\tTol2: 0.12712\tTol3: 0.10372\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13563.947\tTol1: 0.02015\tTol2: 0.05946\tTol3: 0.03821\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13221.261\tTol1: 0.02526\tTol2: 0.07239\tTol3: 0.06350\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12908.660\tTol1: 0.02364\tTol2: 0.03555\tTol3: 0.03565\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12758.410\tTol1: 0.01164\tTol2: 0.04466\tTol3: 0.02963\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12625.889\tTol1: 0.01039\tTol2: 0.03131\tTol3: 0.02722\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12330.271\tTol1: 0.02341\tTol2: 0.02843\tTol3: 0.02544\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.42 sec)\tlogL: -17306.594\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16561.412\tTol1: 0.04306\tTol2: 0.36346\tTol3: 0.03935\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15848.489\tTol1: 0.04305\tTol2: 0.05302\tTol3: 0.05328\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15467.131\tTol1: 0.02406\tTol2: 0.06381\tTol3: 0.04400\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15113.926\tTol1: 0.02284\tTol2: 0.04820\tTol3: 0.03413\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14747.416\tTol1: 0.02425\tTol2: 0.06527\tTol3: 0.04961\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14510.393\tTol1: 0.01607\tTol2: 0.04765\tTol3: 0.03168\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14220.530\tTol1: 0.01998\tTol2: 0.05378\tTol3: 0.04777\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13901.985\tTol1: 0.02240\tTol2: 0.04312\tTol3: 0.02672\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13551.963\tTol1: 0.02518\tTol2: 0.03174\tTol3: 0.02337\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13475.637\tTol1: 0.00563\tTol2: 0.02631\tTol3: 0.02113\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.45 sec)\tlogL: -17339.293\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -16617.951\tTol1: 0.04160\tTol2: 0.38827\tTol3: 0.07065\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15979.942\tTol1: 0.03839\tTol2: 0.05944\tTol3: 0.03660\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15573.291\tTol1: 0.02545\tTol2: 0.06775\tTol3: 0.04198\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15157.834\tTol1: 0.02668\tTol2: 0.07199\tTol3: 0.07475\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15012.552\tTol1: 0.00958\tTol2: 0.06766\tTol3: 0.06908\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14742.412\tTol1: 0.01799\tTol2: 0.06955\tTol3: 0.06870\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14243.838\tTol1: 0.03382\tTol2: 0.11810\tTol3: 0.08982\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13915.066\tTol1: 0.02308\tTol2: 0.04719\tTol3: 0.04279\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13727.621\tTol1: 0.01347\tTol2: 0.04163\tTol3: 0.03292\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13450.458\tTol1: 0.02019\tTol2: 0.05327\tTol3: 0.04795\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13128.758\tTol1: 0.02392\tTol2: 0.03048\tTol3: 0.02527\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12957.710\tTol1: 0.01303\tTol2: 0.02231\tTol3: 0.01367\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.43 sec)\tlogL: -17435.894\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -16721.697\tTol1: 0.04096\tTol2: 0.36177\tTol3: 0.03012\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16077.843\tTol1: 0.03850\tTol2: 0.06293\tTol3: 0.05091\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15549.106\tTol1: 0.03289\tTol2: 0.06249\tTol3: 0.04285\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15241.797\tTol1: 0.01976\tTol2: 0.05105\tTol3: 0.04104\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15047.474\tTol1: 0.01275\tTol2: 0.06312\tTol3: 0.05770\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14637.116\tTol1: 0.02727\tTol2: 0.07646\tTol3: 0.07163\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14253.725\tTol1: 0.02619\tTol2: 0.04802\tTol3: 0.04182\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13928.965\tTol1: 0.02278\tTol2: 0.04096\tTol3: 0.03006\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -14113.095\tTol1: 0.01322\tTol2: 0.02586\tTol3: 0.02070\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.36 sec)\tlogL: -17481.646\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16805.548\tTol1: 0.03867\tTol2: 0.35962\tTol3: 0.03083\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16163.593\tTol1: 0.03820\tTol2: 0.06501\tTol3: 0.06248\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15667.059\tTol1: 0.03072\tTol2: 0.07172\tTol3: 0.04174\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15480.444\tTol1: 0.01191\tTol2: 0.05004\tTol3: 0.04966\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14896.807\tTol1: 0.03770\tTol2: 0.08482\tTol3: 0.07702\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15266.308\tTol1: 0.02480\tTol2: 0.05786\tTol3: 0.03814\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14357.405\tTol1: 0.05954\tTol2: 0.13979\tTol3: 0.10699\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13874.466\tTol1: 0.03364\tTol2: 0.06984\tTol3: 0.04258\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13498.134\tTol1: 0.02712\tTol2: 0.04785\tTol3: 0.03410\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13964.721\tTol1: 0.03457\tTol2: 0.05485\tTol3: 0.04286\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13493.719\tTol1: 0.03373\tTol2: 0.15650\tTol3: 0.11186\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12919.507\tTol1: 0.04255\tTol2: 0.07837\tTol3: 0.05252\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12546.096\tTol1: 0.02890\tTol2: 0.05685\tTol3: 0.03587\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -12421.164\tTol1: 0.00996\tTol2: 0.03810\tTol3: 0.03532\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12255.474\tTol1: 0.01334\tTol2: 0.08155\tTol3: 0.07470\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12102.141\tTol1: 0.01251\tTol2: 0.03612\tTol3: 0.03103\t\n",
