TF-UNS.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Ultrasound Nerve Segmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "import os, glob, math, cv2, time\n",
    "import numpy as np\n",
    "import sys\n",
    "import re\n",
    "import csv\n",
    
    "from tensorflow.contrib import slim\n",
    "from tensorflow.contrib.layers.python.layers import layers\n",
    "from scipy.misc import imresize\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's load the training data, images will be resized to 128x128 and will later be converted to TFrecords format so we dont need to hold them into memory."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train_path='.../train/' #path where the training images are\n",
    "test_path='.../test/' #path to where the testing images are\n",
    "record_path='.../tfrecords/' #where the TFrecords will be saved\n",
    "tensorb_path='.../events/'#where the tensorboard events will be saved\n",
    "ckpt_path=\".../checkpoints/\" #where checkpoints will be saved"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "height=128  #original size : 420*580\n",
    "width=128\n",
    "def process_image(img_file):  \n",
    " \"rezize images\"\n",
    " img = cv2.imread(img_file,1)\n",
    " img = (cv2.resize(img,(height,width)).astype('uint8'))\n",
    " img = (cv2.cvtColor(img, cv2.COLOR_BGR2GRAY))\n",
    " return img\n",
    "\n",
    "def load_training():\n",
    " start = time.time()\n",
    " id_patient=[]\n",
    " X_train=[] \n",
    " Y_train=[]\n",
    "\n",
    " print('Loading training data')\n",
    " path = train_path+'*[0-9].tif'\n",
    " files = glob.glob(path)\n",
    " for im_file in files:\n",
    "  id_patient.append(os.path.basename(im_file).split('_')[0])\n",
    "  img=process_image(im_file)\n",
    "  mask=os.path.splitext(im_file)[0]+'_mask.tif'\n",
    "  X_train.append(img)\n",
    "  imgmask= process_image(mask)\n",
    "  Y_train.append(imgmask)   \n",
    " \n",
    " end = time.time() - start\n",
    " print(\"Number of training examples : %.4r\" % (len(X_train)))\n",
    " print(\"Time: %.2f seconds\" % end)\n",
    " X_train=np.asarray(X_train)\n",
    " Y_train=np.asarray(Y_train)\n",
    " return X_train,Y_train,id_patient\n",
    "\n",
    "\n",
    "def load_testing():\n",
    " start = time.time()\n",
    " test=[]\n",
    " idtest=[]\n",
    " print('Loading testing data')\n",
    " path = test_path+'*.tif'\n",
    " files = glob.glob(path)\n",
    " for im_file in files:\n",
    "  img=process_image(im_file)\n",
    "  test.append(img)\n",
    "  idtest.append((im_file.split('/')[-1].split('.')[0]))\n",
    " end = time.time() - start\n",
    " print(\"Number of testing examples : %.4r\" % (len(test)))\n",
    " print(\"Time: %.2f seconds\" % end)\n",
    " test=np.asarray(test)\n",
    " return test,idtest\n",
    "\n",
    "def show_image_and_mask(idx):\n",
    "    \"\"\" Displays a training image and the mask. \"\"\"\n",
    "    im = X[idx]\n",
    "    plt.subplot(121)\n",
    "    plt.imshow(im,cmap=\"gray\")\n",
    "\n",
    "    mask = Y[idx]\n",
    "    plt.subplot(122)\n",
    "    plt.imshow(mask,cmap=\"gray\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading training data\n",
      "Number of training examples : 5635\n",
      "Time: 103.16 seconds\n"
     ]
    }
   ],
   "source": [
    "X,Y,id_patient=load_training()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading testing data\n",
      "Number of testing examples : 5508\n",
      "Time: 62.50 seconds\n"
     ]
    }
   ],
   "source": [
    "test,idtest=load_testing()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is what our images looks like"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAgYAAAEBCAYAAAAD0m2FAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0739df8ed0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image_and_mask(42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we split the images into training and validation sets. We want to split on patients rather than randomly to avoid overfitting. There are 47 patients, each have either 119 or 120 images. We takes 39 patients as training and the remaining 8 as validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "Xtrain=X[0:4552]\n",
    "Ytrain=Y[0:4552]\n",
    "Xvalid=X[4552:]\n",
    "Yvalid=Y[4552:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's convert them to TFrecords format"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def _int64_feature(value):\n",
