cnn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88\n",
      "  return f(*args, **kwds)\n",
      "/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88\n",
      "  return f(*args, **kwds)\n",
      "/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88\n",
      "  return f(*args, **kwds)\n",
      "/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88\n",
      "  return f(*args, **kwds)\n"
     ]
    }
   ],
   "source": [
    "import argparse\n",
    "import logging\n",
    "import os\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torchvision.models as models\n",
    "import torchvision.transforms as transforms\n",
    "\n",
    "import dill as pickle\n",
    "\n",
    "from torch.autograd import Variable\n",
    "from tqdm import tqdm, tqdm_notebook\n",
    "\n",
    "from sklearn.metrics import f1_score, recall_score, precision_score, roc_curve, roc_auc_score, auc\n",
    "\n",
    "import utils\n",
    "import models.net as net\n",
    "import models.data_loader as data_loader\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# This is a bit of magic to make matplotlib figures appear inline in the notebook\n",
    "# rather than in a new window.\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = (30.0, 16.0) # set default size of plots\n",
    "plt.rcParams['image.interpolation'] = 'nearest'\n",
    "plt.rcParams['image.cmap'] = 'gray'\n",
    "\n",
    "# Some more magic so that the notebook will reload external python modules;\n",
    "# see http://stackoverflow.com/questions/1907993/autoreload-of-modules-in-ipython\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Configure Project\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Directory containing the .txt metadata about dataset\n",
    "data_dir = 'data/4326'\n",
    "layer_name = 'VIIRS_SNPP_CorrectedReflectance_BandsM11-I2-I1'\n",
    "label_split = os.path.join(data_dir, layer_name + \".txt\")\n",
    "\n",
    "# Directory containing params.json\n",
    "model_dir = 'experiments/base_model'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training Function Definitions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def evaluate(model, loss_fn, dataloader, metrics, params):\n",
    "    \"\"\"Evaluate the model on `num_steps` batches.\n",
    "    Args:\n",
    "        model: (torch.nn.Module) the neural network\n",
    "        loss_fn: a function that takes batch_output and batch_labels and computes the loss for the batch\n",
    "        dataloader: (DataLoader) a torch.utils.data.DataLoader object that fetches data\n",
    "        metrics: (dict) a dictionary of functions that compute a metric using the output and labels of each batch\n",
    "        params: (Params) hyperparameters\n",
    "    \"\"\"\n",
    "\n",
    "    # set model to evaluation mode\n",
    "    model.eval()\n",
    "\n",
    "    # summary for current eval loop\n",
    "    summ = []\n",
    "    \n",
    "    X = None\n",
    "    outputs_prob = None\n",
    "    outputs_pred = None\n",
    "    labels = None\n",
    "\n",
    "    # compute metrics over the dataset\n",
    "    with tqdm_notebook(total=len(dataloader)) as t:\n",
    "        for data_batch, labels_batch in dataloader:\n",
    "            # move to GPU if available\n",
    "            if params.cuda:\n",
    "                data_batch, labels_batch = data_batch.cuda(async=True), labels_batch.cuda(async=True)\n",
    "            # fetch the next evaluation batch\n",
    "            data_batch, labels_batch = Variable(data_batch), Variable(labels_batch)\n",
    "\n",
    "            # compute model output\n",
    "            output_batch = model(data_batch)\n",
    "            loss = loss_fn(output_batch, labels_batch)\n",
    "\n",
    "            # extract data from torch Variable, move to cpu, convert to numpy arrays\n",
    "            output_batch = output_batch.data.cpu().numpy()\n",
    "            output_pred_batch = np.argmax(output_batch, axis=1)\n",
    "            labels_batch = labels_batch.data.cpu().numpy()\n",
    "\n",
    "            # save the outputs and labels\n",
    "            if X is None:\n",
    "                X = data_batch\n",
    "                outputs_prob = output_batch\n",
    "                outputs_pred = output_pred_batch\n",
    "                labels = labels_batch\n",
    "            else:\n",
    "                X = np.append(X, data_batch, axis=0)\n",
    "                outputs_prob = np.append(outputs_prob, output_batch, axis=0)\n",
    "                outputs_pred = np.append(outputs_pred, output_pred_batch, axis=0)\n",
    "                labels = np.append(labels, labels_batch, axis=0)\n",
    "\n",
    "            # compute all metrics on this batch\n",
    "            summary_batch = {metric: metrics[metric](output_batch, labels_batch)\n",
    "                             for metric in metrics}\n",
    "            summary_batch['loss'] = loss.data[0]\n",
    "            summ.append(summary_batch)\n",
    "\n",
    "    # compute mean of all metrics in summary\n",
    "    metrics_mean = {metric:np.mean([x[metric] for x in summ]) for metric in summ[0]} \n",
    "    metrics_string = \" ; \".join(\"{}: {:05.3f}\".format(k, v) for k, v in metrics_mean.items())\n",
    "    logging.info(\"- Eval metrics : \" + metrics_string)\n",
    "    \n",
    "    return X, outputs_prob, outputs_pred, labels, metrics_mean\n",
    "\n",
    "def train(model, optimizer, loss_fn, dataloader, metrics, params):\n",
    "    \"\"\"Train the model on `num_steps` batches\n",
    "    Args:\n",
    "        model: (torch.nn.Module) the neural network\n",
    "        optimizer: (torch.optim) optimizer for parameters of model\n",
    "        loss_fn: a function that takes batch_output and batch_labels and computes the loss for the batch\n",
    "        dataloader: (DataLoader) a torch.utils.data.DataLoader object that fetches training data\n",
    "        metrics: (dict) a dictionary of functions that compute a metric using the output and labels of each batch\n",
    "        params: (Params) hyperparameters\n",
    "    \"\"\"\n",
    "\n",
    "    # set model to training mode\n",
    "    model.train()\n",
    "\n",
    "    # summary for current training loop and a running average object for loss\n",
    "    summ = []\n",
    "    loss_avg = utils.RunningAverage()\n",
    "\n",
    "    # Use tqdm for progress bar\n",
    "    with tqdm_notebook(total=len(dataloader)) as t:\n",
    "        for i, (train_batch, labels_batch) in enumerate(dataloader):\n",
    "            # move to GPU if available\n",
    "            if params.cuda:\n",
    "                train_batch, labels_batch = train_batch.cuda(async=True), labels_batch.cuda(async=True)\n",
    "            # convert to torch Variables\n",
    "            train_batch, labels_batch = Variable(train_batch), Variable(labels_batch)\n",
    "\n",
    "            # compute model output and loss\n",
    "            output_batch = model(train_batch)\n",
    "            loss = loss_fn(output_batch, labels_batch)\n",
    "\n",
    "            # clear previous gradients, compute gradients of all variables wrt loss\n",
    "            optimizer.zero_grad()\n",
    "            loss.backward()\n",
    "\n",
    "            # performs updates using calculated gradients\n",
    "            optimizer.step()\n",
    "\n",
    "            # Evaluate summaries only once in a while\n",
    "            if i % params.save_summary_steps == 0:\n",