      "Iter 18/50\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11825.482\tTol1: 0.02286\tTol2: 0.03109\tTol3: 0.02473\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -11775.665\tTol1: 0.00421\tTol2: 0.02587\tTol3: 0.02085\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.37 sec)\tlogL: -17528.521\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16827.209\tTol1: 0.04001\tTol2: 0.37511\tTol3: 0.05611\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16054.686\tTol1: 0.04591\tTol2: 0.07342\tTol3: 0.04770\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15483.371\tTol1: 0.03559\tTol2: 0.06369\tTol3: 0.04570\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15051.080\tTol1: 0.02792\tTol2: 0.05483\tTol3: 0.04535\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14786.085\tTol1: 0.01761\tTol2: 0.04861\tTol3: 0.04281\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14686.870\tTol1: 0.00671\tTol2: 0.05243\tTol3: 0.03737\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13985.458\tTol1: 0.04776\tTol2: 0.08621\tTol3: 0.07268\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14251.986\tTol1: 0.01906\tTol2: 0.06177\tTol3: 0.03904\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13432.254\tTol1: 0.05752\tTol2: 0.17402\tTol3: 0.14293\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13799.947\tTol1: 0.02737\tTol2: 0.05013\tTol3: 0.03781\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12950.233\tTol1: 0.06157\tTol2: 0.15031\tTol3: 0.12759\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12648.460\tTol1: 0.02330\tTol2: 0.08518\tTol3: 0.05419\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12378.927\tTol1: 0.02131\tTol2: 0.03498\tTol3: 0.02555\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12109.665\tTol1: 0.02175\tTol2: 0.01608\tTol3: 0.01172\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.47 sec)\tlogL: -17095.622\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16489.383\tTol1: 0.03546\tTol2: 0.36389\tTol3: 0.03512\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15900.399\tTol1: 0.03572\tTol2: 0.04899\tTol3: 0.04138\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15634.256\tTol1: 0.01674\tTol2: 0.05328\tTol3: 0.04488\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15380.148\tTol1: 0.01625\tTol2: 0.06276\tTol3: 0.05112\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -14827.129\tTol1: 0.03596\tTol2: 0.07003\tTol3: 0.06511\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14620.374\tTol1: 0.01394\tTol2: 0.07038\tTol3: 0.04618\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14143.677\tTol1: 0.03260\tTol2: 0.05030\tTol3: 0.03128\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13850.427\tTol1: 0.02073\tTol2: 0.05382\tTol3: 0.03912\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13526.434\tTol1: 0.02339\tTol2: 0.04815\tTol3: 0.04701\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13624.149\tTol1: 0.00722\tTol2: 0.05903\tTol3: 0.03731\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12972.031\tTol1: 0.04786\tTol2: 0.12749\tTol3: 0.09773\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12879.895\tTol1: 0.00710\tTol2: 0.05458\tTol3: 0.03059\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12381.587\tTol1: 0.03869\tTol2: 0.09438\tTol3: 0.07750\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12143.854\tTol1: 0.01920\tTol2: 0.02342\tTol3: 0.01071\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.34 sec)\tlogL: -17203.944\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -16448.672\tTol1: 0.04390\tTol2: 0.37162\tTol3: 0.05941\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16025.056\tTol1: 0.02575\tTol2: 0.06604\tTol3: 0.03700\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15650.310\tTol1: 0.02339\tTol2: 0.05736\tTol3: 0.03911\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -15226.182\tTol1: 0.02710\tTol2: 0.04258\tTol3: 0.03749\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -15055.831\tTol1: 0.01119\tTol2: 0.03953\tTol3: 0.03481\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14700.703\tTol1: 0.02359\tTol2: 0.05927\tTol3: 0.04846\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14507.781\tTol1: 0.01312\tTol2: 0.04116\tTol3: 0.02840\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -14149.247\tTol1: 0.02471\tTol2: 0.07218\tTol3: 0.05999\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14057.155\tTol1: 0.00651\tTol2: 0.03777\tTol3: 0.03520\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13531.856\tTol1: 0.03737\tTol2: 0.11878\tTol3: 0.09539\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -13415.954\tTol1: 0.00857\tTol2: 0.05017\tTol3: 0.03212\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12927.715\tTol1: 0.03639\tTol2: 0.10569\tTol3: 0.08532\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -12892.757\tTol1: 0.00270\tTol2: 0.05894\tTol3: 0.03888\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12571.364\tTol1: 0.02493\tTol2: 0.13186\tTol3: 0.10149\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -12249.293\tTol1: 0.02562\tTol2: 0.04948\tTol3: 0.03970\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.54 sec)\tlogL: -12224.751\tTol1: 0.00200\tTol2: 0.04113\tTol3: 0.02555\t\n",
      "Iter 18/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12063.524\tTol1: 0.01319\tTol2: 0.07313\tTol3: 0.06162\t\n",
      "Iter 19/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11911.198\tTol1: 0.01263\tTol2: 0.09163\tTol3: 0.06757\t\n",
      "Iter 20/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -11692.031\tTol1: 0.01840\tTol2: 0.06410\tTol3: 0.05301\t\n",
      "Iter 21/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -11765.347\tTol1: 0.00627\tTol2: 0.01584\tTol3: 0.01284\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Doing clustering...Done!\n",
      "Alphas:  (3,)\n",
      "Betas:  (3, 2, 2)\n",
