convnet_2.0.0.35b.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import os\n",
    "import wget\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow as tf\n",
    "from mammo_utils import flatten\n",
    "from training_utils import download_file, get_batches, read_and_decode_single_example, load_validation_data, \\\n",
    "    download_data, evaluate_model, get_training_data, load_weights, flatten, _scale_input_data, augment, _conv2d_batch_norm\n",
    "from tensorboard import summary as summary_lib"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# arguments\n",
    "epochs = 40\n",
    "dataset = 10\n",
    "init_model = None\n",
    "how = \"normal\"\n",
    "action = \"train\"\n",
    "threshold = 0.5\n",
    "contrast = 1.1\n",
    "distort = False\n",
    "weight = 8.0\n",
    "freeze = False\n",
    "stop = False\n",
    "version = \"\"\n",
    "# mean to use when centering data\n",
    "mu = 127.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# download the data\n",
    "batch_size = 32\n",
    "train_files, total_records = get_training_data(what=dataset)\n",
    "\n",
    "if how == \"label\":\n",
    "    model_label = \"l\"\n",
    "elif how == \"normal\":\n",
    "    model_label = \"b\"\n",
    "else:\n",
    "    model_label = \"x\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Steps per epoch: 1746\n",
      "Number of classes: 2\n"
     ]
    }
   ],
   "source": [
    "## Hyperparameters\n",
    "epsilon = 1e-8\n",
    "\n",
    "# learning rate\n",
    "epochs_per_decay = 5\n",
    "starting_rate = 0.001\n",
    "decay_factor = 0.80\n",
    "staircase = True\n",
    "\n",
    "# learning rate decay variables\n",
    "steps_per_epoch = int(total_records / batch_size)\n",
    "print(\"Steps per epoch:\", steps_per_epoch)\n",
    "\n",
    "# lambdas\n",
    "lamC = 0.00001\n",
    "lamF = 0.00250\n",
    "\n",
    "# use dropout\n",
    "dropout = True\n",
    "fcdropout_rate = 0.5\n",
    "convdropout_rate = 0.001\n",
    "pooldropout_rate = 0.1\n",
    "\n",
    "if how == \"label\":\n",
    "    num_classes = 5\n",
    "elif how == \"normal\":\n",
    "    num_classes = 2\n",
    "elif how == \"mass\":\n",
    "    num_classes = 3\n",
    "elif how == \"benign\":\n",
    "    num_classes = 3\n",
    "\n",
    "print(\"Number of classes:\", num_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Graph created...\n"
     ]
    }
   ],
   "source": [
    "## Build the graph\n",
    "graph = tf.Graph()\n",
    "\n",
    "model_name = \"model_s2.0.0.35\" + model_label + \".\" + str(dataset) + str(version)\n",
    "## Change Log\n",
    "# 0.0.0.4 - increase pool3 to 3x3 with stride 3\n",
    "# 0.0.0.6 - reduce pool 3 stride back to 2\n",
    "# 0.0.0.7 - reduce lambda for l2 reg\n",
    "# 0.0.0.8 - increase conv1 to 7x7 stride 2\n",
    "# 0.0.0.9 - disable per image normalization\n",
    "# 0.0.0.10 - commented out batch norm in conv layers, added conv4 and changed stride of convs to 1, increased FC lambda\n",
    "# 0.0.0.11 - turn dropout for conv layers on\n",
    "# 0.0.0.12 - added batch norm after pooling layers, increase pool dropout, decrease conv dropout, added extra conv layer to reduce data dimensionality\n",
    "# 0.0.0.13 - added precision and f1 summaries\n",
    "# 0.0.0.14 - fixing batch normalization, I don't think it's going to work after each pool\n",
    "# 0.0.0.15 - reduced xentropy weighting term\n",
    "# 0.0.0.17 - replaced initial 5x5 conv layers with 3 3x3 layers\n",
    "# 0.0.0.18 - changed stride of first conv to 2 from 1\n",
    "# 0.0.0.19 - doubled units in two fc layers\n",
    "# 0.0.0.20 - lowered learning rate, put a batch norm back in\n",
    "# 0.0.0.21 - put all batch norms back in\n",
    "# 0.0.0.22 - increased lambdaC, removed dropout from conv layers\n",
    "# 1.0.0.23 - added extra conv layers\n",
    "# 1.0.0.27 - updates to training code and metrics\n",
    "# 1.0.0.28 - using weighted x-entropy to improve recall\n",
    "# 1.0.0.29 - updated code to work training to classify for multiple classes\n",
    "# 1.0.0.29f - putting weighted x-entropy back\n",
    "# 1.0.0.30b - changed some hyperparameters\n",
    "# 1.0.0.31l - added decision threshold to predictions\n",
    "# 1.0.0.33 - scaling input data\n",
    "# 1.0.0.34 - centering data by 127, not by mean\n",
    "# 1.0.0.35 - not centering data, just scaling it\n",
    "# 2.0.0.35 - turning into fcn\n",
    "\n",
    "with graph.as_default():\n",
    "    training = tf.placeholder(dtype=tf.bool, name=\"is_training\")\n",
    "    is_testing = tf.placeholder(dtype=bool, shape=(), name=\"is_testing\")\n",
    "\n",
    "    # create global step for decaying learning rate\n",
    "    global_step = tf.Variable(0, trainable=False)\n",
    "\n",
    "    learning_rate = tf.train.exponential_decay(starting_rate,\n",
    "                                               global_step,\n",
    "                                               steps_per_epoch * epochs_per_decay,\n",
    "                                               decay_factor,\n",
    "                                               staircase=staircase)\n",
    "\n",
    "    with tf.name_scope('inputs') as scope:\n",
    "        image, label = read_and_decode_single_example(train_files, label_type=how, normalize=False, distort=False)\n",
    "\n",
    "        X_def, y_def = tf.train.shuffle_batch([image, label], batch_size=batch_size, capacity=2000,\n",
    "                                              seed=None,\n",
    "                                              min_after_dequeue=1000)\n",
    "\n",
    "        # Placeholders\n",
    "        X = tf.placeholder_with_default(X_def, shape=[None, None, None, 1])\n",
    "        y = tf.placeholder_with_default(y_def, shape=[None])\n",
    "\n",
    "        X_adj = tf.cast(X, dtype=tf.float32)\n",
    "        X_adj = _scale_input_data(X_adj, contrast=contrast, mu=mu, scale=255.0)\n",
    "\n",
    "        # data augmentation\n",
    "        if distort:\n",
    "            X_adj, y = augment(X_adj, y, horizontal_flip=True, vertical_flip=True, mixup=0)\n",
    "\n",
    "    # Convolutional layer 1\n",
    "    with tf.name_scope('conv1') as scope:\n",
    "        conv1 = tf.layers.conv2d(\n",
    "            X_adj,  # Input data\n",
    "            filters=32,  # 32 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(2, 2),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=100),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv1'\n",
    "        )\n",
    "\n",
    "        conv1 = tf.layers.batch_normalization(\n",
    "            conv1,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn1'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv1_bn_relu = tf.nn.relu(conv1, name='relu1')\n",
    "\n",
    "    with tf.name_scope('conv1.1') as scope:\n",
    "        conv11 = tf.layers.conv2d(\n",
    "            conv1_bn_relu,  # Input data\n",
    "            filters=32,  # 32 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(1, 1),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=101),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv1.1'\n",
    "        )\n",
    "\n",
    "        conv11 = tf.layers.batch_normalization(\n",
    "            conv11,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn1.1'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv11 = tf.nn.relu(conv11, name='relu1.1')\n",
    "\n",
    "\n",
    "    with tf.name_scope('conv1.2') as scope:\n",
    "        conv12 = tf.layers.conv2d(\n",
    "            conv11,  # Input data\n",
    "            filters=32,  # 32 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(1, 1),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=1101),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv1.2'\n",
    "        )\n",
    "\n",
    "        conv12 = tf.layers.batch_normalization(\n",