    "  return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))\n",
    "def _bytes_feature(value):\n",
    "  return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))\n",
    "\n",
    "def convert_to(images,masks,name):\n",
    "  num_examples = images.shape[0]\n",
    "  rows = images.shape[1]\n",
    "  cols = images.shape[2]\n",
    "  depth = 1\n",
    "\n",
    "  filename = os.path.join(record_path, name + '.tfrecords')\n",
    "  print('Writing', filename)\n",
    "  writer = tf.python_io.TFRecordWriter(filename)\n",
    "  for index in range(num_examples):\n",
    "    image_raw = images[index].tostring()\n",
    "    mask_raw=masks[index].tostring()\n",
    "    example = tf.train.Example(features=tf.train.Features(feature={\n",
    "        'height': _int64_feature(rows),\n",
    "        'width': _int64_feature(cols),\n",
    "        'depth': _int64_feature(depth),\n",
    "        'image_raw': _bytes_feature(image_raw),\n",
    "        'mask_raw': _bytes_feature(mask_raw)}))\n",
    "    writer.write(example.SerializeToString())\n",
    "  writer.close()\n",
    "\n",
    "def convert_topred(images,idtest,name):\n",
    "  num_examples = images.shape[0]\n",
    "  rows = images.shape[1]\n",
    "  cols = images.shape[2]\n",
    "  depth = 1\n",
    "\n",
    "  filename = os.path.join(record_path, name + '.tfrecords')\n",
    "  print('Writing', filename)\n",
    "  writer = tf.python_io.TFRecordWriter(filename)\n",
    "  for index in range(num_examples):\n",
    "    image_raw = images[index].tostring()\n",
    "    example = tf.train.Example(features=tf.train.Features(feature={\n",
    "        'height': _int64_feature(rows),\n",
    "        'width': _int64_feature(cols),\n",
    "        'depth': _int64_feature(depth),\n",
    "        'image_raw': _bytes_feature(image_raw),\n",
    "        'idtest': _int64_feature(int(idtest[index]))}))\n",
    "    writer.write(example.SerializeToString())\n",
    "  writer.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Writing', '/media/florian/My Passport/git/trainrecord.tfrecords')\n",
      "('Writing', '/media/florian/My Passport/git/validrecord.tfrecords')\n",
      "('Writing', '/media/florian/My Passport/git/testrecord.tfrecords')\n"
     ]
    }
   ],
   "source": [
    "convert_to(Xtrain,Ytrain,'trainrecord')\n",
    "convert_to(Xvalid,Yvalid,'validrecord')\n",
    "convert_topred(test,idtest,'testrecord')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Writing', '/media/florian/My Passport/git/testrecord.tfrecords')\n"
     ]
    }
   ],
   "source": [
    "convert_topred(test,idtest,'testrecord')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "del X,Y,test,Xtrain,Ytrain,Xvalid,Yvalid"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we have our data as TFrecords , we need to be able to read and decode them"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def read_and_decode(filename_queue):\n",
    "  reader = tf.TFRecordReader()\n",
    "  _, serialized_example = reader.read(filename_queue)\n",
    "  features = tf.parse_single_example(\n",
    "      serialized_example,\n",
    "      features={\n",
    "          'image_raw': tf.FixedLenFeature([], tf.string),\n",
    "          'mask_raw':tf.FixedLenFeature([], tf.string)})\n",
    "\n",
    " \n",
    "  image = tf.decode_raw(features['image_raw'], tf.uint8)\n",
    "  image = tf.reshape(image, [height,width])\n",
    " \n",
    "  mask=tf.decode_raw(features['mask_raw'], tf.uint8)\n",
    "  mask = tf.reshape(mask, [height,width])\n",
    "\n",
    "  # Convert from [0, 255] -> [-0.5, 0.5] floats.\n",
    "  image = tf.cast(image, tf.float32) * (1. / 255) - 0.5\n",
    "  \n",
    "  #Convert from [0, 255] -> [0, 1] floats.\n",
    "  mask=tf.cast(mask, tf.float32) * (1. / 255)\n",
    "  return image,mask\n",
    "\n",
    "def read_and_decode_pred(filename_queue):\n",
    "  reader = tf.TFRecordReader()\n",
    "  _, serialized_example = reader.read(filename_queue)\n",
    "  features = tf.parse_single_example(\n",
    "      serialized_example,\n",
    "      features={\n",
    "          'image_raw': tf.FixedLenFeature([], tf.string),\n",
    "          'idtest': tf.FixedLenFeature([], tf.int64)})\n",
    "\n",
    " \n",
    "  image = tf.decode_raw(features['image_raw'], tf.uint8)\n",