    "                # extract data from torch Variable, move to cpu, convert to numpy arrays\n",
    "                output_batch = output_batch.data.cpu().numpy()\n",
    "                labels_batch = labels_batch.data.cpu().numpy()\n",
    "\n",
    "                # compute all metrics on this batch\n",
    "                summary_batch = {metric:metrics[metric](output_batch, labels_batch)\n",
    "                                 for metric in metrics}\n",
    "                summary_batch['loss'] = loss.data[0]\n",
    "                summ.append(summary_batch)\n",
    "\n",
    "            # update the average loss\n",
    "            loss_avg.update(loss.data[0])\n",
    "\n",
    "            t.set_postfix(loss='{:05.3f}'.format(loss_avg()))\n",
    "            t.update()\n",
    "\n",
    "    # compute mean of all metrics in summary\n",
    "    metrics_mean = {metric:np.mean([x[metric] for x in summ]) for metric in summ[0]}\n",
    "    metrics_string = \" ; \".join(\"{}: {:05.3f}\".format(k, v) for k, v in metrics_mean.items())\n",
    "    logging.info(\"- Train metrics: \" + metrics_string)\n",
    "\n",
    "\n",
    "def train_and_evaluate(model, train_dataloader, val_dataloader, optimizer, loss_fn, metrics, params, model_dir,\n",
    "                       restore_file=None, map_location='cpu'):\n",
    "    \"\"\"Train the model and evaluate every epoch.\n",
    "    Args:\n",
    "        model: (torch.nn.Module) the neural network\n",
    "        train_dataloader: (DataLoader) a torch.utils.data.DataLoader object that fetches training data\n",
    "        val_dataloader: (DataLoader) a torch.utils.data.DataLoader object that fetches validation data\n",
    "        optimizer: (torch.optim) optimizer for parameters of model\n",
    "        loss_fn: a function that takes batch_output and batch_labels and computes the loss for the batch\n",
    "        metrics: (dict) a dictionary of functions that compute a metric using the output and labels of each batch\n",
    "        params: (Params) hyperparameters\n",
    "        model_dir: (string) directory containing config, weights and log\n",
    "        restore_file: (string) optional- name of file to restore from (without its extension .pth.tar)\n",
    "    \"\"\"\n",
    "    # reload weights from restore_file if specified\n",
    "    if restore_file is not None:\n",
    "        restore_path = os.path.join(model_dir, restore_file + '.pth.tar')\n",
    "        logging.info(\"Restoring parameters from {}\".format(restore_path))\n",
    "        utils.load_checkpoint(restore_path, model, optimizer, map_location=map_location)\n",
    "\n",
    "    best_val_score = 0.0\n",
    "\n",
    "    for epoch in range(params.num_epochs):\n",
    "        # Run one epoch\n",
    "        logging.info(\"Epoch {}/{}\".format(epoch + 1, params.num_epochs))\n",
    "\n",
    "        # compute number of batches in one epoch (one full pass over the training set)\n",
    "        train(model, optimizer, loss_fn, train_dataloader, metrics, params)\n",
    "\n",
    "        # Evaluate for one epoch on validation set\n",
    "        _, _, _, _, val_metrics = evaluate(model, loss_fn, val_dataloader, metrics, params)\n",
    "\n",
    "        val_score = val_metrics['AUC']\n",
    "        is_best = val_score>best_val_score\n",
    "\n",
    "        # Save weights\n",
    "        utils.save_checkpoint({'epoch': epoch + 1,\n",
    "                               'state_dict': model.state_dict(),\n",
    "                               'optim_dict' : optimizer.state_dict()},\n",
    "                               is_best=is_best,\n",
    "                               checkpoint=model_dir)\n",
    "\n",
    "        # If best_eval, best_save_path\n",
    "        if is_best:\n",
    "            logging.info(\"- Found new best accuracy\")\n",
    "            print(\"best_val_score {:05.3f}\".format(best_val_score))\n",
    "            print(\"val_score {:05.3f}\".format(val_score))\n",
    "            best_val_score = val_score\n",
    "\n",
    "            # Save best val metrics in a json file in the model directory\n",
    "            best_json_path = os.path.join(model_dir, \"metrics_val_best_weights.json\")\n",
    "            utils.save_dict_to_json(val_metrics, best_json_path)\n",
    "\n",
    "        # Save latest val metrics in a json file in the model directory\n",
    "        last_json_path = os.path.join(model_dir, \"metrics_val_last_weights.json\")\n",
    "        utils.save_dict_to_json(val_metrics, last_json_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load Hyperparameters from File"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GPU available: False\n"
     ]
    }
   ],
   "source": [
    "json_path = os.path.join(model_dir, 'params.json')\n",
    "assert os.path.isfile(json_path), \"No json configuration file found at {}\".format(json_path)\n",
    "params = utils.Params(json_path)\n",
    "\n",
    "# use GPU if available\n",
    "params.cuda = torch.cuda.is_available()\n",
    "print(\"GPU available: {}\".format(params.cuda))\n",
    "\n",
    "# Set the random seed for reproducible experiments\n",
    "torch.manual_seed(230)\n",
    "if params.cuda: torch.cuda.manual_seed(230)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Define the Model, Loss, and Optimizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define the model and optimizer\n",
    "model = net.Net(params, num_classes=2).cuda() if params.cuda else net.Net(params, num_classes=2)\n",
    "\n",
    "# Define the loss\n",
    "loss_fn = net.loss_fn\n",
    "\n",
    "# Optimize only the classifier\n",
    "optimizer = optim.Adam(model.parameters(), lr=params.learning_rate)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Define Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Fetch metrics\n",
    "def accuracy(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute the accuracy, given the outputs and labels for all images.\n",
    "\n",
    "    Args:\n",
    "        outputs: (np.ndarray) dimension batch_size x 2 - log softmax output of the model\n",
    "        labels: (np.ndarray) dimension batch_size, where each element is a value in [0, 1]\n",
    "\n",
    "    Returns: (float) accuracy in [0,1]\n",
    "    \"\"\"\n",
    "    outputs = np.argmax(outputs, axis=1)\n",
    "    return np.sum(outputs==labels)/float(labels.size)\n",
    "\n",
    "def recall(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute the recall, given the outputs and labels for all images.\n",
    "\n",
    "    Args:\n",
    "        outputs: (np.ndarray) dimension batch_size x 2 - log softmax output of the model\n",
    "        labels: (np.ndarray) dimension batch_size, where each element is a value in [0, 1]\n",
    "\n",
    "    Returns: (float) recall in [0,1]\n",
    "    \"\"\"\n",
    "    outputs = np.argmax(outputs, axis=1)\n",
    "    return recall_score(labels, outputs)\n",
    "\n",
    "def precision(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute the precision, given the outputs and labels for all images. \n",
    "    \n",
    "    The precision is the ratio tp / (tp + fp) where tp is the number \n",
    "    of true positives and fp the number of false positives. \n",
    "    \n",
    "    The precision is intuitively the ability of the classifier not to \n",
    "    label as positive a sample that is negative.\n",
    "    \n",
    "    The best value is 1 and the worst value is 0.\n",
    "    \n",