    "            conv12,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn1.2'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv12 = tf.nn.relu(conv12, name='relu1.1')\n",
    "\n",
    "    # Max pooling layer 1\n",
    "    with tf.name_scope('pool1') as scope:\n",
    "        pool1 = tf.layers.max_pooling2d(\n",
    "            conv12,  # Input\n",
    "            pool_size=(3, 3),  # Pool size: 3x3\n",
    "            strides=(2, 2),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            name='pool1'\n",
    "        )\n",
    "\n",
    "        # optional dropout\n",
    "        if dropout:\n",
    "            pool1 = tf.layers.dropout(pool1, rate=pooldropout_rate, seed=103, training=training)\n",
    "\n",
    "    # Convolutional layer 2\n",
    "    with tf.name_scope('conv2.1') as scope:\n",
    "        conv2 = tf.layers.conv2d(\n",
    "            pool1,  # Input data\n",
    "            filters=64,  # 32 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 9x9\n",
    "            strides=(1, 1),  # Stride: 1\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=104),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv2.1'\n",
    "        )\n",
    "\n",
    "        conv2 = tf.layers.batch_normalization(\n",
    "            conv2,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn2.1'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv2 = tf.nn.relu(conv2, name='relu2.1')\n",
    "\n",
    "    # Convolutional layer 2\n",
    "    with tf.name_scope('conv2.2') as scope:\n",
    "        conv22 = tf.layers.conv2d(\n",
    "            conv2,  # Input data\n",
    "            filters=64,  # 32 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 9x9\n",
    "            strides=(1, 1),  # Stride: 1\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=1104),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv2.2'\n",
    "        )\n",
    "\n",
    "        conv22 = tf.layers.batch_normalization(\n",
    "            conv22,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn2.2'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv22 = tf.nn.relu(conv22, name='relu2.2')\n",
    "\n",
    "    # Max pooling layer 2\n",
    "    with tf.name_scope('pool2') as scope:\n",
    "        pool2 = tf.layers.max_pooling2d(\n",
    "            conv22,  # Input\n",
    "            pool_size=(2, 2),  # Pool size: 3x3\n",
    "            strides=(2, 2),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            name='pool2'\n",
    "        )\n",
    "\n",
    "        # optional dropout\n",
    "        if dropout:\n",
    "            pool2 = tf.layers.dropout(pool2, rate=pooldropout_rate, seed=106, training=training)\n",
    "\n",
    "    # Convolutional layer 3\n",
    "    with tf.name_scope('conv3.1') as scope:\n",
    "        conv3 = tf.layers.conv2d(\n",
    "            pool2,  # Input data\n",
    "            filters=128,  # 48 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(1, 1),  # Stride: 1\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=107),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv3.1'\n",
    "        )\n",
    "\n",
    "        conv3 = tf.layers.batch_normalization(\n",
    "            conv3,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn3.1'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv3 = tf.nn.relu(conv3, name='relu3.1')\n",
    "\n",
    "    # Convolutional layer 3\n",
    "    with tf.name_scope('conv3.2') as scope:\n",
    "        conv32 = tf.layers.conv2d(\n",
    "            conv3,  # Input data\n",
    "            filters=128,  # 48 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(1, 1),  # Stride: 1\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=1107),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv3.2'\n",
    "        )\n",
    "\n",
    "        conv32 = tf.layers.batch_normalization(\n",
    "            conv32,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn3.2'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv32 = tf.nn.relu(conv32, name='relu3.2')\n",
    "\n",
    "    # Max pooling layer 3\n",
    "    with tf.name_scope('pool3') as scope:\n",
    "        pool3 = tf.layers.max_pooling2d(\n",
    "            conv32,  # Input\n",
    "            pool_size=(2, 2),  # Pool size: 2x2\n",
    "            strides=(2, 2),  # Stride: 2\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            name='pool3'\n",
    "        )\n",
    "\n",
    "        if dropout:\n",
    "            pool3 = tf.layers.dropout(pool3, rate=pooldropout_rate, seed=109, training=training)\n",
    "\n",
    "    # Convolutional layer 4\n",
    "    with tf.name_scope('conv4') as scope:\n",
    "            conv4 = tf.layers.conv2d(\n",
    "                pool3,  # Input data\n",
    "                filters=256,  # 48 filters\n",
    "                kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "                strides=(1, 1),  # Stride: 1\n",
    "                padding='SAME',  # \"same\" padding\n",
    "                activation=None,  # None\n",
    "                kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=110),\n",
    "                kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "                name='conv4'\n",
    "            )\n",
    "\n",
    "            conv4 = tf.layers.batch_normalization(\n",
    "                conv4,\n",
    "                axis=-1,\n",
    "                momentum=0.99,\n",
    "                epsilon=epsilon,\n",
    "                center=True,\n",
    "                scale=True,\n",
    "                beta_initializer=tf.zeros_initializer(),\n",
    "                gamma_initializer=tf.ones_initializer(),\n",
    "                moving_mean_initializer=tf.zeros_initializer(),\n",
    "                moving_variance_initializer=tf.ones_initializer(),\n",
    "                training=training,\n",
    "                name='bn4'\n",
    "            )\n",
    "\n",
    "            # apply relu\n",
    "            conv4_bn_relu = tf.nn.relu(conv4, name='relu4')\n",
    "\n",
    "            #if dropout:\n",
    "            #    conv4_bn_relu = tf.layers.dropout(conv4_bn_relu, rate=convdropout_rate, seed=111, training=training)\n",
    "\n",
    "    # Max pooling layer 4\n",
    "    with tf.name_scope('pool4') as scope:\n",
    "            pool4 = tf.layers.max_pooling2d(\n",
    "                conv4_bn_relu,  # Input\n",
    "                pool_size=(2, 2),  # Pool size: 2x2\n",
    "                strides=(2, 2),  # Stride: 2\n",
    "                padding='SAME',  # \"same\" padding\n",
    "                name='pool4'\n",
    "            )\n",
    "\n",
    "            if dropout:\n",
    "                pool4 = tf.layers.dropout(pool4, rate=pooldropout_rate, seed=112, training=training)\n",
    "\n",
    "            # Convolutional layer 4\n",
    "    with tf.name_scope('conv5') as scope:\n",
    "        conv5 = tf.layers.conv2d(\n",
    "            pool4,  # Input data\n",
    "            filters=512,  # 48 filters\n",
    "            kernel_size=(3, 3),  # Kernel size: 5x5\n",
    "            strides=(1, 1),  # Stride: 1\n",
    "            padding='SAME',  # \"same\" padding\n",
    "            activation=None,  # None\n",
    "            kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2, seed=113),\n",
    "            kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),\n",
    "            name='conv5'\n",
    "        )\n",
    "\n",
    "        conv5 = tf.layers.batch_normalization(\n",
    "            conv5,\n",
    "            axis=-1,\n",
    "            momentum=0.99,\n",
    "            epsilon=epsilon,\n",
    "            center=True,\n",
    "            scale=True,\n",
    "            beta_initializer=tf.zeros_initializer(),\n",
    "            gamma_initializer=tf.ones_initializer(),\n",
    "            moving_mean_initializer=tf.zeros_initializer(),\n",