    "  image = tf.reshape(image, [height,width])\n",
    " \n",
    " # Convert from [0, 255] -> [-0.5, 0.5] floats.\n",
    "  image = tf.cast(image, tf.float32) * (1. / 255) - 0.5\n",
    "  idtest = tf.cast(features['idtest'], tf.int32)\n",
    "  return image,idtest"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To improve the network's performance we make use of data augmentation,when the images are loaded they are randomly flipped vertically and/or horizontaly ,there are already built-in functions to do that in tensorflow, but they take only one image at a time,this is fine for  classficiation problems where the target of an input is just a number (the class) but not for segmentation problems where the target of an input is another image, we need to make sure the input and the target are distorted the same way,the next function is there to do that (credit to \"mtourne\" on stackoverflow) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def image_distortions(image, distortions):\n",
    "    distort_left_right_random = distortions[0]\n",
    "    mirror = tf.less(tf.pack([1.0, distort_left_right_random, 1.0]), 0.5)\n",
    "    image = tf.reverse(image, mirror)\n",
    "    distort_up_down_random = distortions[1]\n",
    "    mirror = tf.less(tf.pack([distort_up_down_random, 1.0, 1.0]), 0.5)\n",
    "    image = tf.reverse(image, mirror)\n",
    "    return image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def distorted_inputs(batch_size, num_epochs):\n",
    "  \"\"\"Construct distorted input for  training\"\"\"\n",
    "  if not num_epochs: num_epochs = None\n",
    "  filename =record_path+ 'trainrecord.tfrecords'\n",
    "\n",
    "  with tf.name_scope('input'):\n",
    "    filename_queue = tf.train.string_input_producer(\n",
    "        [filename], num_epochs=num_epochs)\n",
    "\n",
    "    image,mask= read_and_decode(filename_queue)\n",
    "    \n",
    "  image= tf.reshape(image, [height,width,1])\n",
    "  mask=tf.reshape(mask, [height,width,1])\n",
    "  \n",
    "  distortions = tf.random_uniform([2], 0, 1.0, dtype=tf.float32)\n",
    "  image = image_distortions(image, distortions)\n",
    "  mask = image_distortions(mask, distortions)# image and mask will be distorted the same way\n",
    "\n",
    "    # Shuffle the examples and collect them into batch_size batches.\n",
    "    # (Internally uses a RandomShuffleQueue.)\n",
    "    # We run this in two threads to avoid being a bottleneck.\n",
    "  with tf.device('/cpu:0'):\n",
    "     images, masks = tf.train.shuffle_batch(\n",
    "        [image,mask], batch_size=batch_size, num_threads=2,\n",
    "        capacity=1000 + 3 * batch_size,allow_smaller_final_batch=False,\n",
    "        # Ensures a minimum amount of shuffling of examples.\n",
    "        min_after_dequeue=1000)\n",
    "  tf.image_summary('image', images,max_images=2) #This will show the images on TensorBoard\n",
    "  tf.image_summary('mask', masks,max_images=2)\n",
    "  masks=tf.cast(masks, tf.int64)  \n",
    "  return images,masks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def inputs(train, batch_size, num_epochs):\n",
    "  \"\"\"same as the previous function but images are not distorted (for evaluation)\"\"\"\n",
    "  if train==True:\n",
    "   filename = record_path+'trainrecord.tfrecords'\n",
    "  else:\n",
    "   filename = record_path+'validrecord.tfrecords'\n",
    "      \n",
    "\n",
    "  with tf.name_scope('input'):\n",
    "    filename_queue = tf.train.string_input_producer(\n",
    "        [filename], num_epochs=num_epochs)\n",
    "    image,mask= read_and_decode(filename_queue)\n",
    "  \n",
    "  image=tf.reshape(image, [height,width,1])\n",
    "  mask=tf.reshape(mask, [height,width,1])  \n",
    "\n",
    "  with tf.device('/cpu:0'):\n",
    "     images, masks = tf.train.batch(\n",
    "        [image,mask], batch_size=batch_size, num_threads=2,\n",
    "        capacity=1000 + 3 * batch_size,allow_smaller_final_batch=False)\n",
    "  tf.image_summary('image', images,max_images=2)\n",
    "  tf.image_summary('mask', masks,max_images=2)\n",
    "  masks=tf.cast(masks, tf.int64) \n",
    "  return images,masks\n",
    "\n",
    "def inputs_topred(batch_size, num_epochs):\n",
    "    filename = record_path+'testrecord.tfrecords'\n",
    "    with tf.name_scope('input'):\n",
    "        filename_queue = tf.train.string_input_producer(\n",