      "Q estimate:  (4999, 2, 3, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 1.45 sec)\tlogL: -17307.670\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -16522.220\tTol1: 0.04538\tTol2: 0.37650\tTol3: 0.05798\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.56 sec)\tlogL: -16061.128\tTol1: 0.02791\tTol2: 0.05411\tTol3: 0.03436\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -15608.364\tTol1: 0.02819\tTol2: 0.05982\tTol3: 0.03949\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15343.424\tTol1: 0.01697\tTol2: 0.04205\tTol3: 0.03738\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -15240.169\tTol1: 0.00673\tTol2: 0.05052\tTol3: 0.03299\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14882.998\tTol1: 0.02344\tTol2: 0.06433\tTol3: 0.05644\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14443.848\tTol1: 0.02951\tTol2: 0.04390\tTol3: 0.02986\t\n",
      "Iter 9/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -14184.954\tTol1: 0.01792\tTol2: 0.04994\tTol3: 0.03349\t\n",
      "Iter 10/50\n",
      "M step: done,  E step: done //  (in 0.55 sec)\tlogL: -13871.614\tTol1: 0.02209\tTol2: 0.04756\tTol3: 0.02777\t\n",
      "Iter 11/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13935.618\tTol1: 0.00461\tTol2: 0.03921\tTol3: 0.02930\t\n",
      "Iter 12/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13287.285\tTol1: 0.04652\tTol2: 0.15542\tTol3: 0.12767\t\n",
      "Iter 13/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -13513.840\tTol1: 0.01705\tTol2: 0.05101\tTol3: 0.02239\t\n",
      "Iter 14/50\n",
      "M step: done,  E step: done //  (in 0.58 sec)\tlogL: -12755.697\tTol1: 0.05610\tTol2: 0.13877\tTol3: 0.10738\t\n",
      "Iter 15/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12673.172\tTol1: 0.00647\tTol2: 0.04840\tTol3: 0.03488\t\n",
      "Iter 16/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12370.150\tTol1: 0.02391\tTol2: 0.09830\tTol3: 0.08379\t\n",
      "Iter 17/50\n",
      "M step: done,  E step: done //  (in 0.57 sec)\tlogL: -12178.355\tTol1: 0.01550\tTol2: 0.01706\tTol3: 0.00610\t\n",
      "Finished training!\n",
      "20 runs over CMM, Epochs to converge=  16.9\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "([array([-17139.5248, -16413.0426, -15915.286 , -15518.0415, -15305.5574,\n",
       "         -14829.1709, -14490.4869, -14189.1728]),\n",
       "  array([-17348.2058, -16549.013 , -15904.9904, -15473.185 , -15124.5424,\n",
       "         -14754.4647, -14514.0862, -14371.9895, -13962.571 , -13729.5137,\n",
       "         -13274.1826, -13101.2028, -12833.0911, -12504.1941, -12497.2335,\n",
       "         -12149.473 , -11960.1263, -12300.3441]),\n",
       "  array([-17164.9199, -16708.7656, -15989.816 , -15544.0926, -15130.6887,\n",
       "         -14776.5449, -14454.8589, -14085.4399, -13905.6935, -13617.3964,\n",
       "         -13403.693 , -13054.986 , -12640.9199, -12729.6855]),\n",
       "  array([-17206.0985, -16493.8015, -15841.4831, -15487.4828, -15030.9144,\n",
       "         -14680.7012, -14622.6769, -14160.394 , -13828.7773, -13549.9885,\n",
       "         -13290.1492, -13369.7713, -12673.9535, -12485.6521, -12264.7707,\n",
       "         -12155.3323]),\n",
       "  array([-17533.6603, -16701.9508, -16087.1613, -15607.9674, -15193.1844,\n",
       "         -14982.3909, -14526.6434, -14346.0714, -14235.4353, -13691.4471,\n",
       "         -13459.3592, -13213.9838, -13161.7222, -12474.8303, -12286.3013,\n",
       "         -12093.9319, -12118.1254, -11895.7684, -11645.5556, -11489.0606,\n",
       "         -11389.8257, -11309.4596]),\n",
       "  array([-17527.615 , -16756.3344, -16299.2536, -15710.8524, -15408.8358,\n",
       "         -14817.8254, -14980.9279, -14428.2848, -13987.1985, -13677.2186,\n",
       "         -13896.2954, -13052.1474, -12868.0361, -12529.7956, -12335.0356,\n",
       "         -12023.5163, -12013.8389, -11768.0645, -11618.0973, -11552.5315,\n",
       "         -11915.0483, -11547.1526, -11500.0838, -11557.4054, -11365.1029,\n",
       "         -11217.5061, -11178.3073, -11133.2259, -11052.5425, -11048.7957]),\n",
       "  array([-17335.2989, -16498.6709, -16000.8907, -15832.8508, -15272.7419,\n",
       "         -15011.2086, -14616.7595, -14271.6031, -14007.9688, -13877.2501,\n",
       "         -13207.2973, -13545.4521, -12636.7422, -13078.3029, -12274.9006,\n",
       "         -12142.155 , -12191.0816, -11840.9992, -11806.0045, -11837.3207,\n",
       "         -11613.1186, -11623.2202, -11663.8342, -11490.9312, -11412.4424,\n",
       "         -11313.5957, -11243.501 , -11146.8692, -11077.5634, -11109.5726,\n",
       "         -11055.644 , -11022.1886, -10983.2365, -10883.0523, -10960.1655,\n",
       "         -10847.4095, -10770.855 ]),\n",
       "  array([-17104.6129, -16584.9661, -15816.7299, -15436.974 , -15134.8904,\n",
       "         -14869.6974, -14547.5797, -14249.9769, -13836.8592, -13513.0745,\n",
       "         -13297.0207]),\n",
       "  array([-17250.975 , -16523.8709, -16115.565 , -15472.2939, -15205.4656,\n",
       "         -14725.9693, -15504.0998, -14373.2784, -14114.013 , -13795.4648,\n",
       "         -13561.7559, -13265.1321, -13156.3282, -12909.7491, -12552.2173,\n",
       "         -12439.7163, -12885.0329, -12191.8113, -12211.1729, -11950.174 ,\n",
       "         -11823.9748, -11752.37  ]),\n",
       "  array([-17631.5403, -16629.4999, -16007.0561, -15611.3005, -15463.1326,\n",
       "         -14921.3944, -14631.9193]),\n",
       "  array([-17477.9796, -16538.2131, -16186.4168, -15524.1062, -15152.344 ,\n",
       "         -14774.2745, -14479.0245, -14230.6628, -13850.9153, -13539.8015,\n",
       "         -13119.9628, -13253.0628, -12924.0387, -12469.9756, -12207.4742,\n",
       "         -11952.0777]),\n",
       "  array([-17307.0357, -16591.7287, -15969.195 , -15698.4093, -15225.1419,\n",
       "         -15128.9349, -14712.6884, -14368.6929, -14582.2154, -13842.9153,\n",
       "         -13563.9472, -13221.2612, -12908.6603, -12758.4102, -12625.8888,\n",
       "         -12330.2712]),\n",
       "  array([-17306.594 , -16561.4117, -15848.4889, -15467.1306, -15113.9257,\n",
       "         -14747.4158, -14510.3926, -14220.5299, -13901.9849, -13551.9625,\n",
       "         -13475.6369]),\n",
       "  array([-17339.2927, -16617.951 , -15979.9421, -15573.2912, -15157.8337,\n",
       "         -15012.5522, -14742.4118, -14243.8382, -13915.0663, -13727.6214,\n",
       "         -13450.4579, -13128.7578, -12957.7096]),\n",
       "  array([-17435.8936, -16721.6975, -16077.843 , -15549.1062, -15241.7966,\n",