    "            moving_variance_initializer=tf.ones_initializer(),\n",
    "            training=training,\n",
    "            name='bn5'\n",
    "        )\n",
    "\n",
    "        # apply relu\n",
    "        conv5_bn_relu = tf.nn.relu(conv5, name='relu5')\n",
    "\n",
    "    # Max pooling layer 4\n",
    "    with tf.name_scope('pool5') as scope:\n",
    "        pool5 = tf.layers.max_pooling2d(\n",
    "            conv5_bn_relu,\n",
    "            pool_size=(2, 2),  # Pool size: 2x2\n",
    "            strides=(2, 2),  # Stride: 2\n",
    "            padding='SAME',\n",
    "            name='pool5'\n",
    "        )\n",
    "\n",
    "        if dropout:\n",
    "            pool5 = tf.layers.dropout(pool5, rate=pooldropout_rate, seed=115, training=training)\n",
    "\n",
    "    fc1 = _conv2d_batch_norm(pool5, 2048, kernel_size=(5, 5), stride=(5, 5), training=training, epsilon=1e-8,\n",
    "                             padding=\"VALID\", seed=1013, lambd=lamC, name=\"fc_1\")\n",
    "\n",
    "    fc2 = _conv2d_batch_norm(fc1, 2048, kernel_size=(1, 1), stride=(1, 1), training=training, epsilon=1e-8,\n",
    "                             padding=\"VALID\", seed=1014, lambd=lamC, name=\"fc_2\")\n",
    "\n",
    "    fc3 = tf.layers.dense(\n",
    "        fc2,\n",
    "        num_classes,  # One output unit per category\n",
    "        activation=None,  # No activation function\n",
    "        kernel_initializer=tf.variance_scaling_initializer(scale=1, seed=121),\n",
    "        bias_initializer=tf.zeros_initializer(),\n",
    "        name=\"fc_logits\"\n",
    "    )\n",
    "\n",
    "    logits = tf.squeeze(fc3, name=\"fc_flat_logits\")\n",
    "\n",
    "    # get the fully connected variables so we can only train them when retraining the network\n",
    "    fc_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, \"fc\")\n",
    "\n",
    "    with tf.variable_scope('conv1', reuse=True):\n",
    "        conv_kernels1 = tf.get_variable('kernel')\n",
    "        kernel_transposed = tf.transpose(conv_kernels1, [3, 0, 1, 2])\n",
    "\n",
    "    with tf.variable_scope('visualization'):\n",
    "        tf.summary.image('conv1/filters', kernel_transposed, max_outputs=32, collections=[\"kernels\"])\n",
    "\n",
    "    ## Loss function options\n",
    "    # Regular mean cross entropy\n",
    "    #mean_ce = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits))\n",
    "\n",
    "    # Different weighting method\n",
    "    # This will weight the positive examples higher so as to improve recall\n",
    "    weights = tf.multiply(weight, tf.cast(tf.greater(y, 0), tf.float32)) + 1\n",
    "    mean_ce = tf.reduce_mean(tf.losses.sparse_softmax_cross_entropy(labels=y, logits=logits, weights=weights))\n",
    "\n",
    "    # Add in l2 loss\n",
    "    loss = mean_ce + tf.losses.get_regularization_loss()\n",
    "\n",
    "    # Adam optimizer\n",
    "    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)\n",
    "\n",
    "    # Minimize cross-entropy - freeze certain layers depending on input\n",
    "    if freeze:\n",
    "        train_op = optimizer.minimize(loss, global_step=global_step, var_list=fc_vars)\n",
    "    else:\n",
    "        train_op = optimizer.minimize(loss, global_step=global_step)\n",
    "\n",
    "    # get the probabilites for the classes\n",
    "    probabilities = tf.nn.softmax(logits, name=\"probabilities\")\n",
    "    abnormal_probability = 1 - probabilities[:,0]\n",
    "    \n",
    "    # Compute predictions from the probabilities\n",
    "    if threshold == 0.5:\n",
    "        predictions = tf.argmax(probabilities, axis=1, output_type=tf.int32)\n",
    "    else:\n",
    "        predictions = tf.cast(tf.greater(abnormal_probability, threshold), tf.int32)\n",
    "    \n",
    "    # get the accuracy\n",
    "    accuracy, acc_op = tf.metrics.accuracy(\n",
    "        labels=y,\n",
    "        predictions=predictions,\n",
    "        updates_collections=tf.GraphKeys.UPDATE_OPS,\n",
    "        name=\"accuracy\",\n",
    "    )\n",
    "\n",
    "    # calculate recall\n",
    "    if num_classes > 2:\n",
    "        # collapse the predictions down to normal or not for our pr metrics\n",
    "        zero = tf.constant(0, dtype=tf.int64)\n",
    "        collapsed_predictions = tf.cast(tf.greater(abnormal_probability, threshold), tf.int32)\n",
    "        collapsed_labels = tf.greater(y, 0)\n",
    "\n",
    "        recall, rec_op = tf.metrics.recall(labels=collapsed_labels, predictions=collapsed_predictions, updates_collections=tf.GraphKeys.UPDATE_OPS, name=\"recall\")\n",
    "        precision, prec_op = tf.metrics.precision(labels=collapsed_labels, predictions=collapsed_predictions, updates_collections=tf.GraphKeys.UPDATE_OPS, name=\"precision\")\n",
    "\n",
    "    else:\n",
    "        recall, rec_op = tf.metrics.recall(labels=y, predictions=predictions, updates_collections=tf.GraphKeys.UPDATE_OPS, name=\"recall\")\n",
    "        precision, prec_op = tf.metrics.precision(labels=y, predictions=predictions, updates_collections=tf.GraphKeys.UPDATE_OPS, name=\"precision\")\n",
    "\n",
    "    f1_score = 2 * ((precision * recall) / (precision + recall))\n",
    "    _, update_op = summary_lib.pr_curve_streaming_op(name='pr_curve',\n",
    "                                                     predictions=abnormal_probability,\n",
    "                                                     labels=y,\n",
    "                                                     updates_collections=tf.GraphKeys.UPDATE_OPS,\n",
    "                                                     num_thresholds=20)\n",
    "\n",
    "    tf.summary.scalar('recall_1', recall, collections=[\"summaries\"])\n",
    "    tf.summary.scalar('precision_1', precision, collections=[\"summaries\"])\n",
    "    tf.summary.scalar('f1_score', f1_score, collections=[\"summaries\"])\n",
    "\n",
    "    # Create summary hooks\n",
    "    tf.summary.scalar('accuracy', accuracy, collections=[\"summaries\"])\n",
    "    tf.summary.scalar('cross_entropy', mean_ce, collections=[\"summaries\"])\n",
    "    tf.summary.scalar('learning_rate', learning_rate, collections=[\"summaries\"])\n",
    "\n",
    "    # add this so that the batch norm gets run\n",
    "    extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)\n",
    "\n",
    "    # Merge all the summaries\n",
    "    merged = tf.summary.merge_all(\"summaries\")\n",
    "    kernel_summaries = tf.summary.merge_all(\"kernels\")\n",
    "    per_epoch_summaries = [[]]\n",
    "\n",
    "    print(\"Graph created...\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## CONFIGURE OPTIONS\n",
    "# if we are initializing the weights from a pre-trained model make sure it exists\n",
    "if init_model is not None:\n",
    "    if os.path.exists(os.path.join(\"model\", init_model + '.ckpt.index')):\n",
    "        init = False\n",
    "    else:\n",
    "        init = True\n",
    "# otherwise if this model exists let's continue training it rather than reinitializing it\n",
    "else:\n",
    "    if os.path.exists(os.path.join(\"model\", model_name + '.ckpt.index')):\n",
    "        init = False\n",
    "    else:\n",
    "        init = True\n",
    "\n",
    "meta_data_every = 1\n",
    "log_to_tensorboard = True\n",
    "print_every = 5  # how often to print metrics\n",
    "checkpoint_every = 1  # how often to save model in epochs\n",
    "use_gpu = False  # whether or not to use the GPU\n",
    "print_metrics = True  # whether to print or plot metrics, if False a plot will be created and updated every epoch\n",
    "\n",
    "# Placeholders for metrics\n",
    "train_cost_values = []\n",
    "train_lr_values = []\n",
    "train_acc_values = []\n",
    "train_recall_values = []\n",
    "valid_acc_values = []\n",
    "valid_cost_values = []\n",
    "valid_recall_values = []\n",