    "        [filename], num_epochs=num_epochs)\n",
    "        image,idtest=read_and_decode_pred(filename_queue)\n",
    "        image=tf.reshape(image, [height,width,1])\n",
    "    with tf.device('/cpu:0'):\n",
    "     images,idtests = tf.train.batch(\n",
    "     [image,idtest],batch_size=batch_size, num_threads=1,allow_smaller_final_batch=True,capacity=1000 + 3 * batch_size)\n",
    "     return images,idtests"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we define the network architecture using TensorFlow-Slim "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def inference(data,is_training,batch_norm,scope=''):\n",
    "  prob=0.60\n",
    "  with slim.arg_scope([slim.conv2d], padding='SAME', normalizer_fn=slim.batch_norm,\n",
    "                     normalizer_params={'decay': 0.9997,'is_training':batch_norm,'updates_collections':None,\n",
    "                     'trainable':is_training},\n",
    "                      weights_initializer=slim.initializers.xavier_initializer()\n",
    "                     ,weights_regularizer=slim.l2_regularizer(0.0005)):    \n",
    "    \n",
    "    conv1 = slim.repeat(data, 2, slim.conv2d, 32, [3, 3], scope='conv1')\n",
    "    conv1 = slim.dropout(conv1, prob,is_training=is_training,scope='dropout1')\n",
    "    pool1 = slim.max_pool2d(conv1, [2, 2], scope='pool1') # 64\n",
    "    conv2 = slim.repeat(pool1, 2, slim.conv2d, 64, [3, 3], scope='conv2')\n",
    "    conv2 = slim.dropout(conv2, prob,is_training=is_training,scope='dropout2')\n",
    "    pool2 = slim.max_pool2d(conv2, [2, 2], scope='pool2') # 32\n",
    "    conv3 = slim.repeat(pool2, 2, slim.conv2d, 128, [3, 3], scope='conv3')\n",
    "    conv3 = slim.dropout(conv3, prob,is_training=is_training,scope='dropout3')\n",
    "    pool3 = slim.max_pool2d(conv3, [2, 2], scope='pool3') # 16\n",
    "    conv4 = slim.repeat(pool3, 2, slim.conv2d, 256, [3, 3], scope='conv4')\n",
    "    conv4 = slim.dropout(conv4, prob,is_training=is_training,scope='dropout4')\n",
    "    pool4 = slim.max_pool2d(conv4, [2, 2], scope='pool4')#8\n",
    "    conv5 = slim.repeat(pool4, 2, slim.conv2d, 512, [3, 3], scope='conv5')\n",
    "    conv5 = slim.dropout(conv5, prob,is_training=is_training,scope='dropout5')\n",
    "\n",
    "  with slim.arg_scope([slim.conv2d,slim.conv2d_transpose], padding='SAME',\n",
    "                     normalizer_fn=slim.batch_norm,\n",
    "                     normalizer_params={'decay': 0.9997,'is_training':batch_norm,'updates_collections':None,\n",
    "                    'trainable':is_training},\n",
    "                     weights_initializer=slim.initializers.xavier_initializer(),\n",
    "                     weights_regularizer=slim.l2_regularizer(0.0005)):\n",
    "    \n",
    "    deconv1=slim.conv2d_transpose(conv5,512,stride=2,kernel_size=2) #16\n",
    "    deconv1 = slim.dropout(deconv1, prob,is_training=is_training,scope='d_dropout1')\n",
    "    concat1=tf.concat(3,[conv4,deconv1],name='concat1')\n",
    "    conv6 = slim.repeat(concat1, 2, slim.conv2d, 256, [3, 3], scope='conv6')  \n",
    "    conv6 = slim.dropout(conv6, prob,is_training=is_training,scope='dropout6')\n",
    "    deconv2=slim.conv2d_transpose(conv6,256,stride=2,kernel_size=2) #32\n",
    "    deconv2 = slim.dropout(deconv2, prob,is_training=is_training,scope='d_dropout2')\n",
    "    concat2=tf.concat(3,[conv3,deconv2],name='concat2')   \n",
    "    conv7 = slim.repeat(concat2, 2, slim.conv2d,128, [3, 3], scope='conv7')\n",
    "    conv7 = slim.dropout(conv7, prob,is_training=is_training,scope='dropout7')\n",
    "    deconv3=slim.conv2d_transpose(conv7,128,stride=2,kernel_size=2) #64\n",
    "    deconv3 = slim.dropout(deconv3, prob,is_training=is_training,scope='d_dropout3')\n",
    "    concat3=tf.concat(3,[conv2,deconv3],name='concat3')\n",
    "    conv8 = slim.repeat(concat3, 2, slim.conv2d, 64, [3, 3], scope='conv8')\n",
    "    conv8 = slim.dropout(conv8, prob,is_training=is_training,scope='dropout8')\n",
    "    deconv4=slim.conv2d_transpose(conv8,64,stride=2,kernel_size=2) #128\n",
    "    deconv4 = slim.dropout(deconv4, prob,is_training=is_training,scope='d_dropout4')\n",
    "    concat4=tf.concat(3,[conv1,deconv4],name='concat4')\n",
    "    conv9 = slim.repeat(concat4, 2, slim.conv2d, 32, [3, 3], scope='conv9')\n",
    "    conv9 = slim.dropout(conv9, prob,is_training=is_training,scope='dropout9')\n",