       "         -15047.4738, -14637.1158, -14253.7253, -13928.9649, -14113.0954]),\n",
       "  array([-17481.6464, -16805.5483, -16163.5932, -15667.0586, -15480.4445,\n",
       "         -14896.8069, -15266.3083, -14357.4053, -13874.4655, -13498.1336,\n",
       "         -13964.7213, -13493.7189, -12919.5071, -12546.0959, -12421.164 ,\n",
       "         -12255.4743, -12102.141 , -11825.482 , -11775.6646]),\n",
       "  array([-17528.5211, -16827.209 , -16054.6856, -15483.3713, -15051.0805,\n",
       "         -14786.0854, -14686.8696, -13985.4577, -14251.9858, -13432.2545,\n",
       "         -13799.9469, -12950.2333, -12648.4604, -12378.9266, -12109.6648]),\n",
       "  array([-17095.6224, -16489.3834, -15900.3993, -15634.2556, -15380.1482,\n",
       "         -14827.1292, -14620.3737, -14143.6766, -13850.4268, -13526.4339,\n",
       "         -13624.1486, -12972.0311, -12879.8949, -12381.587 , -12143.8538]),\n",
       "  array([-17203.944 , -16448.6719, -16025.0561, -15650.3097, -15226.1816,\n",
       "         -15055.831 , -14700.7029, -14507.7808, -14149.2474, -14057.1553,\n",
       "         -13531.8561, -13415.9544, -12927.7146, -12892.7566, -12571.3645,\n",
       "         -12249.2927, -12224.7506, -12063.5238, -11911.1977, -11692.0312,\n",
       "         -11765.3469]),\n",
       "  array([-17307.6695, -16522.2203, -16061.1282, -15608.3644, -15343.4237,\n",
       "         -15240.1688, -14882.9982, -14443.8479, -14184.9535, -13871.6143,\n",
       "         -13935.6179, -13287.2854, -13513.8404, -12755.6974, -12673.1718,\n",
       "         -12370.1501, -12178.3547])],\n",
       " 6)"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from codeE.methods import ModelInf_EM_CMM as  CMM\n",
    "CMM_model = CMM(M=M_seted) \n",
    "#CMM_model = CMM(M=M_seted, init_Z='model', n_init_Z=5, priors=0) \n",
    "\n",
    "args = {'epochs':1, 'batch_size':BATCH_SIZE, 'optimizer':OPT}\n",
    "CMM_model.set_model(aux_model, **args)\n",
    "\n",
    "# if you want you can initialize step E forcing some method\n",
    "#CMM_model.init_E(r_obs, method=\"hardmv\")\n",
    "#CMM_model.get_qestimation()\n",
    "\n",
    "CMM_model.fit(X_train, r_obs, runs =20,max_iter=EPOCHS_BASE,tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.8857771554310863\n",
      "Test accuracy of model = 0.7494373593398349\n"
     ]
    }
   ],
   "source": [
    "cmm_fx = CMM_model.get_basemodel()\n",
    "\n",
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(cmm_fx, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(cmm_fx, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Global Behavior - Label Noise\n",
    "As CMM with *M=1*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 300)           4200300   \n",
      "_________________________________________________________________\n",
      "cu_dnngru_1 (CuDNNGRU)       (None, 50, 128)           165120    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 50, 128)           0         \n",
      "_________________________________________________________________\n",
      "cu_dnngru_2 (CuDNNGRU)       (None, 64)                37248     \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 2)                 130       \n",
      "=================================================================\n",
      "Total params: 4,402,798\n",
      "Trainable params: 202,498\n",
      "Non-trainable params: 4,200,300\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "aux_model = clone_UB.get_model()\n",
    "aux_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.58 sec)\tlogL: -15471.541\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14804.922\tTol1: 0.04309\tTol2: 0.14822\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -14445.659\tTol1: 0.02427\tTol2: 0.05170\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14291.207\tTol1: 0.01069\tTol2: 0.00385\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.17 sec)\tlogL: -15563.266\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14679.100\tTol1: 0.05681\tTol2: 0.14399\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14816.685\tTol1: 0.00937\tTol2: 0.02538\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.15 sec)\tlogL: -15457.817\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -14681.268\tTol1: 0.05024\tTol2: 0.14887\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14423.291\tTol1: 0.01757\tTol2: 0.01957\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.24 sec)\tlogL: -15376.046\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14683.768\tTol1: 0.04502\tTol2: 0.14667\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14494.078\tTol1: 0.01292\tTol2: 0.03074\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14312.185\tTol1: 0.01255\tTol2: 0.03195\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14237.385\tTol1: 0.00523\tTol2: 0.01625\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.21 sec)\tlogL: -15375.529\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.68 sec)\tlogL: -14731.082\tTol1: 0.04191\tTol2: 0.14817\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14574.217\tTol1: 0.01065\tTol2: 0.03217\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14364.404\tTol1: 0.01440\tTol2: 0.04249\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14257.738\tTol1: 0.00743\tTol2: 0.01316\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.20 sec)\tlogL: -15320.158\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14728.066\tTol1: 0.03865\tTol2: 0.14802\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14704.597\tTol1: 0.00159\tTol2: 0.03645\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14337.328\tTol1: 0.02498\tTol2: 0.05131\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14182.970\tTol1: 0.01077\tTol2: 0.04002\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14067.068\tTol1: 0.00817\tTol2: 0.01255\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.26 sec)\tlogL: -15362.242\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14755.736\tTol1: 0.03948\tTol2: 0.14572\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14534.695\tTol1: 0.01498\tTol2: 0.07400\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14359.676\tTol1: 