    "\n",
    "if use_gpu:\n",
    "    config = tf.ConfigProto()\n",
    "    config.gpu_options.allocator_type = 'BFC'\n",
    "    config.gpu_options.per_process_gpu_memory_fraction = 0.7\n",
    "else:\n",
    "    config = tf.ConfigProto(device_count = {'GPU': 0})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import PIL\n",
    "from PIL import ImageMath\n",
    "import matplotlib.pyplot as plt\n",
    "from scipy.misc import imresize\n",
    "%matplotlib inline\n",
    "\n",
    "def softmax(x):\n",
    "    \"\"\"\n",
    "    Compute softmax values for each sets of scores in x.\n",
    "    \n",
    "    Rows are scores for each class. \n",
    "    Columns are predictions (samples).\n",
    "    \"\"\"\n",
    "    scoreMatExp = np.exp(np.asarray(x))\n",
    "    return scoreMatExp / scoreMatExp.sum(0)\n",
    "\n",
    "def scale_data(input_data, new_min=0.0, new_max=255.0):\n",
    "    old_max = np.max(input_data)\n",
    "    old_min = np.min(input_data)\n",
    "    \n",
    "    adjustment_factor = (new_max - new_min) / (old_max - old_min)\n",
    "    scaled_data = adjustment_factor * (input_data - old_min) + new_min\n",
    "    \n",
    "    return scaled_data\n",
    "\n",
    "def summarize(data):\n",
    "    print(\"Mean:\", np.mean(data))\n",
    "    print(\"Min:\", np.min(data))\n",
    "    print(\"Max:\", np.max(data))\n",
    "    print(\"Std:\", np.std(data))\n",
    "    \n",
    "def resize_image(image, factor):\n",
    "    h = image.shape[0]\n",
    "    w = image.shape[1]\n",
    "    \n",
    "    new_size = (int(h * factor), int(w * factor))\n",
    "    \n",
    "    image = imresize(image, new_size)\n",
    "    \n",
    "    return image\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def load_image(image, mask=None, scale=0.5, margin=0.15, base_path= \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-ddsm\\\\\"):\n",
    "    image_path = os.path.join(base_path, image)\n",
    "    image = PIL.Image.open(image_path)\n",
    "    \n",
    "    # if image is png convert it to RGB so pixels are same scale\n",
    "    if \".png\" in image_path:\n",
    "        image = ImageMath.eval('im/256', {'im':image}).convert('L')\n",
    "    \n",
    "    image = np.array(image)\n",
    "    \n",
    "    # try to lose the extra channels, if there is an error they don't exist\n",
    "    try:\n",
    "        image = image[:,:,0]\n",
    "    except:\n",
    "        pass\n",
    "\n",
    "    if mask is not None:\n",
    "        mask_path = base_path + mask\n",
    "        mask = PIL.Image.open(mask_path)\n",
    "        mask = np.array(mask)\n",
    "        mask = mask[:,:,0]\n",
    "        mask = imresize(mask, (h // 2, w // 2))\n",
    "    \n",
    "    h = image.shape[0]\n",
    "    w = image.shape[1]\n",
    "    \n",
    "    new_h = int(h * scale)\n",
    "    new_w = int(w * scale)\n",
    "    \n",
    "    image = imresize(image, (new_h, new_w))\n",
    "    \n",
    "    # trim margin off each side\n",
    "    h_border = int(h * margin)\n",
    "    w_border = int(w * margin)\n",
    "\n",
    "    image = image[h_border:(h-h_border),w_border:(w-w_border)]\n",
    "    \n",
    "    return image, mask\n",
    "\n",
    "def load_images(dir_name, base_path= \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-ddsm\\\\\"):\n",
    "    directory = base_path + dir_name + \"\\\\\"\n",
    "    files = os.listdir(directory)\n",
    "    \n",
    "    images = []\n",
    "    \n",
    "    for file in files:\n",
    "        image, _ = load_image(file, base_path = directory)\n",
    "        images.append(image)\n",
    "        \n",
    "    return images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./model/model_s2.0.0.36b.10.ckpt\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\eric\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\ipykernel_launcher.py:25: DeprecationWarning: `imresize` is deprecated!\n",
      "`imresize` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.\n",
      "Use ``skimage.transform.resize`` instead.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image: A_1955_1.LEFT_CC.LJPEG.png - Probability: 0.38754448\n",
      "Image: A_1955_1.LEFT_MLO.LJPEG.png - Probability: 0.22468972\n",
      "Image: A_1955_1.RIGHT_CC.LJPEG.png - Probability: 0.5569822\n",
      "Image: A_1955_1.RIGHT_MLO.LJPEG.png - Probability: 0.9949004\n",
      "Image: A_1956_1.LEFT_CC.LJPEG.png - Probability: 0.9614304\n",
      "Image: A_1956_1.LEFT_MLO.LJPEG.png - Probability: 0.9903154\n",
      "Image: A_1956_1.RIGHT_CC.LJPEG.png - Probability: 0.3896608\n",
      "Image: A_1956_1.RIGHT_MLO.LJPEG.png - Probability: 0.9854769\n",
      "Image: A_1957_1.LEFT_CC.LJPEG.png - Probability: 0.7640514\n",
      "Image: A_1957_1.LEFT_MLO.LJPEG.png - Probability: 0.76438993\n",
      "Image: A_1957_1.RIGHT_CC.LJPEG.png - Probability: 0.8690084\n",
      "Image: A_1957_1.RIGHT_MLO.LJPEG.png - Probability: 0.76918286\n",
      "Image: A_1958_1.LEFT_CC.LJPEG.png - Probability: 0.9002469\n",
      "Image: A_1958_1.LEFT_MLO.LJPEG.png - Probability: 0.70626676\n",
      "Image: A_1958_1.RIGHT_CC.LJPEG.png - Probability: 0.98961115\n",
      "Image: A_1958_1.RIGHT_MLO.LJPEG.png - Probability: 0.23822545\n",
      "Image: A_1959_1.LEFT_CC.LJPEG.png - Probability: 0.9775288\n",
      "Image: A_1959_1.LEFT_MLO.LJPEG.png - Probability: 0.096545056\n",
      "Image: A_1959_1.RIGHT_CC.LJPEG.png - Probability: 0.8524244\n",
      "Image: A_1959_1.RIGHT_MLO.LJPEG.png - Probability: 0.36463055\n",
      "Image: A_1960_1.LEFT_CC.LJPEG.png - Probability: 0.98334724\n",
      "Image: A_1960_1.LEFT_MLO.LJPEG.png - Probability: 0.96082044\n",
      "Image: A_1960_1.RIGHT_CC.LJPEG.png - Probability: 0.86329204\n",
      "Image: A_1960_1.RIGHT_MLO.LJPEG.png - Probability: 0.83464664\n",
      "Image: A_1961_1.LEFT_CC.LJPEG.png - Probability: 0.3519853\n",
      "Image: A_1961_1.LEFT_MLO.LJPEG.png - Probability: 0.93646884\n",
      "Image: A_1961_1.RIGHT_CC.LJPEG.png - Probability: 0.9207018\n",
      "Image: A_1961_1.RIGHT_MLO.LJPEG.png - Probability: 0.9181448\n",
      "Image: A_1962_1.LEFT_CC.LJPEG.png - Probability: 0.07387081\n",
      "Image: A_1962_1.LEFT_MLO.LJPEG.png - Probability: 0.8646925\n",
      "Image: A_1962_1.RIGHT_CC.LJPEG.png - Probability: 0.578613\n",
      "Image: A_1962_1.RIGHT_MLO.LJPEG.png - Probability: 0.9662809\n",
      "Image: A_1963_1.LEFT_CC.LJPEG.png - Probability: 0.011976663\n",
      "Image: A_1963_1.LEFT_MLO.LJPEG.png - Probability: 0.87252825\n",
      "Image: A_1963_1.RIGHT_CC.LJPEG.png - Probability: 0.30318114\n",
      "Image: A_1963_1.RIGHT_MLO.LJPEG.png - Probability: 0.9801665\n",
      "Image: A_1964_1.LEFT_CC.LJPEG.png - Probability: 0.918552\n",
      "Image: A_1964_1.LEFT_MLO.LJPEG.png - Probability: 0.20107213\n",
      "Image: A_1964_1.RIGHT_CC.LJPEG.png - Probability: 0.38980743\n",
      "Image: A_1964_1.RIGHT_MLO.LJPEG.png - Probability: 0.6711491\n",
      "Image: A_1965_1.LEFT_CC.LJPEG.png - Probability: 0.36368215\n",
      "Image: A_1965_1.LEFT_MLO.LJPEG.png - Probability: 0.05466447\n",
      "Image: A_1965_1.RIGHT_CC.LJPEG.png - Probability: 0.050107066\n",
      "Image: A_1965_1.RIGHT_MLO.LJPEG.png - Probability: 0.62227714\n",
      "Image: A_1966_1.LEFT_CC.LJPEG.png - Probability: 0.05856569\n",
      "Image: A_1966_1.LEFT_MLO.LJPEG.png - Probability: 0.09052744\n",
      "Image: A_1966_1.RIGHT_CC.LJPEG.png - Probability: 0.7851554\n",
      "Image: A_1966_1.RIGHT_MLO.LJPEG.png - Probability: 0.8047352\n",
      "Image: A_1967_1.LEFT_CC.LJPEG.png - Probability: 0.81144714\n",
      "Image: A_1967_1.LEFT_MLO.LJPEG.png - Probability: 0.32032314\n",
      "Image: A_1967_1.RIGHT_CC.LJPEG.png - Probability: 0.48594594\n",
      "Image: A_1967_1.RIGHT_MLO.LJPEG.png - Probability: 0.83167535\n",
      "Image: A_1968_1.LEFT_CC.LJPEG.png - Probability: 0.8734962\n",