    "    conv1x1=slim.conv2d(conv9, 2, [1, 1],  activation_fn=tf.nn.sigmoid,scope='conv1x1')\n",
    "  \n",
    "  return conv1x1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def dice_coef(y_true, y_pred):\n",
    "    \"\"\"Compute the mean(batch-wise) of dice coefficients\"\"\"\n",
    "    md=tf.constant(0.0)\n",
    "    y_true=tf.cast(y_true, tf.float32)\n",
    "    y_true_f = slim.flatten(y_true)\n",
    "    y_pred_f = slim.flatten(y_pred)\n",
    "    for i in xrange(batch_size):\n",
    "     union=tf.reduce_sum(y_true_f[i]) + tf.reduce_sum(y_pred_f[i]) \n",
    "     md = tf.cond(tf.equal(union,0.0), lambda: tf.add(md,1.0), \n",
    "                 lambda: tf.add(md,tf.div(2.*tf.reduce_sum(tf.mul(y_true_f[i],y_pred_f[i])),union))) \n",
    "\n",
    "    return tf.div(md,batch_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now start the training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "num_epochs=55\n",
    "batch_size=24\n",
    "\n",
    "def train():\n",
    "  \n",
    "  with tf.Graph().as_default():\n",
    "  \n",
    "    images, masks =distorted_inputs(batch_size=batch_size,num_epochs=num_epochs)\n",
    "    masks=tf.reshape(masks,[batch_size,height,width], name=None)\n",
    "    logits=inference(images,is_training=True,batch_norm=True)\n",
    "    \n",
    "    \n",
    "    pred=tf.argmax(logits, dimension=3)\n",
    "    pred=tf.reshape(pred, [batch_size,height,width,1])\n",
    "    pred=tf.cast(pred, tf.float32)\n",
    "    tf.image_summary('predicted', pred,max_images=2) # This will display predicted images in tensorboard\n",
    "    \n",
    "    class_weight = tf.constant([0.75,0.25])\n",
    "    weighted_logits =tf.mul(logits, class_weight) \n",
    "    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(weighted_logits, masks)\n",
    "    \n",
    "    loss = tf.reduce_mean(cross_entropy, name='xentropy_mean')\n",
    "    dice=dice_coef(masks,pred)\n",
    "    \n",
    "    tf.scalar_summary('loss', loss)#tensorboard will keep track of loss and dice during training\n",
    "    tf.scalar_summary('dice', dice)\n",
    "\n",
    "\n",
    "    train_op=tf.train.AdamOptimizer(learning_rate=0.0005, beta1=0.9, beta2=0.999,\n",
    "                                  epsilon=1e-08, use_locking=False).minimize(loss)\n",
    "    \n",
    "\n",
    "\n",
    "\n",
    "    # Build the summary operation based on the TF collection of Summaries.\n",
    "    tf.add_to_collection('train_op', train_op)\n",
    "    saver = tf.train.Saver(tf.all_variables())\n",
    "    summary_op = tf.merge_all_summaries()\n",
    "    # Build an initialization operation to run below.\n",
    "    init = tf.group(tf.initialize_all_variables(),\n",
    "                    tf.initialize_local_variables())\n",
    "    \n",
    "\n",
    "\n",
    "    # Start running operations on the Graph.\n",
    "    sess = tf.Session(config=tf.ConfigProto(\n",
    "        log_device_placement=False))\n",
    "    sess.run(init)\n",
    "   \n",
    "    coord = tf.train.Coordinator()\n",
    "    threads = tf.train.start_queue_runners(sess=sess, coord=coord)\n",
    "    summary_writer = tf.train.SummaryWriter(tensorb_path, sess.graph)\n",
    "    try: \n",
    "      step = 0\n",
    "      while not coord.should_stop():\n",
    "          \n",
    "    \n",
    "        start_time = time.time()\n",
    "        _, loss_value,dice_value= sess.run([train_op, loss,dice])\n",
    "        \n",
    "        duration = time.time() - start_time\n",
    "        if step % 10 == 0:\n",
    "          \n",
    "          #print('Step %d: loss = %.10f (%.3f sec) dice=%.2f' % (step, loss_value,duration,dice_value))\n",
    "          summary_str = sess.run(summary_op)\n",
    "          summary_writer.add_summary(summary_str, step)\n",
    "        if step % 1000 == 0 :\n",
    "         saver.save(sess,ckpt_path+ 'model.ckpt',step) #makes a checkpoint every 1000 step       \n",
    "        if step % 100==0:\n",
    "         print('Step %d: loss = %.5f dice=%.2f (%.3f sec) ' % (step, loss_value,dice_value,duration))   \n",
    "        step += 1\n",
    "    except tf.errors.OutOfRangeError:\n",
    "      print('Done training for %d epochs, %d steps.' % (num_epochs, step))\n",
    "    finally:\n",
    "      # When done, ask the threads to stop.\n",
    "      coord.request_stop()\n",
    "\n",
    "    # Wait for threads to finish.\n",