    "    Args:\n",
    "        outputs: (np.ndarray) dimension batch_size x 2 - log softmax output of the model\n",
    "        labels: (np.ndarray) dimension batch_size, where each element is a value in [0, 1]\n",
    "\n",
    "    Returns: (float) precision in [0,1]\n",
    "    \"\"\"\n",
    "    outputs = np.argmax(outputs, axis=1)\n",
    "    return precision_score(labels, outputs)\n",
    "\n",
    "def f1_metric(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute the F1 score, given the outputs and labels for all images.\n",
    "\n",
    "    Args:\n",
    "        outputs: (np.ndarray) dimension batch_size x 2 - log softmax output of the model\n",
    "        labels: (np.ndarray) dimension batch_size, where each element is a value in [0, 1]\n",
    "\n",
    "    Returns: (float) F1 score in [0,1]\n",
    "    \"\"\"\n",
    "    outputs = np.argmax(outputs, axis=1)\n",
    "    return f1_score(labels, outputs)\n",
    "\n",
    "def roc_graph(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute Receiver operating characteristic (ROC).\n",
    "    \n",
    "    Args:\n",
    "        outputs: (np.ndarray) dimension batch_size x 2 - log softmax output of the model\n",
    "        labels: (np.ndarray) dimension batch_size, where each element is a value in [0, 1]\n",
    "        \n",
    "    Returns: fpr, tpr, thresholds\n",
    "    \"\"\"\n",
    "    outputs = outputs[:,1]\n",
    "    return roc_curve(labels, outputs)\n",
    "\n",
    "def roc_auc_metric(outputs, labels):\n",
    "    \"\"\"\n",
    "    Compute Area Under the Receiver Operating Characteristic Curve (ROC AUC) from prediction scores.\n",
    "    \"\"\"\n",
    "    outputs = outputs[:,1]\n",
    "    return roc_auc_score(labels, outputs)\n",
    "\n",
    "# maintain all metrics required in this dictionary- these are used in the training and evaluation loops\n",
    "metrics = {\n",
    "    'AUC': roc_auc_metric, # area under receiver operator characteristic curve\n",
    "    'F1': f1_metric, # weighted average of precision and recall\n",
    "    'recall': recall, \n",
    "    'precision': precision,\n",
    "    'accuracy': accuracy,\n",
    "    # could add more metrics such as accuracy for each token type\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Run Experiment"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load the Datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading the datasets...\n",
      "- done.\n"
     ]
    }
   ],
   "source": [
    "# Set the logger\n",
    "utils.set_logger(os.path.join(model_dir, 'train.log'))\n",
    "\n",
    "# Create the input data pipeline\n",
    "logging.info(\"Loading the datasets...\")\n",
    "\n",
    "# Set the model input size\n",
    "IMG_DIM = (128, 256)\n",
    "IMG_PADDING = (0, 64, 0, 64) # left, top, right, bottom borders\n",
    "IMG_BINZARIZE = True # use black and white image\n",
    "\n",
    "# borrowed from http://pytorch.org/tutorials/advanced/neural_style_tutorial.html\n",
    "# and http://pytorch.org/tutorials/beginner/data_loading_tutorial.html\n",
    "# define a training image loader that specifies transforms on images. See documentation for more details.\n",
    "train_transformer = transforms.Compose([\n",
    "    transforms.Resize(IMG_DIM),  # resize the image \n",
    "    transforms.Pad(padding=IMG_PADDING, fill=0), # pad to be square!\n",
    "    transforms.Grayscale(),\n",
    "    transforms.RandomHorizontalFlip(),  # randomly flip image horizontally\n",
    "    transforms.RandomVerticalFlip(), # randomly flip image vertically\n",
    "    transforms.ToTensor(), # transform it into a torch tensor\n",
    "    # lambda x: x>0,\n",
    "    # lambda x: x.float(),\n",
    "    ])  \n",
    "\n",
    "# loader for evaluation, no data augmentation (e.g. horizontal flip)\n",
    "eval_transformer = transforms.Compose([\n",
    "    transforms.Resize(IMG_DIM),  # resize the image\n",
    "    transforms.Pad(padding=IMG_PADDING, fill=0), # pad to be square!\n",
    "    transforms.Grayscale(),\n",
    "    transforms.ToTensor(), # transform it into a torch tensor\n",
    "    # lambda x: x>0,\n",
    "    # lambda x: x.float(),\n",
    "    ])  \n",
    "\n",
    "params.img_binarize = IMG_BINZARIZE\n",
    "\n",
    "# fetch dataloaders\n",
    "train_dl = data_loader.fetch_dataloader('train', train_transformer, data_dir, layer_name, label_split, params)\n",
    "val_dl = data_loader.fetch_dataloader('val', eval_transformer, data_dir, layer_name, label_split, params)\n",
    "test_dl = data_loader.fetch_dataloader('test', eval_transformer, data_dir, layer_name, label_split, params)\n",
    "\n",
    "logging.info(\"- done.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train the Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Starting training for 100 epoch(s)\n",
      "Restoring parameters from experiments/base_model/last.pth.tar\n",
      "Epoch 1/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b36072f69db140fc9382bfeba09d5992",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel/__main__.py:112: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "/anaconda3/lib/python3.6/site-packages/ipykernel/__main__.py:116: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "- Train metrics: AUC: 0.761 ; F1: 0.414 ; recall: 0.750 ; precision: 0.286 ; accuracy: 0.734 ; loss: 0.717\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "90ddaefbc0e643fc8507eb03f4bcf28b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel/__main__.py:55: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "- Eval metrics : AUC: 0.606 ; F1: 0.168 ; recall: 1.000 ; precision: 0.094 ; accuracy: 0.094 ; loss: 5.944\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 2/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.000\n",
      "val_score 0.606\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8927dc702f6a4f0d8b9defc662217276",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.777 ; F1: 0.385 ; recall: 0.625 ; precision: 0.278 ; accuracy: 0.750 ; loss: 0.636\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "602551803c84486787f5208115850ac9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.627 ; F1: 0.210 ; recall: 0.889 ; precision: 0.123 ; accuracy: 0.413 ; loss: 1.337\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 3/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.606\n",
      "val_score 0.627\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9b9b1acbe850462196ce40b22f7ca6f3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.777 ; F1: 0.385 ; recall: 0.625 ; precision: 0.278 ; accuracy: 0.750 ; loss: 0.605\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7e0aa85c20fe4f1e9bca39d2b33ed275",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.645 ; F1: 0.252 ; recall: 0.433 ; precision: 0.181 ; accuracy: 0.697 ; loss: 0.640\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 4/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.627\n",
      "val_score 0.645\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "77af7a575eb54d989ba4126a8f7299fc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.827 ; F1: 0.606 ; recall: 0.909 ; precision: 0.455 ; accuracy: 0.797 ; loss: 0.581\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "49f2bf44c4d34a0f818a7be697a8d7ce",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.651 ; F1: 0.306 ; recall: 0.456 ; precision: 0.235 ; accuracy: 0.782 ; loss: 0.521\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 5/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.645\n",