      "Image: A_1968_1.LEFT_MLO.LJPEG.png - Probability: 0.9923959\n",
      "Image: A_1968_1.RIGHT_CC.LJPEG.png - Probability: 0.18573551\n",
      "Image: A_1968_1.RIGHT_MLO.LJPEG.png - Probability: 0.5027686\n",
      "Image: A_1969_1.LEFT_CC.LJPEG.png - Probability: 0.7083251\n",
      "Image: A_1969_1.LEFT_MLO.LJPEG.png - Probability: 0.3015819\n",
      "Image: A_1969_1.RIGHT_CC.LJPEG.png - Probability: 0.265887\n",
      "Image: A_1969_1.RIGHT_MLO.LJPEG.png - Probability: 0.5964965\n",
      "Image: A_1970_1.LEFT_CC.LJPEG.png - Probability: 0.17231146\n",
      "Image: A_1970_1.LEFT_MLO.LJPEG.png - Probability: 0.008115733\n",
      "Image: A_1970_1.RIGHT_CC.LJPEG.png - Probability: 0.018466603\n",
      "Image: A_1970_1.RIGHT_MLO.LJPEG.png - Probability: 0.95229214\n",
      "Image: A_1971_1.LEFT_CC.LJPEG.png - Probability: 0.9806368\n",
      "Image: A_1971_1.LEFT_MLO.LJPEG.png - Probability: 0.983526\n",
      "Image: A_1971_1.RIGHT_CC.LJPEG.png - Probability: 0.9544166\n",
      "Image: A_1971_1.RIGHT_MLO.LJPEG.png - Probability: 0.9973205\n",
      "Image: A_1972_1.LEFT_CC.LJPEG.png - Probability: 0.9812727\n",
      "Image: A_1972_1.LEFT_MLO.LJPEG.png - Probability: 0.7731539\n",
      "Image: A_1972_1.RIGHT_CC.LJPEG.png - Probability: 0.58414394\n",
      "Image: A_1972_1.RIGHT_MLO.LJPEG.png - Probability: 0.72490734\n",
      "Image: A_1973_1.LEFT_CC.LJPEG.png - Probability: 0.5445953\n",
      "Image: A_1973_1.LEFT_MLO.LJPEG.png - Probability: 0.984594\n",
      "Image: A_1973_1.RIGHT_CC.LJPEG.png - Probability: 0.8536173\n",
      "Image: A_1973_1.RIGHT_MLO.LJPEG.png - Probability: 0.22430909\n",
      "Image: A_1974_1.LEFT_CC.LJPEG.png - Probability: 0.98435485\n",
      "Image: A_1974_1.LEFT_MLO.LJPEG.png - Probability: 0.98664874\n",
      "Image: A_1974_1.RIGHT_CC.LJPEG.png - Probability: 0.98983216\n",
      "Image: A_1974_1.RIGHT_MLO.LJPEG.png - Probability: 0.81014436\n",
      "Image: A_1975_1.LEFT_CC.LJPEG.png - Probability: 0.9771438\n",
      "Image: A_1975_1.LEFT_MLO.LJPEG.png - Probability: 0.99873835\n",
      "Image: A_1975_1.RIGHT_CC.LJPEG.png - Probability: 0.069703504\n",
      "Image: A_1975_1.RIGHT_MLO.LJPEG.png - Probability: 0.79240227\n",
      "Image: A_1976_1.LEFT_CC.LJPEG.png - Probability: 0.03257164\n",
      "Image: A_1976_1.LEFT_MLO.LJPEG.png - Probability: 0.6267845\n",
      "Image: A_1976_1.RIGHT_CC.LJPEG.png - Probability: 0.34425476\n",
      "Image: A_1976_1.RIGHT_MLO.LJPEG.png - Probability: 0.88076735\n",
      "Image: A_1977_1.LEFT_CC.LJPEG.png - Probability: 0.06557006\n",
      "Image: A_1977_1.LEFT_MLO.LJPEG.png - Probability: 0.7046083\n",
      "Image: A_1977_1.RIGHT_CC.LJPEG.png - Probability: 0.41738403\n",
      "Image: A_1977_1.RIGHT_MLO.LJPEG.png - Probability: 0.100843035\n",
      "Image: A_1978_1.LEFT_CC.LJPEG.png - Probability: 0.6230633\n",
      "Image: A_1978_1.LEFT_MLO.LJPEG.png - Probability: 0.44845074\n",
      "Image: A_1978_1.RIGHT_CC.LJPEG.png - Probability: 0.91535014\n",
      "Image: A_1978_1.RIGHT_MLO.LJPEG.png - Probability: 0.2272308\n",
      "Image: A_1979_1.LEFT_CC.LJPEG.png - Probability: 0.005341584\n",
      "Image: A_1979_1.LEFT_MLO.LJPEG.png - Probability: 0.64943856\n",
      "Image: A_1979_1.RIGHT_CC.LJPEG.png - Probability: 0.28510666\n",
      "Image: A_1979_1.RIGHT_MLO.LJPEG.png - Probability: 0.63785714\n",
      "Image: A_1980_1.LEFT_CC.LJPEG.png - Probability: 0.037422944\n",
      "Image: A_1980_1.LEFT_MLO.LJPEG.png - Probability: 0.4510757\n",
      "Image: A_1980_1.RIGHT_CC.LJPEG.png - Probability: 0.8181417\n",
      "Image: A_1980_1.RIGHT_MLO.LJPEG.png - Probability: 0.69657165\n",
      "Image: A_1988_1.LEFT_CC.LJPEG.png - Probability: 0.2736695\n",
      "Image: A_1988_1.LEFT_MLO.LJPEG.png - Probability: 0.63629127\n",
      "Image: A_1988_1.RIGHT_CC.LJPEG.png - Probability: 0.37090537\n",
      "Image: A_1988_1.RIGHT_MLO.LJPEG.png - Probability: 0.9914825\n",
      "Image: A_1989_1.LEFT_CC.LJPEG.png - Probability: 0.21771708\n",
      "Image: A_1989_1.LEFT_MLO.LJPEG.png - Probability: 0.26907226\n",
      "Image: A_1989_1.RIGHT_CC.LJPEG.png - Probability: 0.4509594\n",
      "Image: A_1989_1.RIGHT_MLO.LJPEG.png - Probability: 0.23131676\n",
      "Image: A_1990_1.LEFT_CC.LJPEG.png - Probability: 0.23792464\n",
      "Image: A_1990_1.LEFT_MLO.LJPEG.png - Probability: 0.70841247\n",
      "Image: A_1990_1.RIGHT_CC.LJPEG.png - Probability: 0.24044417\n",
      "Image: A_1990_1.RIGHT_MLO.LJPEG.png - Probability: 0.9313499\n",
      "Image: A_1991_1.LEFT_CC.LJPEG.png - Probability: 0.23602472\n",
      "Image: A_1991_1.LEFT_MLO.LJPEG.png - Probability: 0.35002136\n",
      "Image: A_1991_1.RIGHT_CC.LJPEG.png - Probability: 0.58051395\n",
      "Image: A_1991_1.RIGHT_MLO.LJPEG.png - Probability: 0.048590153\n",
      "Image: A_1992_1.LEFT_CC.LJPEG.png - Probability: 0.9112068\n",
      "Image: A_1992_1.LEFT_MLO.LJPEG.png - Probability: 0.8836826\n",
      "Image: A_1992_1.RIGHT_CC.LJPEG.png - Probability: 0.7953117\n",
      "Image: A_1992_1.RIGHT_MLO.LJPEG.png - Probability: 0.8841313\n",
      "Image: A_1993_1.LEFT_MLO.LJPEG.png - Probability: 0.81994325\n",
      "Image: A_1993_1.RIGHT_CC.LJPEG.png - Probability: 0.99758923\n",
      "Image: A_1993_1.RIGHT_MLO.LJPEG.png - Probability: 0.5357797\n",
      "Image: A_2028_1.LEFT_CC.LJPEG.png - Probability: 0.93607855\n",
      "Image: A_2028_1.LEFT_MLO.LJPEG.png - Probability: 0.90748596\n",
      "Image: A_2028_1.RIGHT_CC.LJPEG.png - Probability: 0.9227241\n",
      "Image: A_2028_1.RIGHT_MLO.LJPEG.png - Probability: 0.82268447\n",
      "Image: A_2029_1.LEFT_CC.LJPEG.png - Probability: 0.96802294\n",
      "Image: A_2029_1.LEFT_MLO.LJPEG.png - Probability: 0.07832986\n",
      "Image: A_2029_1.RIGHT_CC.LJPEG.png - Probability: 0.16041973\n",
      "Image: A_2029_1.RIGHT_MLO.LJPEG.png - Probability: 0.93243086\n",
      "Image: A_2030_1.LEFT_CC.LJPEG.png - Probability: 0.97541\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image: A_2030_1.LEFT_MLO.LJPEG.png - Probability: 0.9088892\n",
      "Image: A_2030_1.RIGHT_CC.LJPEG.png - Probability: 0.9865666\n",
      "Image: A_2030_1.RIGHT_MLO.LJPEG.png - Probability: 0.91849434\n",
      "Image: A_2031_1.LEFT_CC.LJPEG.png - Probability: 0.12958908\n",
      "Image: A_2031_1.LEFT_MLO.LJPEG.png - Probability: 0.5006222\n",
      "Image: A_2031_1.RIGHT_CC.LJPEG.png - Probability: 0.6682431\n",
      "Image: A_2031_1.RIGHT_MLO.LJPEG.png - Probability: 0.7689034\n",
      "Image: A_2032_1.LEFT_CC.LJPEG.png - Probability: 0.064146765\n",
      "Image: A_2032_1.LEFT_MLO.LJPEG.png - Probability: 0.98727745\n",
      "Image: A_2032_1.RIGHT_CC.LJPEG.png - Probability: 0.54698527\n",
      "Image: A_2032_1.RIGHT_MLO.LJPEG.png - Probability: 0.8573726\n",
      "Image: A_2033_1.LEFT_CC.LJPEG.png - Probability: 0.08535098\n",
      "Image: A_2033_1.LEFT_MLO.LJPEG.png - Probability: 0.9077915\n",
      "Image: A_2033_1.RIGHT_CC.LJPEG.png - Probability: 0.5375555\n",
      "Image: A_2033_1.RIGHT_MLO.LJPEG.png - Probability: 0.93672496\n",
      "Image: A_2034_1.LEFT_CC.LJPEG.png - Probability: 0.768679\n",
      "Image: A_2034_1.LEFT_MLO.LJPEG.png - Probability: 0.01657121\n",
      "Image: A_2034_1.RIGHT_CC.LJPEG.png - Probability: 0.10556134\n",
      "Image: A_2034_1.RIGHT_MLO.LJPEG.png - Probability: 0.52720404\n",
      "Image: A_2035_1.LEFT_CC.LJPEG.png - Probability: 0.037043102\n",
      "Image: A_2035_1.LEFT_MLO.LJPEG.png - Probability: 0.5132224\n",
      "Image: A_2035_1.RIGHT_CC.LJPEG.png - Probability: 0.09506644\n",
      "Image: A_2035_1.RIGHT_MLO.LJPEG.png - Probability: 0.324348\n",
      "Image: A_2036_1.LEFT_CC.LJPEG.png - Probability: 0.21167693\n",
      "Image: A_2036_1.LEFT_MLO.LJPEG.png - Probability: 0.008984342\n",
      "Image: A_2036_1.RIGHT_CC.LJPEG.png - Probability: 0.7630898\n",
      "Image: A_2036_1.RIGHT_MLO.LJPEG.png - Probability: 0.14646313\n",
      "Image: A_2037_1.LEFT_CC.LJPEG.png - Probability: 0.60630226\n",