    "    coord.join(threads)\n",
    "    sess.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Step 0: loss = 0.58497 dice=0.02 (4.481 sec) \n",
      "Step 100: loss = 0.56066 dice=0.07 (0.836 sec) \n",
      "Step 200: loss = 0.55534 dice=0.11 (0.839 sec) \n",
      "Step 300: loss = 0.54780 dice=0.34 (0.839 sec) \n",
      "Step 400: loss = 0.54458 dice=0.11 (0.836 sec) \n",
      "Step 500: loss = 0.53950 dice=0.22 (0.838 sec) \n",
      "Step 600: loss = 0.53494 dice=0.26 (0.900 sec) \n",
      "Step 700: loss = 0.52935 dice=0.48 (0.838 sec) \n",
      "Step 800: loss = 0.52610 dice=0.17 (0.835 sec) \n",
      "Step 900: loss = 0.52185 dice=0.43 (0.839 sec) \n",
      "Step 1000: loss = 0.51742 dice=0.45 (0.855 sec) \n",
      "Step 1100: loss = 0.51179 dice=0.53 (0.848 sec) \n",
      "Step 1200: loss = 0.50924 dice=0.38 (0.874 sec) \n",
      "Step 1300: loss = 0.50423 dice=0.66 (0.875 sec) \n",
      "Step 1400: loss = 0.50147 dice=0.49 (0.874 sec) \n",
      "Step 1500: loss = 0.49834 dice=0.35 (0.870 sec) \n",
      "Step 1600: loss = 0.49523 dice=0.55 (0.876 sec) \n",
      "Step 1700: loss = 0.49250 dice=0.50 (0.867 sec) \n",
      "Step 1800: loss = 0.48814 dice=0.38 (0.846 sec) \n",
      "Step 1900: loss = 0.48444 dice=0.48 (0.885 sec) \n",
      "Step 2000: loss = 0.48213 dice=0.53 (0.879 sec) \n",
      "Step 2100: loss = 0.48019 dice=0.46 (0.880 sec) \n",
      "Step 2200: loss = 0.47566 dice=0.45 (0.878 sec) \n",
      "Step 2300: loss = 0.47323 dice=0.42 (0.870 sec) \n",
      "Step 2400: loss = 0.46983 dice=0.59 (0.869 sec) \n",
      "Step 2500: loss = 0.46858 dice=0.42 (0.869 sec) \n",
      "Step 2600: loss = 0.46498 dice=0.66 (0.870 sec) \n",
      "Step 2700: loss = 0.46464 dice=0.37 (0.898 sec) \n",
      "Step 2800: loss = 0.46104 dice=0.56 (0.877 sec) \n",
      "Step 2900: loss = 0.45765 dice=0.68 (0.903 sec) \n",
      "Step 3000: loss = 0.45658 dice=0.68 (0.875 sec) \n",
      "Step 3100: loss = 0.45475 dice=0.72 (0.844 sec) \n",
      "Step 3200: loss = 0.45206 dice=0.73 (0.899 sec) \n",
      "Step 3300: loss = 0.45073 dice=0.49 (0.865 sec) \n",
      "Step 3400: loss = 0.44864 dice=0.60 (0.836 sec) \n",
      "Step 3500: loss = 0.44656 dice=0.64 (0.854 sec) \n",
      "Step 3600: loss = 0.44496 dice=0.48 (0.840 sec) \n",
      "Step 3700: loss = 0.44373 dice=0.41 (0.896 sec) \n",
      "Step 3800: loss = 0.44313 dice=0.46 (0.838 sec) \n",
      "Step 3900: loss = 0.44026 dice=0.41 (0.838 sec) \n",
      "Step 4000: loss = 0.43856 dice=0.64 (0.888 sec) \n",
      "Step 4100: loss = 0.43677 dice=0.63 (0.839 sec) \n",
      "Step 4200: loss = 0.43528 dice=0.61 (0.866 sec) \n",
      "Step 4300: loss = 0.43578 dice=0.67 (0.862 sec) \n",
      "Step 4400: loss = 0.43277 dice=0.63 (0.864 sec) \n",
      "Step 4500: loss = 0.43097 dice=0.71 (0.866 sec) \n",
      "Step 4600: loss = 0.43203 dice=0.60 (0.856 sec) \n",
      "Step 4700: loss = 0.42877 dice=0.53 (0.858 sec) \n",
      "Step 4800: loss = 0.42763 dice=0.56 (0.835 sec) \n",
      "Step 4900: loss = 0.42808 dice=0.61 (0.836 sec) \n",
      "Step 5000: loss = 0.42449 dice=0.69 (0.911 sec) \n",
      "Step 5100: loss = 0.42497 dice=0.73 (0.860 sec) \n",
      "Step 5200: loss = 0.42346 dice=0.54 (0.838 sec) \n",
      "Step 5300: loss = 0.42095 dice=0.66 (0.862 sec) \n",
      "Step 5400: loss = 0.42287 dice=0.70 (0.860 sec) \n",
      "Step 5500: loss = 0.42105 dice=0.56 (0.837 sec) \n",
      "Step 5600: loss = 0.42034 dice=0.61 (0.890 sec) \n",
      "Step 5700: loss = 0.41996 dice=0.71 (0.920 sec) \n",
      "Step 5800: loss = 0.41804 dice=0.58 (0.863 sec) \n",
      "Step 5900: loss = 0.41692 dice=0.66 (0.885 sec) \n",
      "Step 6000: loss = 0.41535 dice=0.73 (0.874 sec) \n",
      "Step 6100: loss = 0.41520 dice=0.65 (0.861 sec) \n",
      "Step 6200: loss = 0.41727 dice=0.67 (0.908 sec) \n",
      "Step 6300: loss = 0.41376 dice=0.75 (0.880 sec) \n",
      "Step 6400: loss = 0.41419 dice=0.70 (0.861 sec) \n",
      "Step 6500: loss = 0.41265 dice=0.72 (0.835 sec) \n",
      "Step 6600: loss = 0.41191 dice=0.81 (0.860 sec) \n",
      "Step 6700: loss = 0.41057 dice=0.63 (0.859 sec) \n",
      "Step 6800: loss = 0.41130 dice=0.70 (0.861 sec) \n",
      "Step 6900: loss = 0.41377 dice=0.59 (0.862 sec) \n",
      "Step 7000: loss = 0.40998 dice=0.64 (0.837 sec) \n",
      "Step 7100: loss = 0.40925 dice=0.76 (0.895 sec) \n",
      "Step 7200: loss = 0.40849 dice=0.77 (0.837 sec) \n",