      "val_score 0.651\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a3ecefcb75f94aeeb95eb5409c39e495",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.807 ; F1: 0.333 ; recall: 0.667 ; precision: 0.222 ; accuracy: 0.750 ; loss: 0.610\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "11e2b9434bf741b1a9c672bb78f484f6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.662 ; F1: 0.314 ; recall: 0.422 ; precision: 0.256 ; accuracy: 0.825 ; loss: 0.484\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 6/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.651\n",
      "val_score 0.662\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b9164328aae84246af587efc615d90f5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.923 ; F1: 0.483 ; recall: 0.778 ; precision: 0.350 ; accuracy: 0.766 ; loss: 0.508\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c7378f4bf3a74b0f8232dae275a17d7d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.688 ; F1: 0.314 ; recall: 0.422 ; precision: 0.256 ; accuracy: 0.825 ; loss: 0.485\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 7/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.662\n",
      "val_score 0.688\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "dd3e593a459044f6941f9e80cd8bf8a8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.875 ; F1: 0.300 ; recall: 0.750 ; precision: 0.188 ; accuracy: 0.781 ; loss: 0.556\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b5307dd18b4d4ce7a37c35995537e2b6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.696 ; F1: 0.274 ; recall: 0.356 ; precision: 0.236 ; accuracy: 0.836 ; loss: 0.475\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 8/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.688\n",
      "val_score 0.696\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b9694ffcb794d8080f745bb67010d34",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.929 ; F1: 0.588 ; recall: 0.625 ; precision: 0.556 ; accuracy: 0.891 ; loss: 0.451\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ddf73a6fcf7c498f90d40f04a0dc105d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.697 ; F1: 0.281 ; recall: 0.322 ; precision: 0.311 ; accuracy: 0.857 ; loss: 0.458\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 9/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.696\n",
      "val_score 0.697\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "42d503a6fe3c460690005600ad8ad085",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.794 ; F1: 0.583 ; recall: 0.778 ; precision: 0.467 ; accuracy: 0.844 ; loss: 0.516\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ada64333537349ba9aff568e82d32bff",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.695 ; F1: 0.300 ; recall: 0.322 ; precision: 0.371 ; accuracy: 0.873 ; loss: 0.443\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 10/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7b9eae9eab5e4f04b82ca1150efcd5cb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.837 ; F1: 0.286 ; recall: 0.750 ; precision: 0.176 ; accuracy: 0.766 ; loss: 0.531\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ad45e92aff3e49cfa96ece2c1aeee8f1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.695 ; F1: 0.375 ; recall: 0.389 ; precision: 0.411 ; accuracy: 0.889 ; loss: 0.436\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 11/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e33508b025bd41d9b9441b41d926dd44",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.986 ; F1: 0.476 ; recall: 1.000 ; precision: 0.312 ; accuracy: 0.828 ; loss: 0.476\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "177c61156f2842c09d15e038191d4e79",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.717 ; F1: 0.321 ; recall: 0.322 ; precision: 0.398 ; accuracy: 0.884 ; loss: 0.432\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 12/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.697\n",
      "val_score 0.717\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a4d88f3b5f66427598007a8d197bf489",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.672 ; F1: 0.308 ; recall: 0.571 ; precision: 0.211 ; accuracy: 0.719 ; loss: 0.620\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "94a6a6e249b7445387629e91d5646831",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.725 ; F1: 0.286 ; recall: 0.267 ; precision: 0.375 ; accuracy: 0.889 ; loss: 0.420\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 13/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.717\n",
      "val_score 0.725\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "476f1019166b40838457c98ab8f0ef1a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.842 ; F1: 0.500 ; recall: 0.556 ; precision: 0.455 ; accuracy: 0.844 ; loss: 0.448\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4de67d635f674f40b5afd03b77515c49",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.758 ; F1: 0.276 ; recall: 0.267 ; precision: 0.325 ; accuracy: 0.890 ; loss: 0.413\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 14/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.725\n",
      "val_score 0.758\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d8b6704e3d7148189ac19d121d9e10c5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.878 ; F1: 0.643 ; recall: 0.750 ; precision: 0.562 ; accuracy: 0.844 ; loss: 0.440\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bf166240f3104cfd8f5e2e7ee35766f8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.763 ; F1: 0.311 ; recall: 0.322 ; precision: 0.348 ; accuracy: 0.895 ; loss: 0.422\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 15/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.758\n",
      "val_score 0.763\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e26a86ae8779459d81b2098c2e3ff935",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.904 ; F1: 0.545 ; recall: 0.750 ; precision: 0.429 ; accuracy: 0.844 ; loss: 0.427\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bfc9e897eaed4c788b3ca15ad55b0cd9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.738 ; F1: 0.381 ; recall: 0.356 ; precision: 0.500 ; accuracy: 0.916 ; loss: 0.418\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 16/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2988a4344c124b348eb95bb03853197f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.992 ; F1: 0.818 ; recall: 1.000 ; precision: 0.692 ; accuracy: 0.938 ; loss: 0.387\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a46067cec03846e49ef808b51592e336",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.714 ; F1: 0.332 ; recall: 0.322 ; precision: 0.481 ; accuracy: 0.905 ; loss: 0.418\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 17/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1858b49ad0aa4aa79294a3bdc24d3d10",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.802 ; F1: 0.500 ; recall: 0.667 ; precision: 0.400 ; accuracy: 0.812 ; loss: 0.487\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "33d8a8eea12e4477b0d2fd1af90c9936",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.767 ; F1: 0.286 ; recall: 0.267 ; precision: 0.375 ; accuracy: 0.895 ; loss: 0.417\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 18/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.763\n",
