Optimized_embedding.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\aylme\\Anaconda3\\lib\\site-packages\\h5py\\__init__.py:34: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "336000\n",
      "336000\n",
      "Success\n"
     ]
    }
   ],
   "source": [
    "import io\n",
    "import numpy as np\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.feature_extraction.text import TfidfTransformer\n",
    "from sklearn.linear_model import LogisticRegression \n",
    "from keras.utils import to_categorical\n",
    "\n",
    "###########################################################################\n",
    "#generate training/test data\n",
    "########################################################################### \n",
    "bragging_file = io.open(\"cleantweets/Happy_clean.txt\",'r',encoding=\"utf-8\") \n",
    "bitching_file = io.open(\"cleantweets/FML_clean.txt\",'r',encoding=\"utf-8\") \n",
    "neither_file = io.open(\"cleantweets/Neither_clean.txt\",'r',encoding=\"utf-8\") \n",
    "\n",
    "training_samples = 30000\n",
    "bragging_tweets = []\n",
    "bitching_tweets = []\n",
    "neither_tweets = []\n",
    "\n",
    "for line in bragging_file:\n",
    "    bragging_tweets.append(line.strip(\"\\n\"))\n",
    "for line in bitching_file:\n",
    "    bitching_tweets.append(line.strip(\"\\n\"))\n",
    "for line in neither_file:\n",
    "    neither_tweets.append(line.strip(\"\\n\"))\n",
    "\n",
    "np.random.seed(42)\n",
    "bragging_tweets = np.array(bragging_tweets)\n",
    "bitching_tweets = np.array(bitching_tweets)\n",
    "neither_tweets = np.array(neither_tweets)\n",
    "np.random.shuffle(bragging_tweets)\n",
    "np.random.shuffle(bitching_tweets)\n",
    "np.random.shuffle(neither_tweets)\n",
    "\n",
    "bi_data = []\n",
    "br_data = []\n",
    "for i in range(240000):\n",
    "    bi_data.append([bitching_tweets[i],1])\n",
    "    br_data.append([bragging_tweets[i],1])\n",
    "    \n",
    "for i in range(240000):\n",
    "    bi_data.append([neither_tweets[i],0])\n",
    "    br_data.append([neither_tweets[i],0])\n",
    "    \n",
    "\n",
    "\n",
    "np.random.shuffle(bi_data)\n",
    "np.random.shuffle(br_data)\n",
    "\n",
    "bi_tweets_train = np.array(bi_data[:336000])\n",
    "bi_tweets_test = np.array(bi_data[336000:480000])\n",
    "\n",
    "br_tweets_train = np.array(br_data[:336000])\n",
    "br_tweets_test = np.array(br_data[336000:480000])\n",
    "\n",
    "bi_X_train = bi_tweets_train[:,0]\n",
    "bi_Y_train = to_categorical(bi_tweets_train[:,1])\n",
    "bi_X_test = bi_tweets_test[:,0]\n",
    "bi_Y_test = to_categorical(bi_tweets_test[:,1])\n",
    "\n",
    "br_X_train = br_tweets_train[:,0]\n",
    "br_Y_train = to_categorical(br_tweets_train[:,1])\n",
    "br_X_test = br_tweets_test[:,0]\n",
    "br_Y_test = to_categorical(br_tweets_test[:,1])\n",
    "\n",
    "print(len(bi_X_train))\n",
    "print(len(bi_Y_train))\n",
    "\n",
    "print(\"Success\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 337459 unique tokens.\n"
     ]
    }
   ],
   "source": [
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "import numpy as np\n",
    "\n",
    "maxlen = 50\n",
    "training_samples = 200000\n",
    "max_words = 10000\n",
    "tokenizer = Tokenizer(num_words = max_words)\n",
    "tokenizer.fit_on_texts(bi_X_train)\n",
    "sequences = tokenizer.texts_to_sequences(bi_X_train)\n",
    "word_index = tokenizer.word_index\n",
    "print ('Found %s unique tokens.' % len(word_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shape of the data tensor: (336000, 50)\n",
      "shape of the label tensor:  (336000, 2)\n"
     ]
    }
   ],
   "source": [
    "data = pad_sequences(sequences, maxlen = maxlen)\n",
    "labels = np.asarray(bi_Y_train)\n",
    "print (\"shape of the data tensor:\", data.shape)\n",
    "print (\"shape of the label tensor: \", labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 1193514 word vectors.\n"
     ]
    }
   ],
   "source": [
    "# splits data into a training set and a validation set\n",
    "# indices = np.arange(data.shape[0])\n",
    "# np.random.shuffle(indices)\n",
    "# data = data[indices]\n",
    "# labels = labels[indices]\n",
    "import os\n",
    "\n",
    "bi_x_train = data[:]\n",
    "bi_y_train = labels[:]\n",
    "# x_val = data[training_samples:]\n",
    "# y_val = labels[training_samples:]\n",
    "\n",