      "Image: A_2037_1.LEFT_MLO.LJPEG.png - Probability: 0.25000995\n",
      "Image: A_2037_1.RIGHT_CC.LJPEG.png - Probability: 0.55016035\n",
      "Image: A_2037_1.RIGHT_MLO.LJPEG.png - Probability: 0.66580474\n",
      "Image: A_2038_1.LEFT_CC.LJPEG.png - Probability: 0.9535138\n",
      "Image: A_2038_1.LEFT_MLO.LJPEG.png - Probability: 0.9993812\n",
      "Image: A_2038_1.RIGHT_CC.LJPEG.png - Probability: 0.9884213\n",
      "Image: A_2038_1.RIGHT_MLO.LJPEG.png - Probability: 0.99538463\n",
      "Image: A_2039_1.LEFT_CC.LJPEG.png - Probability: 0.8047081\n",
      "Image: A_2039_1.LEFT_MLO.LJPEG.png - Probability: 0.51075757\n",
      "Image: A_2039_1.RIGHT_CC.LJPEG.png - Probability: 0.10769964\n",
      "Image: A_2039_1.RIGHT_MLO.LJPEG.png - Probability: 0.74223894\n",
      "Image: A_2040_1.LEFT_CC.LJPEG.png - Probability: 0.23956065\n",
      "Image: A_2040_1.LEFT_MLO.LJPEG.png - Probability: 0.15809649\n",
      "Image: A_2040_1.RIGHT_CC.LJPEG.png - Probability: 0.20582293\n",
      "Image: A_2040_1.RIGHT_MLO.LJPEG.png - Probability: 0.16349614\n",
      "Image: A_2041_1.LEFT_CC.LJPEG.png - Probability: 0.7371066\n",
      "Image: A_2041_1.LEFT_MLO.LJPEG.png - Probability: 0.2783282\n",
      "Image: A_2041_1.RIGHT_CC.LJPEG.png - Probability: 0.06781866\n",
      "Image: A_2041_1.RIGHT_MLO.LJPEG.png - Probability: 0.31557792\n",
      "Image: A_2042_1.LEFT_CC.LJPEG.png - Probability: 0.55100065\n",
      "Image: A_2042_1.LEFT_MLO.LJPEG.png - Probability: 0.906779\n",
      "Image: A_2042_1.RIGHT_CC.LJPEG.png - Probability: 0.54536504\n",
      "Image: A_2042_1.RIGHT_MLO.LJPEG.png - Probability: 0.18409681\n",
      "Image: A_2043_1.LEFT_CC.LJPEG.png - Probability: 0.6635581\n",
      "Image: A_2043_1.LEFT_MLO.LJPEG.png - Probability: 0.34454563\n",
      "Image: A_2043_1.RIGHT_CC.LJPEG.png - Probability: 0.95724326\n",
      "Image: A_2043_1.RIGHT_MLO.LJPEG.png - Probability: 0.24144545\n",
      "Image: D_4500_1.LEFT_CC.LJPEG.png - Probability: 0.85343856\n",
      "Image: D_4500_1.LEFT_MLO.LJPEG.png - Probability: 0.3547532\n",
      "Image: D_4500_1.RIGHT_CC.LJPEG.png - Probability: 0.28101563\n",
      "Image: D_4500_1.RIGHT_MLO.LJPEG.png - Probability: 0.18678351\n",
      "Image: D_4501_1.LEFT_CC.LJPEG.png - Probability: 0.81050825\n",
      "Image: D_4501_1.LEFT_MLO.LJPEG.png - Probability: 0.8854268\n",
      "Image: D_4501_1.RIGHT_CC.LJPEG.png - Probability: 0.9614435\n",
      "Image: D_4501_1.RIGHT_MLO.LJPEG.png - Probability: 0.8545884\n",
      "Image: D_4502_1.LEFT_CC.LJPEG.png - Probability: 0.49321812\n",
      "Image: D_4502_1.LEFT_MLO.LJPEG.png - Probability: 0.13582237\n",
      "Image: D_4502_1.RIGHT_CC.LJPEG.png - Probability: 0.282686\n",
      "Image: D_4502_1.RIGHT_MLO.LJPEG.png - Probability: 0.107562415\n",
      "Image: D_4503_1.LEFT_CC.LJPEG.png - Probability: 0.78172356\n",
      "Image: D_4503_1.LEFT_MLO.LJPEG.png - Probability: 0.003621916\n",
      "Image: D_4503_1.RIGHT_CC.LJPEG.png - Probability: 0.012440541\n",
      "Image: D_4503_1.RIGHT_MLO.LJPEG.png - Probability: 0.6827681\n",
      "Image: D_4505_1.LEFT_CC.LJPEG.png - Probability: 0.31666848\n",
      "Image: D_4505_1.LEFT_MLO.LJPEG.png - Probability: 0.5112745\n",
      "Image: D_4505_1.RIGHT_CC.LJPEG.png - Probability: 0.92456347\n",
      "Image: D_4505_1.RIGHT_MLO.LJPEG.png - Probability: 0.7286086\n",
      "Image: D_4506_1.LEFT_CC.LJPEG.png - Probability: 0.9777485\n",
      "Image: D_4506_1.LEFT_MLO.LJPEG.png - Probability: 0.027455613\n",
      "Image: D_4506_1.RIGHT_CC.LJPEG.png - Probability: 0.5627477\n",
      "Image: D_4506_1.RIGHT_MLO.LJPEG.png - Probability: 0.07804874\n",
      "Image: D_4508_1.LEFT_CC.LJPEG.png - Probability: 0.2743794\n",
      "Image: D_4508_1.LEFT_MLO.LJPEG.png - Probability: 0.60690427\n",
      "Image: D_4508_1.RIGHT_CC.LJPEG.png - Probability: 0.51942474\n",
      "Image: D_4508_1.RIGHT_MLO.LJPEG.png - Probability: 0.22763151\n",
      "Image: D_4509_1.LEFT_CC.LJPEG.png - Probability: 0.8480802\n",
      "Image: D_4509_1.LEFT_MLO.LJPEG.png - Probability: 0.22559513\n",
      "Image: D_4509_1.RIGHT_CC.LJPEG.png - Probability: 0.5561981\n",
      "Image: D_4509_1.RIGHT_MLO.LJPEG.png - Probability: 0.13729547\n",
      "Image: D_4510_1.LEFT_CC.LJPEG.png - Probability: 0.9391029\n",
      "Image: D_4510_1.LEFT_MLO.LJPEG.png - Probability: 0.99643576\n",
      "Image: D_4510_1.RIGHT_CC.LJPEG.png - Probability: 0.4012142\n",
      "Image: D_4510_1.RIGHT_MLO.LJPEG.png - Probability: 0.82342577\n",
      "Image: D_4511_1.LEFT_CC.LJPEG.png - Probability: 0.9726238\n",
      "Image: D_4511_1.LEFT_MLO.LJPEG.png - Probability: 0.99700195\n",
      "Image: D_4511_1.RIGHT_CC.LJPEG.png - Probability: 0.92976236\n",
      "Image: D_4511_1.RIGHT_MLO.LJPEG.png - Probability: 0.99683785\n",
      "Image: D_4512_1.LEFT_CC.LJPEG.png - Probability: 0.31413758\n",
      "Image: D_4512_1.LEFT_MLO.LJPEG.png - Probability: 0.9048087\n",
      "Image: D_4512_1.RIGHT_CC.LJPEG.png - Probability: 0.23240595\n",
      "Image: D_4512_1.RIGHT_MLO.LJPEG.png - Probability: 0.40476257\n",
      "Image: D_4513_1.LEFT_CC.LJPEG.png - Probability: 0.021899026\n",
      "Image: D_4513_1.LEFT_MLO.LJPEG.png - Probability: 0.023455787\n",
      "Image: D_4513_1.RIGHT_CC.LJPEG.png - Probability: 0.6220343\n",
      "Image: D_4513_1.RIGHT_MLO.LJPEG.png - Probability: 0.045628533\n",
      "Image: D_4514_1.LEFT_CC.LJPEG.png - Probability: 0.7924938\n",
      "Image: D_4514_1.LEFT_MLO.LJPEG.png - Probability: 0.76808614\n",
      "Image: D_4514_1.RIGHT_CC.LJPEG.png - Probability: 0.140951\n",
      "Image: D_4514_1.RIGHT_MLO.LJPEG.png - Probability: 0.9982529\n",
      "Image: D_4515_1.LEFT_CC.LJPEG.png - Probability: 0.002031478\n",
      "Image: D_4515_1.LEFT_MLO.LJPEG.png - Probability: 0.16807817\n",
      "Image: D_4515_1.RIGHT_CC.LJPEG.png - Probability: 0.011122936\n",
      "Image: D_4515_1.RIGHT_MLO.LJPEG.png - Probability: 0.58385885\n",
      "Image: D_4516_1.LEFT_CC.LJPEG.png - Probability: 0.022381708\n",
      "Image: D_4516_1.LEFT_MLO.LJPEG.png - Probability: 0.07078796\n",
      "Image: D_4516_1.RIGHT_CC.LJPEG.png - Probability: 0.053560227\n",
      "Image: D_4516_1.RIGHT_MLO.LJPEG.png - Probability: 0.44621858\n",
      "Image: D_4517_1.LEFT_CC.LJPEG.png - Probability: 0.023811255\n",
      "Image: D_4517_1.LEFT_MLO.LJPEG.png - Probability: 0.002047023\n",
      "Image: D_4517_1.RIGHT_CC.LJPEG.png - Probability: 0.3574041\n",
      "Image: D_4517_1.RIGHT_MLO.LJPEG.png - Probability: 0.013620861\n",
      "Image: D_4518_1.LEFT_CC.LJPEG.png - Probability: 0.22283475\n",
      "Image: D_4518_1.LEFT_MLO.LJPEG.png - Probability: 0.04050203\n",
      "Image: D_4518_1.RIGHT_CC.LJPEG.png - Probability: 0.0789868\n",
      "Image: D_4518_1.RIGHT_MLO.LJPEG.png - Probability: 0.046954334\n",
      "Image: D_4519_1.LEFT_CC.LJPEG.png - Probability: 0.9457932\n",
      "Image: D_4519_1.LEFT_MLO.LJPEG.png - Probability: 0.12769385\n",
      "Image: D_4519_1.RIGHT_CC.LJPEG.png - Probability: 0.9717794\n",
      "Image: D_4519_1.RIGHT_MLO.LJPEG.png - Probability: 0.8159869\n",
      "Image: D_4520_1.LEFT_CC.LJPEG.png - Probability: 0.93129355\n",
      "Image: D_4520_1.LEFT_MLO.LJPEG.png - Probability: 0.019133072\n",
      "Image: D_4520_1.RIGHT_CC.LJPEG.png - Probability: 0.0031947526\n",
      "Image: D_4520_1.RIGHT_MLO.LJPEG.png - Probability: 0.18562196\n",
      "Image: D_4521_1.LEFT_CC.LJPEG.png - Probability: 0.08205832\n",
      "Image: D_4521_1.LEFT_MLO.LJPEG.png - Probability: 0.75327003\n",
      "Image: D_4521_1.RIGHT_CC.LJPEG.png - Probability: 0.14782085\n",
      "Image: D_4521_1.RIGHT_MLO.LJPEG.png - Probability: 0.93413967\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image: D_4522_1.LEFT_CC.LJPEG.png - Probability: 0.41085225\n",
      "Image: D_4522_1.LEFT_MLO.LJPEG.png - Probability: 0.0071904785\n",