      "Step 7300: loss = 0.40823 dice=0.77 (0.853 sec) \n",
      "Step 7400: loss = 0.41045 dice=0.75 (0.838 sec) \n",
      "Step 7500: loss = 0.40713 dice=0.59 (0.836 sec) \n",
      "Step 7600: loss = 0.40675 dice=0.83 (0.838 sec) \n",
      "Step 7700: loss = 0.40760 dice=0.57 (0.891 sec) \n",
      "Step 7800: loss = 0.40604 dice=0.73 (0.886 sec) \n",
      "Step 7900: loss = 0.40677 dice=0.55 (0.891 sec) \n",
      "Step 8000: loss = 0.40588 dice=0.65 (0.889 sec) \n",
      "Step 8100: loss = 0.40351 dice=0.75 (0.863 sec) \n",
      "Step 8200: loss = 0.40394 dice=0.80 (0.869 sec) \n",
      "Step 8300: loss = 0.40339 dice=0.75 (0.863 sec) \n",
      "Step 8400: loss = 0.40267 dice=0.80 (0.863 sec) \n",
      "Step 8500: loss = 0.40280 dice=0.73 (0.861 sec) \n",
      "Step 8600: loss = 0.40189 dice=0.81 (0.864 sec) \n",
      "Step 8700: loss = 0.40169 dice=0.72 (0.860 sec) \n",
      "Step 8800: loss = 0.40418 dice=0.58 (0.860 sec) \n",
      "Step 8900: loss = 0.40204 dice=0.72 (0.837 sec) \n",
      "Step 9000: loss = 0.40381 dice=0.69 (0.879 sec) \n",
      "Step 9100: loss = 0.40356 dice=0.74 (0.876 sec) \n",
      "Step 9200: loss = 0.39959 dice=0.75 (0.911 sec) \n",
      "Step 9300: loss = 0.40178 dice=0.64 (0.864 sec) \n",
      "Step 9400: loss = 0.40319 dice=0.73 (0.859 sec) \n",
      "Done training for 55 epochs, 9482 steps.\n"
     ]
    }
   ],
   "source": [
    "train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def eval_once(saver,loss,dice,summary_writer, summary_op):\n",
    "  num_examples=1082\n",
    "  with tf.Session() as sess:\n",
    "    ckpt = tf.train.get_checkpoint_state(ckpt_path)\n",
    "    saver.restore(sess, ckpt.model_checkpoint_path)\n",
    "    sess.run(tf.initialize_local_variables()) \n",
    "    # Start the queue runners.\n",
    "    coord = tf.train.Coordinator()\n",
    "    try:\n",
    "      threads = []\n",
    "      for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS):\n",
    "        threads.extend(qr.create_threads(sess, coord=coord, daemon=True,\n",
    "                                         start=True))\n",
    "\n",
    "      num_iter=int(math.ceil(num_examples /batch_size))\n",
    "      step=0\n",
    "      mdice=[]\n",
    "      while step < num_iter and not coord.should_stop():\n",
    "        loss_value,dice_value = sess.run([loss,dice])\n",
    "        mdice.append(dice_value)\n",
    "        if step % 5==0: \n",
    "         summary_str = sess.run(summary_op)\n",
    "         summary_writer.add_summary(summary_str, step)\n",
    "   \n",
    "        \n",
    "        if step==num_iter-1:\n",
    "         print('total dice = %.2f '% (np.mean(mdice))) \n",
    "        step+=1\n",
    "\n",
    "\n",
    "    except Exception as e:  \n",
    "      coord.request_stop(e)\n",
    "\n",
    "    coord.request_stop()\n",
    "    coord.join(threads, stop_grace_period_secs=10)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "batch_size=24\n",
    "def evaluate():\n",
    "  \"\"\"Evaluate model on Dataset for a number of steps.\"\"\"\n",
    "  with tf.Graph().as_default() as g:\n",
    "    \n",
    "    images, masks=inputs(train=False,batch_size=batch_size,num_epochs=None)\n",
    "    masks=tf.reshape(masks,[batch_size,height,width], name=None)\n",
    "    logits=inference(images,is_training=False,batch_norm=True)\n",
    "    \n",
    "    pred=tf.argmax(logits, dimension=3)\n",
    "    pred=tf.reshape(pred, [batch_size,height,width,1])\n",
    "    pred=tf.cast(pred, tf.float32)\n",
    "    tf.image_summary('predicted', pred,max_images=2)\n",
    "    \n",
    "    class_weight = tf.constant([0.75,0.25])\n",
    "    weighted_logits = tf.mul(logits, class_weight) \n",
    "    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(weighted_logits, masks)\n",
    "    \n",
    "    loss = tf.reduce_mean(cross_entropy, name='xentropy_mean')\n",
    "    dice=dice_coef(masks,pred)\n",
    "    \n",
    "    tf.scalar_summary('loss', loss)\n",
    "    tf.scalar_summary('dice', dice)\n",
    "    \n",
    "   \n",
    "\n",
    "     #Build the summary operation based on the TF collection of Summaries.\n",
    "    summary_op = tf.merge_all_summaries()\n",
    "        \n",
    "    saver = tf.train.Saver(tf.all_variables())\n",
    "   \n",
    "    summary_writer = tf.train.SummaryWriter(tensorb_path,g)\n",