      "val_score 0.767\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "72bca459169246c498c3cee89e9c1e39",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.805 ; F1: 0.526 ; recall: 0.714 ; precision: 0.417 ; accuracy: 0.859 ; loss: 0.493\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "6d8b5a2bfdf742db853b2b1750116182",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.765 ; F1: 0.286 ; recall: 0.267 ; precision: 0.375 ; accuracy: 0.900 ; loss: 0.417\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 19/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7975867fd693443d86dda9c61368f7d1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.739 ; F1: 0.240 ; recall: 0.600 ; precision: 0.150 ; accuracy: 0.703 ; loss: 0.646\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "27aa1b3e976c4d37af9b3903dac00539",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.765 ; F1: 0.297 ; recall: 0.267 ; precision: 0.458 ; accuracy: 0.905 ; loss: 0.413\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 20/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "65235cb537c946d3bcdde66fd3513d70",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.758 ; F1: 0.211 ; recall: 0.500 ; precision: 0.133 ; accuracy: 0.766 ; loss: 0.522\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d174adfa9c6f46d4b61fa117a2b555c3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.764 ; F1: 0.297 ; recall: 0.267 ; precision: 0.458 ; accuracy: 0.905 ; loss: 0.408\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 21/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1dc5ff69984041d18beb7553d27f45bc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.943 ; F1: 0.640 ; recall: 0.889 ; precision: 0.500 ; accuracy: 0.859 ; loss: 0.484\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a4330f46a2f245dfb438e128a1d6fa79",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1135: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1135: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "- Eval metrics : AUC: 0.764 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.916 ; loss: 0.400\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 22/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f0ece494212d4d0f834622e0e109e29e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.882 ; F1: 0.615 ; recall: 0.667 ; precision: 0.571 ; accuracy: 0.922 ; loss: 0.386\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ef5f64fb3a0449459b750e5de7a676c6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.694 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.911 ; loss: 0.403\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 23/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4ee6c15fc9344684990d7e9291bfea07",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.826 ; F1: 0.545 ; recall: 0.667 ; precision: 0.462 ; accuracy: 0.844 ; loss: 0.453\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b55045babcd345109b122a09910ce449",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.743 ; F1: 0.297 ; recall: 0.267 ; precision: 0.458 ; accuracy: 0.905 ; loss: 0.407\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 24/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e17e35650ae547f8be12d4763def3fd3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.947 ; F1: 0.706 ; recall: 0.667 ; precision: 0.750 ; accuracy: 0.922 ; loss: 0.352\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "fc990c19d3e14b009a31c1b708c2c101",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.737 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.911 ; loss: 0.407\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 25/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9b093e98dc74497798ffb06f431804de",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.898 ; F1: 0.571 ; recall: 0.400 ; precision: 1.000 ; accuracy: 0.953 ; loss: 0.396\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4ebd0a39503248ec873f930aa16c3676",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.737 ; F1: 0.278 ; recall: 0.233 ; precision: 0.500 ; accuracy: 0.911 ; loss: 0.399\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 26/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5cc7a20d197f44a3b9a5ba725f576d04",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.900 ; F1: 0.526 ; recall: 0.625 ; precision: 0.455 ; accuracy: 0.859 ; loss: 0.462\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "68dd63be44f4465590536bdc0520d923",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.728 ; F1: 0.321 ; recall: 0.267 ; precision: 0.500 ; accuracy: 0.916 ; loss: 0.399\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 27/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9f665bdfd0474fbeb15e46a476d8b77d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.848 ; F1: 0.750 ; recall: 0.667 ; precision: 0.857 ; accuracy: 0.938 ; loss: 0.413\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4f35929e0ce64b8ba34228cdab41ea64",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.745 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.916 ; loss: 0.402\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 28/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "21a4cf8767464276883eeee24079f1d8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.972 ; F1: 0.667 ; recall: 0.889 ; precision: 0.533 ; accuracy: 0.875 ; loss: 0.439\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0ff2b98dd59b482298638768d1c0f3e3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.744 ; F1: 0.278 ; recall: 0.233 ; precision: 0.500 ; accuracy: 0.916 ; loss: 0.394\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 29/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "825cecfb5f264cecbade987c719092fc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.911 ; F1: 0.762 ; recall: 0.727 ; precision: 0.800 ; accuracy: 0.922 ; loss: 0.390\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ecbcd84d46994616a27b2da02ab9ac77",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.771 ; F1: 0.278 ; recall: 0.233 ; precision: 0.500 ; accuracy: 0.916 ; loss: 0.396\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 30/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.767\n",
      "val_score 0.771\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e754e823a2ac49b48da6dd8ea90f0f9d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.931 ; F1: 0.800 ; recall: 0.667 ; precision: 1.000 ; accuracy: 0.922 ; loss: 0.349\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4818f3459efb493282b05ae889780449",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.787 ; F1: 0.321 ; recall: 0.267 ; precision: 0.500 ; accuracy: 0.921 ; loss: 0.398\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 31/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.771\n",