    "glove_dir = '/Users/aylme/workspace/CS175/glove.twitter.27B'\n",
    "embeddings_index = {}\n",
    "#your maxlen and you .txt files must correspond\n",
    "f = open(os.path.join(glove_dir, 'glove.twitter.27B.50d.txt'), encoding=\"utf8\")\n",
    "for line in f:\n",
    "    values = line.split()\n",
    "    word = values[0]\n",
    "    coefs=np.asarray(values[1:], dtype='float32')\n",
    "    embeddings_index[word] = coefs\n",
    "f.close()\n",
    "print('Found %s word vectors.'%len(embeddings_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "embedding_dim = maxlen\n",
    "embedding_matrix = np.zeros((max_words, embedding_dim))\n",
    "for word, i in word_index.items():\n",
    "    if (i < max_words):\n",
    "        embedding_vector= embeddings_index.get(word)\n",
    "        if (embedding_vector is not None):\n",
    "            embedding_matrix[i] = embedding_vector\n",
    "            "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 50, 50)            500000    \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 2500)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 32)                80032     \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 2)                 66        \n",
      "=================================================================\n",
      "Total params: 580,098\n",
      "Trainable params: 580,098\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Embedding, Flatten, Dense\n",
    "\n",
    "bi_model = Sequential()\n",
    "bi_model.add(Embedding(max_words, embedding_dim, input_length = maxlen))\n",
    "bi_model.add(Flatten())\n",
    "bi_model.add(Dense(32, activation = 'relu'))\n",
    "bi_model.add(Dense(2, activation = 'softmax'))\n",
    "bi_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#bi_model.layers[0].set_weights([embedding_matrix])\n",
    "#bi_model.layers[0].trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 302400 samples, validate on 33600 samples\n",
      "Epoch 1/3\n",
      "302400/302400 [==============================] - 123s 406us/step - loss: 0.1993 - acc: 0.9159 - val_loss: 0.1321 - val_acc: 0.9517\n",
      "Epoch 2/3\n",
      "302400/302400 [==============================] - 119s 394us/step - loss: 0.1113 - acc: 0.9595 - val_loss: 0.1088 - val_acc: 0.9602\n",
      "Epoch 3/3\n",
      "302400/302400 [==============================] - 121s 400us/step - loss: 0.0982 - acc: 0.9642 - val_loss: 0.1042 - val_acc: 0.9621\n"
     ]
    }
   ],
   "source": [
    "import keras.optimizers\n",
    "sgd = keras.optimizers.SGD(lr=0.0001*400)\n",
    "bi_model.compile(optimizer = sgd,\n",
    "             loss = 'categorical_crossentropy',\n",
    "             metrics = ['acc'])\n",
    "\n",
    "bi_history = bi_model.fit(bi_x_train, bi_y_train, epochs = 3, batch_size= 32, validation_split = 0.1 )\n",
    "bi_model.save_weights('bi_pre-trained_glove_twitter_model.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAEICAYAAACzliQjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIhHyv5FmP+m7ROxUAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23252fd6908>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAEICAYAAACzliQjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSroAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x232c4a9f6d8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "acc = bi_history.history['acc']\n",
    "val_acc = bi_history.history['val_acc']\n",
    "loss = bi_history.history['loss']\n",
    "val_loss = bi_history.history['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc) +1)\n",
    "\n",
    "plt.plot(epochs, acc, 'bo', label = 'Training acc')\n",
    "plt.plot(epochs, val_acc, 'b', label = 'Validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(epochs, loss, 'bo', label = 'Training loss')\n",
    "plt.plot(epochs, val_loss, 'b', label = 'Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "144000/144000 [==============================] - 12s 80us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.10447153980222841, 0.961625]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## These will be results on the test\n",
    "sequences = tokenizer.texts_to_sequences(bi_X_test)\n",
    "bi_x_test = pad_sequences(sequences, maxlen = maxlen)\n",
    "bi_y_test = np.asarray(bi_Y_test)\n",
    "bi_model.load_weights('bi_pre-trained_glove_twitter_model.h5')\n",