      "Image: D_4522_1.RIGHT_CC.LJPEG.png - Probability: 0.03202089\n",
      "Image: D_4522_1.RIGHT_MLO.LJPEG.png - Probability: 0.03281414\n",
      "Image: D_4523_1.LEFT_CC.LJPEG.png - Probability: 0.013299606\n",
      "Image: D_4523_1.LEFT_MLO.LJPEG.png - Probability: 0.23269843\n",
      "Image: D_4523_1.RIGHT_CC.LJPEG.png - Probability: 0.011878761\n",
      "Image: D_4523_1.RIGHT_MLO.LJPEG.png - Probability: 0.79496783\n",
      "Image: D_4524_1.LEFT_CC.LJPEG.png - Probability: 0.6773831\n",
      "Image: D_4524_1.LEFT_MLO.LJPEG.png - Probability: 0.0005195638\n",
      "Image: D_4524_1.RIGHT_CC.LJPEG.png - Probability: 0.54744655\n",
      "Image: D_4524_1.RIGHT_MLO.LJPEG.png - Probability: 0.042927258\n",
      "Image: D_4525_1.LEFT_CC.LJPEG.png - Probability: 0.865803\n",
      "Image: D_4525_1.LEFT_MLO.LJPEG.png - Probability: 0.68207014\n",
      "Image: D_4525_1.RIGHT_CC.LJPEG.png - Probability: 0.25873443\n",
      "Image: D_4525_1.RIGHT_MLO.LJPEG.png - Probability: 0.416618\n",
      "Image: D_4526_1.LEFT_CC.LJPEG.png - Probability: 0.01539698\n",
      "Image: D_4526_1.LEFT_MLO.LJPEG.png - Probability: 0.16480984\n",
      "Image: D_4526_1.RIGHT_CC.LJPEG.png - Probability: 0.02108919\n",
      "Image: D_4526_1.RIGHT_MLO.LJPEG.png - Probability: 0.32705808\n",
      "Image: D_4527_1.LEFT_CC.LJPEG.png - Probability: 0.0046995874\n",
      "Image: D_4527_1.LEFT_MLO.LJPEG.png - Probability: 0.21055308\n",
      "Image: D_4527_1.RIGHT_CC.LJPEG.png - Probability: 0.017913425\n",
      "Image: D_4527_1.RIGHT_MLO.LJPEG.png - Probability: 0.38607955\n",
      "Image: D_4528_1.LEFT_CC.LJPEG.png - Probability: 0.3444694\n",
      "Image: D_4528_1.LEFT_MLO.LJPEG.png - Probability: 0.2344284\n",
      "Image: D_4528_1.RIGHT_CC.LJPEG.png - Probability: 0.99209726\n",
      "Image: D_4528_1.RIGHT_MLO.LJPEG.png - Probability: 0.7108349\n",
      "Image: D_4529_1.LEFT_CC.LJPEG.png - Probability: 0.02220124\n",
      "Image: D_4529_1.LEFT_MLO.LJPEG.png - Probability: 0.120977595\n",
      "Image: D_4529_1.RIGHT_CC.LJPEG.png - Probability: 0.9952992\n",
      "Image: D_4529_1.RIGHT_MLO.LJPEG.png - Probability: 0.65849215\n",
      "Image: D_4530_1.LEFT_CC.LJPEG.png - Probability: 0.0063714115\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-40-97302f2012ae>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m()\u001b[0m\n\u001b[0;32m     30\u001b[0m         yhat = sess.run(probabilities, feed_dict = {\n\u001b[0;32m     31\u001b[0m             \u001b[0mX\u001b[0m\u001b[1;33m:\u001b[0m \u001b[0mimage\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m---> 32\u001b[1;33m             \u001b[0mtraining\u001b[0m\u001b[1;33m:\u001b[0m \u001b[1;32mFalse\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m     33\u001b[0m             })\n\u001b[0;32m     34\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36mrun\u001b[1;34m(self, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[0;32m    903\u001b[0m     \u001b[1;32mtry\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    904\u001b[0m       result = self._run(None, fetches, feed_dict, options_ptr,\n\u001b[1;32m--> 905\u001b[1;33m                          run_metadata_ptr)\n\u001b[0m\u001b[0;32m    906\u001b[0m       \u001b[1;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    907\u001b[0m         \u001b[0mproto_data\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m_run\u001b[1;34m(self, handle, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[0;32m   1138\u001b[0m     \u001b[1;32mif\u001b[0m \u001b[0mfinal_fetches\u001b[0m \u001b[1;32mor\u001b[0m \u001b[0mfinal_targets\u001b[0m \u001b[1;32mor\u001b[0m \u001b[1;33m(\u001b[0m\u001b[0mhandle\u001b[0m \u001b[1;32mand\u001b[0m \u001b[0mfeed_dict_tensor\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1139\u001b[0m       results = self._do_run(handle, final_targets, final_fetches,\n\u001b[1;32m-> 1140\u001b[1;33m                              feed_dict_tensor, options, run_metadata)\n\u001b[0m\u001b[0;32m   1141\u001b[0m     \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1142\u001b[0m       \u001b[0mresults\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;33m[\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m_do_run\u001b[1;34m(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)\u001b[0m\n\u001b[0;32m   1319\u001b[0m     \u001b[1;32mif\u001b[0m \u001b[0mhandle\u001b[0m \u001b[1;32mis\u001b[0m \u001b[1;32mNone\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1320\u001b[0m       return self._do_call(_run_fn, feeds, fetches, targets, options,\n\u001b[1;32m-> 1321\u001b[1;33m                            run_metadata)\n\u001b[0m\u001b[0;32m   1322\u001b[0m     \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1323\u001b[0m       \u001b[1;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_do_call\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0m_prun_fn\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfeeds\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfetches\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m_do_call\u001b[1;34m(self, fn, *args)\u001b[0m\n\u001b[0;32m   1325\u001b[0m   \u001b[1;32mdef\u001b[0m \u001b[0m_do_call\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1326\u001b[0m     \u001b[1;32mtry\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1327\u001b[1;33m       \u001b[1;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   1328\u001b[0m     \u001b[1;32mexcept\u001b[0m \u001b[0merrors\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mOpError\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1329\u001b[0m       \u001b[0mmessage\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mcompat\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mas_text\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0me\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mmessage\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m_run_fn\u001b[1;34m(feed_dict, fetch_list, target_list, options, run_metadata)\u001b[0m\n\u001b[0;32m   1310\u001b[0m       \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_extend_graph\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1311\u001b[0m       return self._call_tf_sessionrun(\n\u001b[1;32m-> 1312\u001b[1;33m           options, feed_dict, fetch_list, target_list, run_metadata)\n\u001b[0m\u001b[0;32m   1313\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1314\u001b[0m     \u001b[1;32mdef\u001b[0m \u001b[0m_prun_fn\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mhandle\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\Anaconda2\\envs\\exts-aml2\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m_call_tf_sessionrun\u001b[1;34m(self, options, feed_dict, fetch_list, target_list, run_metadata)\u001b[0m\n\u001b[0;32m   1418\u001b[0m         return tf_session.TF_Run(\n\u001b[0;32m   1419\u001b[0m             \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_session\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0moptions\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtarget_list\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1420\u001b[1;33m             status, run_metadata)\n\u001b[0m\u001b[0;32m   1421\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1422\u001b[0m   \u001b[1;32mdef\u001b[0m \u001b[0m_call_tf_sessionprun\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "## Predictions for directory of images\n",
    "directory = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\ddsm\\\\Howtek pngs\"\n",
    "# directory = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\MIAS\"\n",
    "files = os.listdir(directory)\n",
    "\n",
    "scale = 0.5\n",
    "how = \"normal\"\n",
    "dataset = 9\n",
    "model_name = \"model_s2.0.0.36b.10\"\n",
    "\n",
    "## Evaluate on test data multi-class\n",
    "with tf.Session(graph=graph, config=config) as sess:\n",
    "    # create the saver\n",
    "    saver = tf.train.Saver()\n",
    "    \n",