0.01204\tTol2: 0.01902\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.25 sec)\tlogL: -15500.555\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14901.170\tTol1: 0.03867\tTol2: 0.14514\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14708.867\tTol1: 0.01291\tTol2: 0.02168\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.18 sec)\tlogL: -15334.652\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14685.270\tTol1: 0.04235\tTol2: 0.14817\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14625.658\tTol1: 0.00406\tTol2: 0.04452\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14330.760\tTol1: 0.02016\tTol2: 0.03945\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14174.102\tTol1: 0.01093\tTol2: 0.03630\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14073.504\tTol1: 0.00710\tTol2: 0.01031\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.24 sec)\tlogL: -15308.357\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14783.238\tTol1: 0.03430\tTol2: 0.14707\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14627.502\tTol1: 0.01053\tTol2: 0.02091\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.19 sec)\tlogL: -15396.867\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14860.095\tTol1: 0.03486\tTol2: 0.14729\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14470.976\tTol1: 0.02619\tTol2: 0.03268\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -14618.597\tTol1: 0.01020\tTol2: 0.03776\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14249.451\tTol1: 0.02525\tTol2: 0.04194\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14078.122\tTol1: 0.01202\tTol2: 0.05034\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14017.884\tTol1: 0.00428\tTol2: 0.00233\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.19 sec)\tlogL: -15388.172\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14804.633\tTol1: 0.03792\tTol2: 0.14881\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14451.291\tTol1: 0.02387\tTol2: 0.03464\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14347.130\tTol1: 0.00721\tTol2: 0.01036\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.21 sec)\tlogL: -15267.811\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14636.523\tTol1: 0.04135\tTol2: 0.14806\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14496.172\tTol1: 0.00959\tTol2: 0.03782\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14277.393\tTol1: 0.01509\tTol2: 0.01431\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.20 sec)\tlogL: -15327.654\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14740.277\tTol1: 0.03832\tTol2: 0.14746\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14533.149\tTol1: 0.01405\tTol2: 0.05613\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14289.869\tTol1: 0.01674\tTol2: 0.03728\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14189.678\tTol1: 0.00701\tTol2: 0.01388\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.19 sec)\tlogL: -15499.179\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14674.459\tTol1: 0.05321\tTol2: 0.14442\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14618.477\tTol1: 0.00381\tTol2: 0.04295\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14348.840\tTol1: 0.01844\tTol2: 0.00991\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.22 sec)\tlogL: -15304.265\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14693.902\tTol1: 0.03988\tTol2: 0.14712\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14475.854\tTol1: 0.01484\tTol2: 0.01549\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.17 sec)\tlogL: -15369.291\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14671.328\tTol1: 0.04541\tTol2: 0.14775\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14434.344\tTol1: 0.01615\tTol2: 0.01575\t\n",
      "Finished training!\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.19 sec)\tlogL: -15465.021\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14793.723\tTol1: 0.04341\tTol2: 0.14837\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14552.440\tTol1: 0.01631\tTol2: 0.02653\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.21 sec)\tlogL: -15424.634\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.63 sec)\tlogL: -14670.667\tTol1: 0.04888\tTol2: 0.14491\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14616.854\tTol1: 0.00367\tTol2: 0.02866\t\n",
      "Finished training!\n",
      "Initializing new EM...\n",
      "Beta:  (2, 2)\n",
      "Q estimate:  (4999, 2)\n",
      "Iter 1/50\n",
      "M step: done,  E step: done //  (in 2.23 sec)\tlogL: -15335.066\t\n",
      "Iter 2/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -15036.240\tTol1: 0.01949\tTol2: 0.14577\t\n",
      "Iter 3/50\n",
      "M step: done,  E step: done //  (in 0.65 sec)\tlogL: -14519.736\tTol1: 0.03435\tTol2: 0.03396\t\n",
      "Iter 4/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -14317.637\tTol1: 0.01392\tTol2: 0.03531\t\n",
      "Iter 5/50\n",
      "M step: done,  E step: done //  (in 0.66 sec)\tlogL: -14498.480\tTol1: 0.01263\tTol2: 0.04320\t\n",
      "Iter 6/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -14079.104\tTol1: 0.02893\tTol2: 0.06814\t\n",
      "Iter 7/50\n",
      "M step: done,  E step: done //  (in 0.67 sec)\tlogL: -13915.473\tTol1: 0.01162\tTol2: 0.04326\t\n",
      "Iter 8/50\n",
      "M step: done,  E step: done //  (in 0.64 sec)\tlogL: -13808.277\tTol1: 0.00770\tTol2: 0.00498\t\n",
      "Finished training!\n",
      "20 runs, Epochs to converge=  4.3\n"
     ]
    }
   ],
   "source": [
    "from codeE.methods import ModelInf_EM_G as G_Noise\n",
    "GNoise_model = G_Noise() \n",
    "#GNoise_model = G_Noise(init_Z='model', n_init_Z=5, priors=0) \n",
    "\n",
    "args = {'epochs':1, 'batch_size':BATCH_SIZE, 'optimizer':OPT}\n",
    "GNoise_model.set_model(aux_model, **args)\n",
    "\n",
    "# if you want you can initialize step E forcing some method\n",
    "#GNoise_model.init_E(r_obs, method=\"hardmv\")\n",
    "#GNoise_model.get_qestimation()\n",
    "\n",
    "logL_hists,i_r = GNoise_model.fit(X_train, r_obs, runs =20,max_iter=EPOCHS_BASE,tolerance=TOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train accuracy of model = 0.8409681936387278\n",