    "bi_model.evaluate(bi_x_test, bi_y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "##################################################################################\n",
    "#BRAGGING MODEL\n",
    "##################################################################################"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 386728 unique tokens.\n"
     ]
    }
   ],
   "source": [
    "tokenizer = Tokenizer(num_words = max_words)\n",
    "tokenizer.fit_on_texts(br_X_train)\n",
    "sequences = tokenizer.texts_to_sequences(br_X_train)\n",
    "word_index = tokenizer.word_index\n",
    "print ('Found %s unique tokens.' % len(word_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shape of the data tensor: (336000, 50)\n",
      "shape of the label tensor:  (336000, 2)\n"
     ]
    }
   ],
   "source": [
    "data = pad_sequences(sequences, maxlen = maxlen)\n",
    "labels = np.asarray(br_Y_train)\n",
    "print (\"shape of the data tensor:\", data.shape)\n",
    "print (\"shape of the label tensor: \", labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 1193514 word vectors.\n"
     ]
    }
   ],
   "source": [
    "br_x_train = data[:]\n",
    "br_y_train = labels[:]\n",
    "# x_val = data[training_samples:]\n",
    "# y_val = labels[training_samples:]\n",
    "\n",
    "glove_dir = '/Users/aylme/workspace/CS175/glove.twitter.27B'\n",
    "embeddings_index = {}\n",
    "#your maxlen and you .txt files must correspond\n",
    "f = open(os.path.join(glove_dir, 'glove.twitter.27B.50d.txt'), encoding=\"utf8\")\n",
    "for line in f:\n",
    "    values = line.split()\n",
    "    word = values[0]\n",
    "    coefs=np.asarray(values[1:], dtype='float32')\n",
    "    embeddings_index[word] = coefs\n",
    "f.close()\n",
    "print('Found %s word vectors.'%len(embeddings_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "embedding_dim = maxlen\n",
    "embedding_matrix = np.zeros((max_words, embedding_dim))\n",
    "for word, i in word_index.items():\n",
    "    if (i < max_words):\n",
    "        embedding_vector= embeddings_index.get(word)\n",
    "        if (embedding_vector is not None):\n",
    "            embedding_matrix[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_2 (Embedding)      (None, 50, 50)            500000    \n",
      "_________________________________________________________________\n",
      "flatten_2 (Flatten)          (None, 2500)              0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 32)                80032     \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 2)                 66        \n",
      "=================================================================\n",
      "Total params: 580,098\n",
      "Trainable params: 580,098\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "br_model = Sequential()\n",
    "br_model.add(Embedding(max_words, embedding_dim, input_length = maxlen))\n",
    "br_model.add(Flatten())\n",
    "br_model.add(Dense(32, activation = 'relu'))\n",
    "br_model.add(Dense(2, activation = 'softmax'))\n",
    "br_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#br_model.layers[0].set_weights([embedding_matrix])\n",
    "#br_model.layers[0].trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 302400 samples, validate on 33600 samples\n",
      "Epoch 1/3\n",
      "302400/302400 [==============================] - 125s 413us/step - loss: 0.2454 - acc: 0.8895 - val_loss: 0.1688 - val_acc: 0.9328\n",
      "Epoch 2/3\n",
      "302400/302400 [==============================] - 120s 396us/step - loss: 0.1619 - acc: 0.9349 - val_loss: 0.1628 - val_acc: 0.9340\n",
      "Epoch 3/3\n",
      "302400/302400 [==============================] - 120s 395us/step - loss: 0.1473 - acc: 0.9411 - val_loss: 0.1601 - val_acc: 0.9363\n"
     ]
    }
   ],
   "source": [
    "br_model.compile(optimizer = sgd,\n",
    "             loss = 'categorical_crossentropy',\n",
    "             metrics = ['acc'])\n",
    "\n",
    "br_history = br_model.fit(br_x_train, br_y_train, epochs = 3, batch_size= 32, validation_split = 0.1 )\n",
    "br_model.save_weights('br_pre-trained_glove_twitter_model.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAEICAYAAACzliQjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nD00P7RulUSaNAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x232465bf748>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAEICAYAAACzliQjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x232469b8630>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "acc = br_history.history['acc']\n",