    "    # restore the model\n",
    "    sess.run(tf.local_variables_initializer())\n",
    "    saver.restore(sess, './model/' + model_name + '.ckpt')\n",
    "    \n",
    "    max_probs = []\n",
    "    all_probs = []\n",
    "    count = 0\n",
    "    \n",
    "    for file in files:\n",
    "        image, _ = load_image(file, base_path=directory, margin=0.0, scale=scale)\n",
    "        \n",
    "        # reshape the input so tensorflow can handle it\n",
    "        image = image.reshape(1,image.shape[0], image.shape[1], 1)\n",
    "        \n",
    "        yhat = sess.run(probabilities, feed_dict = {\n",
    "            X: image,\n",
    "            training: False,\n",
    "            })\n",
    "        \n",
    "        abnormal_probs = yhat[:,:,1]\n",
    "        all_probs.append(abnormal_probs)\n",
    "        max_prob = np.max(abnormal_probs)\n",
    "        max_probs.append(max_prob)\n",
    "        print(\"Image:\", file, \"- Probability:\", max_prob)\n",
    "        \n",
    "        if max_prob > 0.75:\n",
    "            count += 1\n",
    "            \n",
    "print(\"Positive Predictions:\", count, \"out of\", len(files))\n",
    "print(\"Mean Max Prob:\", np.mean(max_probs))        \n",
    "print(\"Min Max Prob:\", np.min(max_probs))        \n",
    "print(\"Max Max Prob:\", np.max(max_probs))   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Positive Predictions: 111 out of 551\n"
     ]
    }
   ],
   "source": [
    "print(\"Positive Predictions:\", count, \"out of\", len(files))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TP: 40\n",
      "Positive Predictions: 110\n",
      "Accuracy: 0.5496894409937888\n",
      "Recall: 0.34782608695652173\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "all_cases_df = pd.read_table('data/Info.txt', delimiter=' ')\n",
    "all_cases_df.drop_duplicates(subset=['REFNUM'], keep='first', inplace=True)\n",
    "preds = max_probs > 0.75\n",
    "actual_abnormal = all_cases_df.CLASS != \"NORM\"\n",
    "correct = (preds == actual_abnormal)\n",
    "tp = (actual_abnormal & preds)\n",
    "\n",
    "print(\"TP:\",np.sum(tp))\n",
    "print(\"Positive Predictions:\",np.sum(preds))\n",
    "print(\"Accuracy:\", np.sum(correct) / len(preds))\n",
    "print(\"Recall:\",np.sum(tp) / np.sum(actual_abnormal))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1d09103d828>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image: (5116, 2641)\n",
      "Max: 255\n",
      "Min: 0\n"
     ]
    }
   ],
   "source": [
    "# load an image to classify\n",
    "# Calc Train Positive Full Image\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-ddsm\\\\Calc Train All Full JPEGs\\\\P_00019_RIGHT_CC_FULL.jpg\"\n",
    "\n",
    "# Mass Train Positive Full Image\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-ddsm\\\\Mass Train All JPEGs Full\\\\P_00064_RIGHT_MLO_FULL.jpg\"\n",
    "\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-ddsm\\\\Mass Test All Full JPEGs\\\\P_00032_RIGHT_CC_FULL.jpg\"\n",
    "\n",
    "# negative howtek image\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\ddsm\\\\Howtek pngs\\\\A_1969_1.LEFT_CC.LJPEG.png\"\n",
    "\n",
    "# negative lumisys image that is predicted positive\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\ddsm\\\\Lumisys pngs\\\\B_3615_1.LEFT_MLO.LJPEG.png\"\n",
    "\n",
    "# negative dba image\n",
    "# image_path = \"E:\\\\Documents\\\\Mammography\\\\DDSM\\\\PNGs\\\\DBA pngs\\\\A_0026_1.LEFT_MLO.LJPEG.png\"\n",
    "\n",
    "# mias image\n",
    "# image_path = \"E:\\\\Documents\\\\Mammography\\\\MIAS\\\\JPEG\\\\mdb004.jpg\"\n",
    "\n",
    "# single tile\n",
    "# image_path = \"C:\\\\Users\\\\eric\\\\Documents\\\\Courses\\\\Applied ML 2\\\\mammography\\\\data\\\\cbis-images\\\\1_P_00024_LEFT_MLO_1.png\"\n",
    "\n",
    "# open the image\n",
    "image = PIL.Image.open(image_path)\n",
    "\n",
    "if \".png\" in image_path:\n",
    "    image = ImageMath.eval('im/256', {'im':image}).convert('L')\n",
    "\n",
    "# convert to numpy array\n",
    "image = np.array(image)\n",
    "\n",
    "# try to lose the extra channels, if there is an error they don't exist\n",
    "try:\n",
    "    image = image[:,:,0]\n",
    "except:\n",
    "    pass\n",
    "\n",
    "# size it down by half\n",
    "# image = resize_image(image, 0.5)\n",
    "# image = resize_image(image, 2.58)\n",
    "\n",
    "h = image.shape[0]\n",
    "w = image.shape[1]\n",
    "\n",
    "trim_borders = False\n",
    "if trim_borders:\n",
    "    # trim 15% off each side\n",
    "    h_border = int(h * 0.15)\n",
    "    w_border = int(w * 0.15)\n",
    "\n",
    "    image = image[h_border:(h-h_border),w_border:(w-w_border)]\n",
    "\n",
    "plt.imshow(image)\n",
    "plt.show()\n",
    "\n",
    "\n",
    "print(\"Image:\", image.shape)\n",
    "print(\"Max:\", np.max(image))\n",
    "print(\"Min:\", np.min(image))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1d0e90ccb38>"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1d085bf37f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "image[image <= 20] = 255\n",
    "plt.imshow(image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./model/model_s2.0.0.36b.101.ckpt\n",
      "[nan nan]\n",
      "Yhat: (2,)\n",
      "Max Abnormal Prob: nan\n"
     ]
    }
   ],
   "source": [
    "## Predictions for single image\n",
    "how = \"normal\"\n",
    "dataset = 10\n",
    "model_name = \"model_s2.0.0.36b.101\"\n",
    "\n",
    "# reshape the input so tensorflow can handle it\n",
    "if len(image.shape) != 4:\n",
    "    image = image.reshape(1,image.shape[0], image.shape[1], 1)\n",
    "\n",
    "## Evaluate on test data multi-class\n",
    "with tf.Session(graph=graph, config=config) as sess:\n",
    "    # create the saver\n",
    "    saver = tf.train.Saver()\n",
    "    \n",
    "    # restore the model\n",
    "    sess.run(tf.local_variables_initializer())\n",
    "    saver.restore(sess, './model/' + model_name + '.ckpt')\n",
    "\n",
    "    yhat = sess.run(probabilities, feed_dict = {\n",
    "        X: image,\n",
    "        training: False,\n",
    "        })\n",
    "    print(yhat)\n",
    "\n",
    "print(\"Yhat:\", yhat.shape)\n",
    "try:\n",
    "    abnormal_probs = yhat[:,:,1]\n",
    "except:\n",
    "    abnormal_probs = yhat[1]\n",
    "print(\"Max Abnormal Prob:\", np.max(abnormal_probs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Positive - R: 2  C: 5 Prob: 0.9158387\n",
      "Positive - R: 3  C: 4 Prob: 0.9550749\n",
      "Positive - R: 4  C: 3 Prob: 0.9460572\n",
      "Positive - R: 7  C: 2 Prob: 0.98671794\n",
      "Positive - R: 7  C: 3 Prob: 0.7244829\n",
      "Positive - R: 8  C: 3 Prob: 0.9021839\n",
      "Positive - R: 8  C: 5 Prob: 0.8037197\n",
      "Positive - R: 9  C: 4 Prob: 0.9974414\n",
      "Positive - R: 10  C: 5 Prob: 0.97893196\n",
      "Max Prob: 0.9974414\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23330d70c18>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "h = yhat.shape[0]\n",
    "w = yhat.shape[1]\n",
    "probs = np.zeros_like(yhat)\n",
    "max_prob = 0\n",
    "\n",
    "for y in range(h):\n",
    "    for x in range(w):\n",
    "        if yhat[y,x,1] > 0.5:\n",
    "            print(\"Positive - R:\", y, \" C:\", x, \"Prob:\", yhat[y,x,1])\n",
    "        \n",
    "        if yhat[y,x,1] > max_prob:\n",
    "            max_prob = yhat[y,x,1]\n",
    "            \n",
    "print(\"Max Prob:\", max_prob)\n",
    "\n",
    "vertlines = image.shape[1] / yhat.shape[0]\n",
    "horlines = image.shape[2] / yhat.shape[1]\n",
    "\n",
    "yticks = np.arange(0, image.shape[1], vertlines)\n",
    "xticks = np.arange(0, image.shape[2], horlines)\n",
    "\n",
    "plt.imshow(image[0,:,:,0])\n",
    "plt.grid(True)\n",
    "plt.yticks(yticks)\n",
    "plt.xticks(xticks)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}