      "val_score 0.787\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d5e44db12824452f989da8e0d6db4e31",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.864 ; F1: 0.700 ; recall: 0.538 ; precision: 1.000 ; accuracy: 0.906 ; loss: 0.391\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7beb8691d89d4e5194a9b987e58d84e6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.763 ; F1: 0.357 ; recall: 0.300 ; precision: 0.500 ; accuracy: 0.927 ; loss: 0.400\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 32/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3b76ae580cdf47a4b0f84eaac3098f80",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 1.000 ; F1: 0.667 ; recall: 1.000 ; precision: 0.500 ; accuracy: 0.922 ; loss: 0.366\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ceb4e8b313d3448a8eaeecf77f7c7d5e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.763 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.911 ; loss: 0.405\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 33/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a5ee8271e9394531bf0aa405c7ececf9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.989 ; F1: 0.667 ; recall: 1.000 ; precision: 0.500 ; accuracy: 0.906 ; loss: 0.392\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f8d927ee326d465cbbc38cae0a1ab45d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.745 ; F1: 0.357 ; recall: 0.300 ; precision: 0.500 ; accuracy: 0.921 ; loss: 0.402\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 34/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "63aa60169be145818d605481ce8dfa0d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.897 ; F1: 0.545 ; recall: 0.750 ; precision: 0.429 ; accuracy: 0.844 ; loss: 0.438\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "6a51c2e1cde044188d550af2c56d9b9a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.757 ; F1: 0.321 ; recall: 0.267 ; precision: 0.500 ; accuracy: 0.921 ; loss: 0.395\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 35/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "acefab76aa5e4d7fa71ed27f3674eeec",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.879 ; F1: 0.632 ; recall: 0.667 ; precision: 0.600 ; accuracy: 0.891 ; loss: 0.401\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e5bbedbf01234bb3bb8e204a6d1f69e9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.756 ; F1: 0.321 ; recall: 0.267 ; precision: 0.500 ; accuracy: 0.921 ; loss: 0.397\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 36/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a87500fb66154acb9b6384c0a6083764",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.807 ; F1: 0.571 ; recall: 0.667 ; precision: 0.500 ; accuracy: 0.906 ; loss: 0.389\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "fed6e826e1944fa4b8ec69deb8db6f5b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.800 ; F1: 0.321 ; recall: 0.267 ; precision: 0.500 ; accuracy: 0.921 ; loss: 0.397\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 37/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.787\n",
      "val_score 0.800\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "544eab8bb98943bc9277965ec677788b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.920 ; F1: 0.833 ; recall: 0.714 ; precision: 1.000 ; accuracy: 0.969 ; loss: 0.354\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c21c46427f954a85b3bee846d3b8a25f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Eval metrics : AUC: 0.810 ; F1: 0.308 ; recall: 0.267 ; precision: 0.476 ; accuracy: 0.916 ; loss: 0.397\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Found new best accuracy\n",
      "Epoch 38/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_val_score 0.800\n",
      "val_score 0.810\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3d42feece28e4344acaf27c876278230",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "- Train metrics: AUC: 0.916 ; F1: 0.706 ; recall: 0.545 ; precision: 1.000 ; accuracy: 0.922 ; loss: 0.349\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a3f2c6a57ce744dc847fafb266bd17f3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Process Process-302:\n",
      "Process Process-304:\n",
      "Process Process-303:\n",
      "Process Process-301:\n",
      "Traceback (most recent call last):\n",
      "Traceback (most recent call last):\n",
      "Traceback (most recent call last):\n",
      "Traceback (most recent call last):\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 258, in _bootstrap\n",
      "    self.run()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 258, in _bootstrap\n",
      "    self.run()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 258, in _bootstrap\n",
      "    self.run()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 258, in _bootstrap\n",
      "    self.run()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 93, in run\n",
      "    self._target(*self._args, **self._kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 93, in run\n",
      "    self._target(*self._args, **self._kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 93, in run\n",
      "    self._target(*self._args, **self._kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 93, in run\n",
      "    self._target(*self._args, **self._kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 52, in _worker_loop\n",
      "    r = index_queue.get()\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 52, in _worker_loop\n",
      "    r = index_queue.get()\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 52, in _worker_loop\n",
      "    r = index_queue.get()\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 52, in _worker_loop\n",
      "    r = index_queue.get()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/queues.py\", line 335, in get\n",
      "    res = self._reader.recv_bytes()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/queues.py\", line 335, in get\n",
      "    res = self._reader.recv_bytes()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/queues.py\", line 335, in get\n",
      "    res = self._reader.recv_bytes()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/queues.py\", line 335, in get\n",
      "    res = self._reader.recv_bytes()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 216, in recv_bytes\n",
      "    buf = self._recv_bytes(maxlength)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 216, in recv_bytes\n",
      "    buf = self._recv_bytes(maxlength)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 216, in recv_bytes\n",
      "    buf = self._recv_bytes(maxlength)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 216, in recv_bytes\n",
      "    buf = self._recv_bytes(maxlength)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 407, in _recv_bytes\n",
      "    buf = self._recv(4)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 407, in _recv_bytes\n",
      "    buf = self._recv(4)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 407, in _recv_bytes\n",
      "    buf = self._recv(4)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 379, in _recv\n",
      "    chunk = read(handle, remaining)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 407, in _recv_bytes\n",
      "    buf = self._recv(4)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 379, in _recv\n",
      "    chunk = read(handle, remaining)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 379, in _recv\n",
      "    chunk = read(handle, remaining)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 379, in _recv\n",
      "    chunk = read(handle, remaining)\n",
      "KeyboardInterrupt\n",
      "KeyboardInterrupt\n",
      "KeyboardInterrupt\n",
      "KeyboardInterrupt\n",
      "- Eval metrics : AUC: 0.809 ; F1: 0.336 ; recall: 0.267 ; precision: 0.533 ; accuracy: 0.927 ; loss: 0.383\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 39/100\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "67e4389608fb482e88802d450e61aa95",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=9), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Process Process-308:\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Traceback (most recent call last):\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 258, in _bootstrap\n",
      "    self.run()\n",
      "Internal Python error in the inspect module.\n",
      "Below is the traceback from this internal error.\n",