    "val_acc = br_history.history['val_acc']\n",
    "loss = br_history.history['loss']\n",
    "val_loss = br_history.history['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc) +1)\n",
    "\n",
    "plt.plot(epochs, acc, 'bo', label = 'Training acc')\n",
    "plt.plot(epochs, val_acc, 'b', label = 'Validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(epochs, loss, 'bo', label = 'Training loss')\n",
    "plt.plot(epochs, val_loss, 'b', label = 'Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "144000/144000 [==============================] - 16s 113us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.1598797175321314, 0.9346944444444445]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sequences = tokenizer.texts_to_sequences(br_X_test)\n",
    "br_x_test = pad_sequences(sequences, maxlen = maxlen)\n",
    "br_y_test = np.asarray(br_Y_test)\n",
    "br_model.load_weights('br_pre-trained_glove_twitter_model.h5')\n",
    "br_model.evaluate(br_x_test, br_y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def predict(bi_pred, br_pred):\n",
    "    predictions = []\n",
    "    for i in range(len(bi_pred)):\n",
    "        if bi_pred[i][0]>=0.5 and br_pred[i][0]>=0.5:\n",
    "            if abs(bi_pred[i][0]-br_pred[i][0])<0.1:\n",
    "                predictions.append(\"BOTH\")\n",
    "            elif bi_pred[i][0]>br_pred[i][0]:\n",
    "                predictions.append(\"BITCHING\")\n",
    "            elif bi_pred[i][0]<br_pred[i][0]:\n",
    "                predictions.append(\"BRAGGING\")\n",
    "        elif bi_pred[i][0]>=0.5 and br_pred[i][0]<=0.5:\n",
    "            predictions.append(\"BITCHING\")\n",
    "        elif bi_pred[i][0]<=0.5 and br_pred[i][0]>=0.5:\n",
    "            predictions.append(\"BRAGGING\")\n",
    "        elif bi_pred[i][0]<=0.5 and br_pred[i][0]<=0.5:\n",
    "            predictions.append(\"NEITHER\")   \n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(BITCHING)  The decision on Sergeant Bergdahl is a complete and total disgrace to our Country and to our Military.\n",
      "(BITCHING)  Does anybody really believe that a reporter, who nobody ever heard of, went to his mailbox and found my tax returns? @NBCNews FAKE NEWS!\n",
      "(BRAGGING)  After Turkey call I will be heading over to Trump National Golf Club, Jupiter, to play golf (quickly) with Tiger Woods and Dustin Johnson. Then back to Mar-a-Lago for talks on bringing even more jobs and companies back to the USA!\n",
      "(BOTH)  “The President has accomplished some absolutely historic things during this past year.” Thank you Charlie Kirk of Turning Points USA. Sadly, the Fake Mainstream Media will NEVER talk about our accomplishments in their end of year reviews. We are compiling a long & beautiful list.\n"
     ]
    }
   ],
   "source": [
    "test_tweets = [\"The decision on Sergeant Bergdahl is a complete and total disgrace to our Country and to our Military.\",\n",
    "                \"Does anybody really believe that a reporter, who nobody ever heard of, went to his mailbox and found my tax returns? @NBCNews FAKE NEWS!\",\n",
    "                \"After Turkey call I will be heading over to Trump National Golf Club, Jupiter, to play golf (quickly) with Tiger Woods and Dustin Johnson. Then back to Mar-a-Lago for talks on bringing even more jobs and companies back to the USA!\",\n",
    "                \"“The President has accomplished some absolutely historic things during this past year.” Thank you Charlie Kirk of Turning Points USA. Sadly, the Fake Mainstream Media will NEVER talk about our accomplishments in their end of year reviews. We are compiling a long & beautiful list.\"]\n",
    "sequences = tokenizer.texts_to_sequences(test_tweets)\n",
    "x_test = pad_sequences(sequences, maxlen = maxlen)\n",
    "br_model.load_weights('br_pre-trained_glove_twitter_model.h5')\n",
    "br_pred=br_model.predict(x_test)\n",
    "bi_model.load_weights('bi_pre-trained_glove_twitter_model.h5')\n",
    "bi_pred=bi_model.predict(x_test)\n",
    "\n",
    "predictions=predict(bi_pred,br_pred)\n",
    "for i in range(len(test_tweets)):\n",
    "    print(\"(\"+predictions[i]+\")  \"+test_tweets[i])\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}