      "Test accuracy of model = 0.7514378594648662\n"
     ]
    }
   ],
   "source": [
    "G_fx = GNoise_model.get_basemodel()\n",
    "\n",
    "from codeE.evaluation import accuracy_model\n",
    "print(\"Train accuracy of model =\", accuracy_model(G_fx, X_train, Z_train))\n",
    "print(\"Test accuracy of model =\", accuracy_model(G_fx, X_test, Z_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Confusion Matrices Visualization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "from codeE.utils import generate_confusionM\n",
    "from codeE.utils import plot_confusion_matrix, compare_conf_ma, compare_set_conf_ma\n",
    "from codeE.evaluation import Individual_D, D_JS, D_NormF"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### Global scenario\n",
    "\n",
    "#### MV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.001091619232014267\n",
      "Global D (based on normalized Frobenius) = 0.031010834500193596\n"
     ]
    }
   ],
   "source": [
    "### SOFT\n",
    "prob_Yz = generate_confusionM(mv_soft, r_obs) #confusion matrix of all annotations\n",
    "compare_conf_ma(prob_Yz, true_conf_mat = confe_matrix_G, text=True)\n",
    "\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.0018885730640643085\n",
      "Global D (based on normalized Frobenius) = 0.03987046331167221\n"
     ]
    }
   ],
   "source": [
    "### HARD\n",
    "prob_Yz = generate_confusionM(mv_hard, r_obs) \n",
    "compare_conf_ma(prob_Yz, true_conf_mat = confe_matrix_G, text=True)\n",
    "\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### CMM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.05722243437245026\n",
      "Global D (based on normalized Frobenius) = 0.16409692932287706\n"
     ]
    }
   ],
   "source": [
    "prob_Yz = CMM_model.get_global_confusionM()\n",
    "\n",
    "compare_conf_ma(prob_Yz, true_conf_mat = confe_matrix_G, text=True)\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Global - Label Noise (without predictive model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.0012072251637555709\n",
      "Global D (based on normalized Frobenius) = 0.032839953899383545\n"
     ]
    }
   ],
   "source": [
    "prob_Yz = LabG_Noise_model.get_global_confusionM()\n",
    "compare_conf_ma(prob_Yz, true_conf_mat = confe_matrix_G, text=True)\n",
    "\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Global - Label Noise (with predictive model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Global D (based on Jensen Shannon) = 0.0012716128481904097\n",
      "Global D (based on normalized Frobenius) = 0.032570142298936844\n"
     ]
    }
   ],
   "source": [
    "prob_Yz = GNoise_model.get_global_confusionM()\n",
    "compare_conf_ma(prob_Yz, true_conf_mat = confe_matrix_G, text=True)\n",
    "\n",
    "print(\"Global D (based on Jensen Shannon) =\",D_JS(confe_matrix_G, prob_Yz))\n",
    "print(\"Global D (based on normalized Frobenius) =\",D_NormF(confe_matrix_G, prob_Yz))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    ">### Individual scenario\n",
    "\n",
    "#### DS"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot 5 random matrices from the set\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.007928023320919486\n",
      "D (based on normalized Frobenius) = 0.05979391932487488\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 3.820828041929696e-05\n",
      "D (based on normalized Frobenius) = 0.003704364877194166\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.001264197104082796\n",
      "D (based on normalized Frobenius) = 0.0332149937748909\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.07904647250908188\n",
      "D (based on normalized Frobenius) = 0.20174914598464966\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 7.610551409113501e-05\n",
      "D (based on normalized Frobenius) = 0.00894396472722292\n"
     ]
    }
   ],
   "source": [
    "prob_Yzt = DS_model.get_confusionM() # estimate of conf_matrix_I\n",
    "\n",
    "compare_set_conf_ma(prob_Yzt, true_set_conf_ma = conf_matrix_I, text=True, n_samp=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual D_JS =  0.020501345468101718\n",
      "Individual D_NormF =  0.07443660041759495\n"
     ]
    }
   ],
   "source": [
    "print(\"Individual D_JS = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_JS))\n",
    "print(\"Individual D_NormF = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_NormF))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Raykar"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot 5 random matrices from the set\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.004903300954412042\n",
      "D (based on normalized Frobenius) = 0.05347186326980591\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAADpCAYAAABxwaO9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAVAklEQVR4nO3de3hU9Z3H8c83mUDkkgACSXNRLkOxoZa6cml3EbpquagEFhWorbr1VqvQXXer9LIVq1XpPtvto4itrSheKlHZakCBgFpBRUWoYktaTR6CJkGIAoaAkMDw2z8yxiQkJJHMzG8m79fz5HnmnPOb33znDN985pw5E8w5JwAAfJMU6wIAAGgJAQUA8BIBBQDwEgEFAPASAQUA8BIBBQDwEgEFIGGZ2Vlm9k6s62iJmX3DzCpiXYfPCCjPmNn+Rj9Hzexgo+Vvx7o+IFrMbHuzf//7zeyeNu7jzCz46bJz7iXn3PAI1bfEzH4RiblRLxDrAtCUc67Xp7fNbLukq5xzz7U23swCzrkj0agNiIGpx/v3j8TGEVScMbNfmNnjZrbUzGokfcfMHjWzWxqNOTccbp8u55jZU2b2oZmVmdn1MSgd6BRmFjSzdWZWbWYfmdnj4fXrw0O2hI+2ZjU/jRY+KrvRzN42swNmttjMMsxslZnVmNlzZta30fgnzWxn+LHWm9mI8PprJH1b0k3hx1oRXp9lZv/XqNd+0Giuk8JHXXvNrFjS6MjvrfhGQMWnf5H0mKR0SY8fb6CZJUt6RtIbkrIlfVPSjWZ2TqSLBCLkNklrJPWVlCNpoSQ558aHt490zvVyzrXWGxeqvg++KGmqpFWSfiKpv+p/J/6g0dhVkoZJGijpz5L+EH6s34Vv/3f4saaaWZKkFZK2qL7XzpH072Y2KTzXfElDwz+TJF1+AvugSyCg4tPLzrkVzrmjzrmDbYz9mqQ059wdzrk651yppMWSZke+TOCEPW1mHzf6uVrSYUmnSspyzh1yzr3cwTkXOud2OecqJb0k6XXn3JvOuVpJT0k649OBzrkHnHM14W23SBppZumtzDta0gDn3K3hXtsm6ff6rNdmSrrdObfHOVcu6e4O1t3l8BlUfCrvwNhTJZ1iZh83Wpcs6cVOrQiIjOnNP4MKn067TdJGM9sr6VfOuQc6MOeuRrcPtrDcK/w4yZJul3SxpAGSjobH9JdU3cK8p0rKaqHXXgrfzlLT3n2vAzV3SQRUfGr+J+gPSOrRaDmz0e1ySSXOuS9FvCogCpxzOyVdLUlmNk7Sc2a2Pnx2oDNdImmapHMlbVf9KfW9kuzTUpqNL5dU5pwb1sp8H0jKlbQ1vHxKZxabiDjFlxjeknS+mfU1sy+o6Tn0VyXVmdl/mlmqmSWb2elmdmZsSgVOjJldbGY54cW9qg+KUHh5l6QhnfRQvSXVStqt+jeAdzTb3vyxNkraZ2bzwhdEJJvZl83s04shnpD043Cf5kia20l1JiwCKjEskfQ31Z8yWC2p4NMN4UvQz5M0RvXvAj+SdJ+ktGgXCXwOK5p9D+op1X/W87qZ7Ze0XNK/OefKwuNvkfRQ+POqmSf42A+rvqcqJRVLeq3Z9sWS8sKP9bRzLqT6iy6+KqlM9b12v+qPvCTp5+H5ylR/kccjJ1hfwjP+w0IAgI84ggIAeImAAgB4iYACAHiJgAIAeCki34M6OZDsclP4ilU0BQa39uV2RMr2HTX66OOD1vbIz6eHJbs+SonU9GhB1pmddYU6OmLz5r995Jwb0Hx9RFIkNyWgtYOzIjE1WtH/0SmxLqHLGf2dZRGdv49S9D2dGtHHQFPzN3HldyyYjWrxr2pwig8A4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgpS4TUCnjxqvvs8+r3+o/6aSrrj1me895/6W+f3y2/mflCzr5tS0N29LvW6KTX9uitHvvj2bJcW/1hvd12ozHNGzaH7TgwT8fs/1/H92iERcVaOSsx3Xutcv13gc1Ddvm3f2qTp9ZoNNnFujxNaXRLBvHkb/4Dv1w1wZ9/y8rWh0z+a6fam7JGl27Zbkyz8hrWD/ysuma826R5rxbpJGXTY9GuQlj9eoNGj58hoLB6VqwYMkx22tr6zRr1o8VDE7X2LGXa/v2HQ3b7rzzQQWD0zV8+AwVFb0axapPXNcIqKQk9f6vW1X9vX/VnqkTlXpevpKHBpsMOfDLX2jvjPO1d8b5OviHh1T73OqGbZ88+DvV/Og/ol11XAuFjmrOgpe08u4LtHXZbBUUlap4254mY84Y3l9vPHKhtjw+SxeeM0Tz7qpvnmdfek9v/v0jvfnYTL320IX6n4ff0r79dbF4GmjmrSV/1KOTr2p1e3DKePUbNkgLh03Uimt+pvN/c4skKbVvuibMn6P7x87U/WMu1oT5c5TaJy1KVce3UCik66//pVatulvFxU9q6dIiFRdvazJm8eJC9e3bW6WlT+uGGy7RvHkLJUnFxdtUULBGW7c+odWrF+q66xYoFArF4ml8Ll0ioAKnj1To/fd0tKJcOnxYh1atULezv9nq+NTzpqr22c/eIR5+bYPcgf3RKDVhbNxapWBuuobkpKlbSrJmTQyq8MXtTcb88+hs9TgpRZL0tdMzVFF1QJJUXLZH4/8hS4FAknqelKKvfPFkrd7wfrSfAlrw/kubdHBPdavbT5t2jt5++GlJUuXrW5TaJ029MgcoOGmctq19RYf2VuvQx/u0be0rCk4+K1plx7WNG7cqGMzVkCE56tYtRbNnT1Rh4bomYwoL1+nyyy+QJF100Tl6/vmNcs6psHCdZs+eqO7du2nw4GwFg7nauHFrLJ7G59IlAiopI1OhnR80LB/duVPJAzNbHpuVraScXB1+fUO0yktIlVUHlJPRs2E5J6OnKj880Or4xYV/1+R/PEWSNHJYf63e8L4+OXhYH+09qBc3Vap8F28Q4kHv7AxVl+9sWN5XsVO9szNaWL9LvbMzYlFi3KmsrFJu7mf7KidnoCorq1odEwgElJ7eS7t3V7frvj4LtGeQmU2WdJekZEn3O+cWRLSqzmbWwkrX4tDuUy5Q3ZpV0tGjka0pwbkWdm+LL4OkR1e+q83FVXrx9/WfS0z8eq7eKK7SP13xlAb0PUlfPz1TgeTEeC8V973UlpZeZOdkraxH21ruJWvHmPbd12dtdr2ZJUtaJGmKpDxJ3zKzvOPfyy9Hd36g5MwvNCwnZWYqVLWrxbHdz5uqQ88uj1ZpCSsno6cqdn12xFSx64Cy+vc8Ztxzr1fojsWbVfjrKereLblh/U+vPFNvLp2pNfdOlXNOw05Jj0rdkZQIvdSWmoqdSs/97OxEWk6manZUad8x6zNUsyN+3snHUk7OQJWXf/b7qqKiSllZA1odc+TIEVVX71e/funtuq/P2vO2dIykUufcNudcnaQCSdMiW1bnOvLXt5V86iAlZedIKSlKnTJVdX967phxyYOGKCktXUfeOvaKM3TM6LyBKin/WGWV+1R3OKTH15Qqf8KgJmPe/PuHuvb2dSr89RQN7NejYX0odFS7Pz4kSXq7ZLfeLt2tiV/LjWb5kRL3vdSWd5a/oK+Er9DLHjtStdU12r/zQ5UWvawhE8cptU+aUvukacjEcSotejnG1caH0aPzVFJSrrKyStXVHVZBwRrl549vMiY/f7weeugZSdKyZc/r7LNHy8yUnz9eBQVrVFtbp7KySpWUlGvMmBGxeBqfS3tO8WVLKm+0XCFpbPNBZnaNpGskKSeQ3HxzbIVC2n/7fKX//mFZUpIOPfWkQqUl6jHnBh3Z+peGsOp+/lTVrjz28tk+jzyh5MFDZD16qt8LG1Tzsx/p8Cvro/0s4kogkKSFN52lyXOeUSjk9N1pp2nE0H66+TcbNSpvgPInDNZNd72q/QcPa+a8NZKkUzJ7qfDX5+nwkaMaf1X9B+1pPVP0yG3nKhBIiFN8bfZS4z5Kb98Z+Kia8divNOgbY9Sjf1/dUL5OL85fqKSU+jo331egkpXrNOy8CZpbulaHPzmowu/+RJJ0aG+11t92r65+Y5kkaf2ti3Rob+sXW+AzgUBA99xzoyZNmqtQKKQrrsjXiBFDdfPNv9WoUV9Sfv4EXXnlNF166c0KBqerX780FRTcIUkaMWKoZs48V3l5FysQSNaiRTcpOdmz38/HYa6N88BmdrGkSc65q8LLl0oa45yb29p9vnpSd7d2cFanForj6//olFiX0OWM/s4ybSquavcJ/Y72Upaluu/p1M4pFu0y3z0W6xK6JLNRm51zo5qvb8/b0gpJjc+v5Eja0cpYAK2jl4AOaE9AvSFpmJkNNrNukmZL4ioCoOPoJaAD2jzJ7Zw7YmZzJBWp/tLYB5xz8fNNL8AT9BLQMe36FNY5t1LSygjXAiQ8eglov4S4NAoAkHgIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcIKACAlwgoAICXCCgAgJcCEZl0xFAN2PRIJKZGK35ul8S6hC5nh/ZFdP6sM4doPn0UVfSRXziCAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeImAAgB4iYACAHiJgAIAeKnLBNTq1Rs0fPgMBYPTtWDBkmO219bWadasHysYnK6xYy/X9u07GrbdeeeDCgana/jwGSoqejWKVcev/MV36Ie7Nuj7f1nR6pjJd/1Uc0vW6Noty5V5Rl7D+pGXTdecd4s0590ijbxsejTKRTvRR9HXlXupSwRUKBTS9df/UqtW3a3i4ie1dGmRiou3NRmzeHGh+vbtrdLSp3XDDZdo3ryFkqTi4m0qKFijrVuf0OrVC3XddQsUCoVi8TTiyltL/qhHJ1/V6vbglPHqN2yQFg6bqBXX/Ezn/+YWSVJq33RNmD9H94+dqfvHXKwJ8+cotU9alKrG8dBHsdGVe6lLBNTGjVsVDOZqyJAcdeuWotmzJ6qwcF2TMYWF63T55RdIki666Bw9//xGOedUWLhOs2dPVPfu3TR4cLaCwVxt3Lg1Fk8jrrz/0iYd3FPd6vbTpp2jtx9+WpJU+foWpfZJU6/MAQpOGqdta1/Rob3VOvTxPm1b+4qCk8+KVtk4DvooNrpyL3WJgKqsrFJubkbDck7OQFVWVrU6JhAIKD29l3bvrm7XfdFxvbMzVF2+s2F5X8VO9c7OaGH9LvXOzmhpCkQZfeSnRO6lNgPKzB4wsyoz+2s0CooE545dZ2btGNO+++JzaGkfOtfyvm3pRYhD8d5L9JGnEriX2nMEtUTS5AjXEVE5OQNVXr6rYbmiokpZWQNaHXPkyBFVV+9Xv37p7bovOq6mYqfSczMbltNyMlWzo0r7jlmfoZodCfNOe4niuJfoIz8lci+1GVDOufWS9kShlogZPTpPJSXlKiurVF3dYRUUrFF+/vgmY/Lzx+uhh56RJC1b9rzOPnu0zEz5+eNVULBGtbV1KiurVElJucaMGRGLp5FQ3ln+gr4Svqooe+xI1VbXaP/OD1Va9LKGTByn1D5pSu2TpiETx6m06OUYV9s54r2X6CM/JXIvBTprIjO7RtI1knTKKZltjI6uQCCge+65UZMmzVUoFNIVV+RrxIihuvnm32rUqC8pP3+Crrxymi699GYFg9PVr1+aCgrukCSNGDFUM2eeq7y8ixUIJGvRopuUnJwc42fkvxmP/UqDvjFGPfr31Q3l6/Ti/IVKSqn/57b5vgKVrFynYedN0NzStTr8yUEVfvcnkqRDe6u1/rZ7dfUbyyRJ629dpEN7W/+AONHQR2iuK/eSuXackzSzQZKecc59uT2TjhqV5zZteuTEKkOH/NwuiXUJXc59ek873KEOfZDSkV6ij6KPPoqNW/TuZufcqObru8RVfACA+ENAAQC81J7LzJdKelXScDOrMLMrI18WkHjoJaBj2rxIwjn3rWgUAiQ6egnoGE7xAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxEQAEAvERAAQC8REABALxkzrnOn9TsQ0nvdfrEkddf0kexLqKLied9fqpzbkCkJo/jPpLi+3WNV/G8z1vspYgEVLwys03OuVGxrqMrYZ8nJl7X6EvEfc4pPgCAlwgoAICXCKimfhfrArog9nli4nWNvoTb53wGBQDwEkdQAAAvEVAAAC8RUGFmNtnM3jGzUjP7UazrSXRm9oCZVZnZX2NdCzoPfRR9idxLBJQkM0uWtEjSFEl5kr5lZnmxrSrhLZE0OdZFoPPQRzGzRAnaSwRUvTGSSp1z25xzdZIKJE2LcU0JzTm3XtKeWNeBTkUfxUAi9xIBVS9bUnmj5YrwOgDtRx+hUxFQ9ayFdVx/D3QMfYRORUDVq5CU22g5R9KOGNUCxCv6CJ2KgKr3hqRhZjbYzLpJmi1peYxrAuINfYRORUBJcs4dkTRHUpGkv0l6wjm3NbZVJTYzWyrpVUnDzazCzK6MdU04MfRRbCRyL/GnjgAAXuIICgDgJQIKAOAlAgoA4CUCCgDgJQIKAOAlAgoA4CUCCgDgpf8HGEkb4PrEBqwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.0800556882021762\n",
      "D (based on normalized Frobenius) = 0.2020305097103119\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.013482437491257137\n",
      "D (based on normalized Frobenius) = 0.08998911827802658\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.01138416853386328\n",
      "D (based on normalized Frobenius) = 0.056896161288022995\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 2.4903886140869635e-05\n",
      "D (based on normalized Frobenius) = 0.004070466384291649\n"
     ]
    }
   ],
   "source": [
    "prob_Yzt = R_model.get_confusionM() # estimate of conf_matrix_I\n",
    "\n",
    "compare_set_conf_ma(prob_Yzt, true_set_conf_ma = conf_matrix_I, text=True, n_samp=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual D_JS =  0.023342278372848514\n",
      "Individual D_NormF =  0.07746179691822985\n"
     ]
    }
   ],
   "source": [
    "print(\"Individual D_JS = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_JS))\n",