      "\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/process.py\", line 93, in run\n",
      "    self._target(*self._args, **self._kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 57, in _worker_loop\n",
      "    samples = collate_fn([dataset[i] for i in batch_indices])\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 57, in <listcomp>\n",
      "    samples = collate_fn([dataset[i] for i in batch_indices])\n",
      "  File \"/Users/davidxue/projects/gibs_ml/models/data_loader.py\", line 102, in __getitem__\n",
      "    image = self.transform(image)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torchvision-0.2.1-py3.6.egg/torchvision/transforms/transforms.py\", line 49, in __call__\n",
      "    img = t(img)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torchvision-0.2.1-py3.6.egg/torchvision/transforms/transforms.py\", line 175, in __call__\n",
      "    return F.resize(img, self.size, self.interpolation)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torchvision-0.2.1-py3.6.egg/torchvision/transforms/functional.py\", line 206, in resize\n",
      "    return img.resize(size[::-1], interpolation)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/PIL/Image.py\", line 1749, in resize\n",
      "    return self._new(self.im.resize(size, resample, box))\n",
      "KeyboardInterrupt\n",
      "\n",
      "Unfortunately, your original traceback can not be constructed.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Traceback (most recent call last):\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/IPython/core/interactiveshell.py\", line 2963, in run_code\n",
      "    exec(code_obj, self.user_global_ns, self.user_ns)\n",
      "  File \"<ipython-input-8-6d0192bd7505>\", line 8, in <module>\n",
      "    train_and_evaluate(model, train_dl, val_dl, optimizer, loss_fn, metrics, params, model_dir, restore_file, map_location)\n",
      "  File \"<ipython-input-3-7aa378bc2faf>\", line 154, in train_and_evaluate\n",
      "    train(model, optimizer, loss_fn, train_dataloader, metrics, params)\n",
      "  File \"<ipython-input-3-7aa378bc2faf>\", line 85, in train\n",
      "    for i, (train_batch, labels_batch) in enumerate(dataloader):\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 280, in __next__\n",
      "    idx, batch = self._get_batch()\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 259, in _get_batch\n",
      "    return self.data_queue.get()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/queues.py\", line 335, in get\n",
      "    res = self._reader.recv_bytes()\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 216, in recv_bytes\n",
      "    buf = self._recv_bytes(maxlength)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 407, in _recv_bytes\n",
      "    buf = self._recv(4)\n",
      "  File \"/anaconda3/lib/python3.6/multiprocessing/connection.py\", line 379, in _recv\n",
      "    chunk = read(handle, remaining)\n",
      "KeyboardInterrupt\n",
      "\n",
      "During handling of the above exception, another exception occurred:\n",
      "\n",
      "Traceback (most recent call last):\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/IPython/core/interactiveshell.py\", line 1863, in showtraceback\n",
      "    stb = value._render_traceback_()\n",
      "AttributeError: 'KeyboardInterrupt' object has no attribute '_render_traceback_'\n",
      "\n",
      "During handling of the above exception, another exception occurred:\n",
      "\n",
      "Traceback (most recent call last):\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/IPython/core/ultratb.py\", line 1095, in get_records\n",
      "    return _fixed_getinnerframes(etb, number_of_lines_of_context, tb_offset)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/IPython/core/ultratb.py\", line 311, in wrapped\n",
      "    return f(*args, **kwargs)\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/IPython/core/ultratb.py\", line 345, in _fixed_getinnerframes\n",
      "    records = fix_frame_records_filenames(inspect.getinnerframes(etb, context))\n",
      "  File \"/anaconda3/lib/python3.6/inspect.py\", line 1483, in getinnerframes\n",
      "    frameinfo = (tb.tb_frame,) + getframeinfo(tb, context)\n",
      "  File \"/anaconda3/lib/python3.6/inspect.py\", line 1441, in getframeinfo\n",
      "    filename = getsourcefile(frame) or getfile(frame)\n",
      "  File \"/anaconda3/lib/python3.6/inspect.py\", line 696, in getsourcefile\n",
      "    if getattr(getmodule(object, filename), '__loader__', None) is not None:\n",
      "  File \"/anaconda3/lib/python3.6/inspect.py\", line 732, in getmodule\n",
      "    for modname, module in list(sys.modules.items()):\n",
      "  File \"/anaconda3/lib/python3.6/site-packages/torch/utils/data/dataloader.py\", line 178, in handler\n",
      "    _error_if_any_worker_fails()\n",
      "RuntimeError: DataLoader worker (pid 18580) exited unexpectedly with exit code 1.\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m"
     ]
    }
   ],
   "source": [
    "# Reload weights from the saved file\n",
    "# Name of the file in --model_dir containing weights to reload before training\n",
    "restore_file = 'last' # 'best' or 'last' or None\n",
    "\n",
    "# Train the model\n",
    "logging.info(\"Starting training for {} epoch(s)\".format(params.num_epochs))\n",
    "map_location = None if params.cuda else 'cpu'\n",
    "train_and_evaluate(model, train_dl, val_dl, optimizer, loss_fn, metrics, params, model_dir, restore_file, map_location)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate Model on Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Starting evaluation\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bbfd3c4ca9d44a78afd2d831dc617602",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel/__main__.py:55: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number\n",
      "- Eval metrics : AUC: 0.823 ; F1: 0.605 ; recall: 0.528 ; precision: 0.722 ; accuracy: 0.937 ; loss: 0.394\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Get the logger\n",
    "utils.set_logger(os.path.join(model_dir, 'evaluate.log'))\n",
    "\n",
    "logging.info(\"Starting evaluation\")\n",
    "\n",
    "# Reload weights from the saved file\n",
    "# Name of the file in --model_dir containing weights to reload before evaluation\n",
    "restore_eval_file = 'best' # 'best' or 'last' or None\n",
    "map_location = None if params.cuda else 'cpu'\n",
    "utils.load_checkpoint(os.path.join(model_dir, restore_eval_file + '.pth.tar'), model, map_location=map_location)\n",
    "\n",
    "# Evaluate\n",
    "X_test, y_test_prob, y_test_pred, y_test, test_metrics = evaluate(model, loss_fn, test_dl, metrics, params)\n",
    "save_path = os.path.join(model_dir, \"metrics_test_{}.json\".format(restore_eval_file))\n",
    "utils.save_dict_to_json(test_metrics, save_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ROC curve (area = 0.817)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "test_fpr, test_tpr, test_thresholds = roc_graph(y_test_prob, y_test)\n",
    "test_roc_auc = roc_auc_metric(y_test_prob, y_test)\n",
    "print('ROC curve (area = %0.3f)' % test_roc_auc)\n",
    "# test_roc_auc = test_metrics['AUC']\n",
    "\n",
    "plt.figure()\n",
    "lw = 2\n",
    "plt.plot(test_fpr, test_tpr, color='darkorange', lw=lw, label='ROC curve (area = %0.2f)' % test_roc_auc)\n",
    "plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic for test set')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 93.6842105263158%\n",