    "print(\"Individual D_NormF = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_NormF))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### C-MoA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot 5 random matrices from the set\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.10963026645604498\n",
      "D (based on normalized Frobenius) = 0.30845511827376915\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.15677884577100173\n",
      "D (based on normalized Frobenius) = 0.3956671896014532\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.01012296733783246\n",
      "D (based on normalized Frobenius) = 0.06791736183155961\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.0930999525758121\n",
      "D (based on normalized Frobenius) = 0.22539713550125665\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "D (based on Jensen Shannon) = 0.019432817111342624\n",
      "D (based on normalized Frobenius) = 0.0536721610702688\n"
     ]
    }
   ],
   "source": [
    "A = np.unique(np.concatenate(A_idx_var)).reshape(-1,1)\n",
    "prob_Yzt = CMOA_model.get_ann_confusionM(A) # estimate of conf_matrix_I\n",
    "\n",
    "compare_set_conf_ma(prob_Yzt, true_set_conf_ma = conf_matrix_I, text=True, n_samp=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Individual D_JS =  0.08213268136578118\n",
      "Individual D_NormF =  0.19953688467811037\n"
     ]
    }
   ],
   "source": [
    "print(\"Individual D_JS = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_JS))\n",
    "print(\"Individual D_NormF = \",Individual_D(conf_matrix_I, prob_Yzt, D=D_NormF))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Reliability"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean score = 0.987019123150916\n",
      "Maximum score = 0.9966951370239258\n",
      "Minimum score = 0.9051119804382324\n"
     ]
    }
   ],
   "source": [
    "sns.distplot(LogReg_crowds.get_ann_rel().flatten())\n",
    "plt.xlabel(\"Reliability score\")\n",
    "plt.show()\n",
    "\n",
    "print(\"Mean score =\", LogReg_crowds.get_ann_rel().mean())\n",
    "print(\"Maximum score =\", LogReg_crowds.get_ann_rel().max())\n",
    "print(\"Minimum score =\", LogReg_crowds.get_ann_rel().min())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The behaviors of the groups found by CMM and CMOA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "from codeE.evaluation import I_sim, R_score, H_conf, S_score, S_bias"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "**For CMM**\n",
      "Group  1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.9929995516330097\n",
      "Expertise Diagonal (R_score) = 0.9861376691360852\n",
      "Randomness (H_conf) = 0.09172793802944372\n",
      "Spammer score (S_score) = 0.9722753382721707\n",
      "Bias for class index 1 with score = 0.013804\n",
      "-------------------------------------------------\n",
      "Group  2\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.5010092213300428\n",
      "Expertise Diagonal (R_score) = 0.500153826540473\n",
      "Randomness (H_conf) = 0.0020184997012979533\n",
      "Spammer score (S_score) = 0.00030765308094613295\n",
      "Bias for class index 0 with score = 0.499846\n",
      "-------------------------------------------------\n",
      "Group  3\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.5000041940002948\n",
      "Expertise Diagonal (R_score) = 0.5000002635989863\n",
      "Randomness (H_conf) = 8.660368989689335e-06\n",
      "Spammer score (S_score) = 5.271979727242487e-07\n",
      "Bias for class index 1 with score = 0.500000\n",
      "-------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "print(\"**For CMM**\")\n",
    "\n",
    "B = CMM_model.get_confusionM()\n",
    "\n",
    "for i, b in enumerate(B):\n",
    "    print(\"Group \",(i+1))\n",
    "    compare_conf_ma(b, text=True)\n",
    "\n",
    "    print(\"Expertise Identity (I_sim) =\", I_sim(b))\n",
    "    print(\"Expertise Diagonal (R_score) =\", R_score(b))\n",
    "    print(\"Randomness (H_conf) =\", H_conf(b))\n",
    "    print(\"Spammer score (S_score) =\", S_score(b))\n",
    "    print(\"Bias for class index %d with score = %f\"%(S_bias(b)))\n",
    "    print(\"-------------------------------------------------\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "**For C-MoA**\n",
      "Group  1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.8111032897771429\n",
      "Expertise Diagonal (R_score) = 0.710064045387271\n",
      "Randomness (H_conf) = 0.49235854488865377\n",
      "Spammer score (S_score) = 0.4201280907745419\n",
      "Bias for class index 1 with score = 0.289707\n",
      "-------------------------------------------------\n",
      "Group  2\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.9837356457797797\n",
      "Expertise Diagonal (R_score) = 0.96821294829132\n",
      "Randomness (H_conf) = 0.1725189041405809\n",
      "Spammer score (S_score) = 0.9364258965826399\n",
      "Bias for class index 0 with score = 0.031383\n",
      "-------------------------------------------------\n",
      "Group  3\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Expertise Identity (I_sim) = 0.9999971000767902\n",
      "Expertise Diagonal (R_score) = 0.9999935607685154\n",
      "Randomness (H_conf) = 0.00011997170394597606\n",
      "Spammer score (S_score) = 0.9999871215370308\n",
      "Bias for class index 1 with score = 0.000002\n",
      "-------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "print(\"**For C-MoA**\")\n",
    "\n",
    "B = CMOA_model.get_confusionM()\n",
    "\n",
    "for i, b in enumerate(B):\n",
    "    print(\"Group \",(i+1))\n",
    "    compare_conf_ma(b, text=True)\n",
    "\n",
    "    print(\"Expertise Identity (I_sim) =\", I_sim(b))\n",
    "    print(\"Expertise Diagonal (R_score) =\", R_score(b))\n",
    "    print(\"Randomness (H_conf) =\", H_conf(b))\n",
    "    print(\"Spammer score (S_score) =\", S_score(b))\n",
    "    print(\"Bias for class index %d with score = %f\"%(S_bias(b)))\n",
    "    print(\"-------------------------------------------------\")"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda env:newpy3_tf1]",
   "language": "python",
   "name": "conda-env-newpy3_tf1-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}