      "Non-normalized conf_matrix:\n",
      "[[169.   9.]\n",
      " [  3.   9.]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x1152 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def confusion_matrix(results, categories, normalizeMatrix = True, plotConfusion=False):\n",
    "    \"\"\" Generate confusion matrix\n",
    "    - results = list of tuples of (correct label, predicted label)\n",
    "    - categories = list of category names\n",
    "    Returns confusion matrix; rows are correct labels and columns are predictions\n",
    "    \"\"\"\n",
    "    # Empty confusion matrix\n",
    "    matrix = np.zeros((len(categories),len(categories)))\n",
    "\n",
    "    # Iterate over all labels and populate matrix\n",
    "    for label, pred in results:\n",
    "        matrix[label, pred] += 1\n",
    "\n",
    "    # Print matrix and percent accuracy\n",
    "    accuracy = float(np.trace(matrix)) * 100.0 / len(results)\n",
    "    print('Accuracy: ' +  str(accuracy) + '%')\n",
    "\n",
    "    print(\"Non-normalized conf_matrix:\")\n",
    "    print(matrix)\n",
    "    \n",
    "    # Normalize confusion matrix\n",
    "    if normalizeMatrix:\n",
    "      s = np.sum(matrix,1) # Sum each row\n",
    "      for i in range(matrix.shape[0]):\n",
    "        # Normalization handles class imbalance in training set\n",
    "        matrix[i,:] /= s[i]\n",
    "\n",
    "    # Save matrix to file:\n",
    "    # np.save(\"confusion_matrix.npy\",matrix)\n",
    "\n",
    "    # Plot the confusion matrix\n",
    "    if plotConfusion:\n",
    "        plot_confusion_matrix(matrix, categories)\n",
    "\n",
    "def plot_confusion_matrix(cm, names=None, title='Confusion Matrix', cmap=plt.cm.Blues):\n",
    "    plt.figure(4)\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "\n",
    "    # Add labels to confusion matrix:\n",
    "    if names is None:\n",
    "        names = range(cm.shape[0])\n",
    "\n",
    "    tick_marks = np.arange(len(names))\n",
    "    plt.xticks(tick_marks, names, rotation=45)\n",
    "    plt.yticks(tick_marks, names)\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('Correct label')\n",
    "    plt.xlabel('Predicted label')\n",
    "    plt.show()\n",
    "\n",
    "    \n",
    "plt.rcParams['figure.figsize'] = (30.0, 16.0) # set default size of plots\n",
    "\n",
    "classes = ['normal', 'abnormal']\n",
    "confusion_matrix(list(zip(y_test_pred, y_test)), classes, plotConfusion=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x1152 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.rcParams['figure.figsize'] = (30.0, 16.0) # set default size of plots\n",
    "\n",
    "# Visualize misclassified examples from the validation dataset.\n",
    "num_classes = len(classes)\n",
    "max_examples = 5\n",
    "\n",
    "incorrect_idx = [idx for idx, incorrect in enumerate(np.array(y_test) != np.array(y_test_pred)) if incorrect]\n",
    "if len(incorrect_idx) > max_examples:\n",
    "    incorrect_idx = incorrect_idx[:max_examples]\n",
    "num_incorrect = len(incorrect_idx)\n",
    "\n",
    "for y, idx in enumerate(incorrect_idx):\n",
    "    for i in range(3):\n",
    "        plt_idx = i * num_incorrect + y + 1\n",
    "        if i == 0:\n",
    "            plt.subplot(3, num_incorrect, plt_idx)\n",
    "            plt.imshow(np.squeeze(X_test[idx]).astype('uint8'))\n",
    "            plt.axis('off')\n",
    "            plt.title(\"Prediction: %s\\n Truth: %s\" % (classes[y_test_pred[idx]], classes[y_test[idx]]))\n",
    "        elif i == 1:\n",
    "            plt.subplot(3, num_incorrect, plt_idx)\n",
    "            log_probabilities = y_test_prob[idx]\n",
    "            probabilities = np.exp(log_probabilities)\n",
    "            y_pos = np.arange(len(classes))\n",
    "            plt.bar(y_pos, probabilities)\n",
    "            plt.xticks(y_pos, classes, rotation=45)\n",
    "            plt.ylabel(\"Probability\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# num_classes = len(classes)\n",
    "# img_size = (16, 8)\n",
    "# num_channels = 1\n",
    "\n",
    "# # Visualize the learned weights for each class.\n",
    "# # Depending on your choice of learning rate and regularization strength, these may\n",
    "# # or may not be nice to look at.\n",
    "# w = model.conv3.weight.data.numpy() # convert to numpy arrays\n",
    "# print(w.shape)\n",
    "# w = w.reshape(num_classes, img_size[1], img_size[0], num_channels)\n",
    "# w_min, w_max = np.min(w), np.max(w)\n",
    "# for i in range(num_classes):\n",
    "#     plt.subplot(2, 5, i + 1)\n",
    "      \n",
    "#     # Rescale the weights to be between 0 and 255\n",
    "#     wimg = 255.0 * (w[i, :, :, :].squeeze() - w_min) / (w_max - w_min)\n",
    "#     plt.imshow(wimg.astype('uint8'))\n",
    "#     plt.axis('off')\n",
    "#     plt.title(classes[i])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test on New Dataset!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Directory containing the .txt metadata about dataset\n",
    "data_dir = 'data/4326'\n",
    "layer_name = 'VIIRS_SNPP_CorrectedReflectance_TrueColor'\n",
    "new_test_imgs = os.path.join(data_dir, \"NEW_\" + layer_name + \".txt\")\n",
    "\n",
    "# Directory containing params.json\n",
    "model_dir = 'experiments/base_model'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set the logger\n",
    "utils.set_logger(os.path.join(model_dir, 'train.log'))\n",
    "\n",
    "# Create the input data pipeline\n",
    "logging.info(\"Loading the datasets...\")\n",
    "\n",
    "# Set the model input size\n",
    "IMG_DIM = (128, 256)\n",
    "IMG_PADDING = (0, 64, 0, 64) # left, top, right, bottom borders\n",
    "IMG_BINZARIZE = True # use black and white image\n",
    "\n",
    "# loader for evaluation, no data augmentation (e.g. horizontal flip)\n",
    "eval_transformer = transforms.Compose([\n",
    "    transforms.Resize(IMG_DIM),  # resize the image\n",
    "    transforms.Pad(padding=IMG_PADDING, fill=0), # pad to be square!\n",
    "    transforms.Grayscale(),\n",
    "    transforms.ToTensor(), # transform it into a torch tensor\n",
    "    # lambda x: x>0,\n",
    "    # lambda x: x.float(),\n",
    "    ])  \n",
    "\n",
    "params.img_binarize = IMG_BINZARIZE\n",
    "\n",
    "# fetch dataloaders\n",
    "new_test_dl = data_loader.fetch_dataloader('new_test', eval_transformer, data_dir, layer_name, label_split, params)\n",
    "\n",
    "logging.info(\"- done.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get the logger\n",
    "utils.set_logger(os.path.join(model_dir, 'evaluate.log'))\n",
    "\n",
    "logging.info(\"Starting evaluation\")\n",
    "\n",
    "# Reload weights from the saved file\n",
    "# Name of the file in --model_dir containing weights to reload before evaluation\n",
    "restore_eval_file = 'last' # 'best' or 'last' or None\n",
    "map_location = None if params.cuda else 'cpu'\n",
    "utils.load_checkpoint(os.path.join(model_dir, restore_eval_file + '.pth.tar'), model, map_location=map_location)\n",
    "\n",
    "# Evaluate\n",
    "X_test, y_test_prob, y_test_pred, y_test, test_metrics = evaluate(model, loss_fn, new_test_dl, metrics, params)\n",
    "save_path = os.path.join(model_dir, \"metrics_test_{}.json\".format(restore_eval_file))\n",
    "utils.save_dict_to_json(test_metrics, save_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [default]",
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}