    "    while True:\n",
    "      eval_once(saver,loss,dice,summary_writer,summary_op)\n",
    "      #pred_once(saver,pred,summary_writer,summary_op)\n",
    "      break\n",
    "      "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total dice = 0.65 \n"
     ]
    }
   ],
   "source": [
    "evaluate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def pred_once(saver,pred,imds):\n",
    "   num_examples=5508\n",
    "   predset=[]\n",
    "   with tf.Session() as sess:\n",
    "    ckpt = tf.train.get_checkpoint_state(ckpt_path)\n",
    "    saver.restore(sess, ckpt.model_checkpoint_path)\n",
    "    sess.run(tf.initialize_local_variables()) \n",
    "    coord = tf.train.Coordinator()\n",
    "    try:\n",
    "      threads = []\n",
    "      for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS):\n",
    "        threads.extend(qr.create_threads(sess, coord=coord, daemon=True,\n",
    "                                         start=True))\n",
    "\n",
    "      num_iter=int(math.ceil(num_examples /batch_size))\n",
    "      step=0\n",
    "      \n",
    "      while step < num_iter and not coord.should_stop():\n",
    "        predictions,idx = sess.run([pred,imds])\n",
    "        predictions = np.array(predictions)\n",
    "        \n",
    "        if step % 100==0:\n",
    "            print'Predicting step %d out of %d ' %(step,num_iter)\n",
    "        for i in xrange (batch_size):\n",
    "            predset.append((predictions[i],idx[i]))\n",
    "        step+=1\n",
    "      print(\"saving to disk\")\n",
    "      np.save(\"/media/florian/My Passport/git/pred\", predset, allow_pickle=True)\n",
    "    except Exception as e:  \n",
    "      coord.request_stop(e)\n",
    "\n",
    "    coord.request_stop()\n",
    "    coord.join(threads, stop_grace_period_secs=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "batch_size=12\n",
    "def pred():\n",
    "  \"\"\"Evaluate model on Dataset for a number of steps.\"\"\"\n",
    "  with tf.Graph().as_default() as g:\n",
    "    \n",
    "    images,imds=inputs_topred(batch_size=batch_size,num_epochs=1)\n",
    "    logits=inference(images,is_training=False,batch_norm=True)\n",
    "    pred=tf.argmax(logits, dimension=3)\n",
    "    saver = tf.train.Saver(tf.all_variables())\n",
    "   \n",
    "    while True:\n",
    "      pred_once(saver,pred,imds)\n",
    "      break\n",
    "      time.sleep(60)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicting step 0 out of 459 \n",
      "Predicting step 100 out of 459 \n",
      "Predicting step 200 out of 459 \n",
      "Predicting step 300 out of 459 \n",
      "Predicting step 400 out of 459 \n",
      "saving to disk\n"
     ]
    }
   ],
   "source": [
    "pred()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's load our predictions see and what they look like."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "predictions=np.load(\"/media/florian/My Passport/git/pred.npy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "del predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fd2c8081910>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAcMAAAFkCAYAAABcn1CIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nZqXnYmhmZqX3/wEOLXZCxyWUqAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fd2de54fa10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.subplot(221)\n",
    "plt.imshow(predictions[6][0],cmap=\"gray\")\n",
    "plt.subplot(222)\n",
    "plt.imshow(predictions[60][0],cmap=\"gray\")\n",
    "plt.subplot(223)\n",
    "plt.imshow(predictions[15][0],cmap=\"gray\")\n",
    "plt.subplot(224)\n",
    "plt.imshow(predictions[3605][0],cmap=\"gray\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "But the original size of the images was 420x580, so let's resize our predictions "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "fpred=[]\n",
    "for i in xrange(len(predictions)):\n",
    " img=imresize((predictions[i][0]), (420,580))/255\n",
    " fpred.append((img,predictions[i][1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": false
   },
   "source": [
    "Then we do a little of post processing , any predicted image that has less than 3500 white pixels is set to a black image (no nerve)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "allblack=np.zeros((420,580))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "postpro=[]\n",
    "for i in xrange(len(fpred)):\n",
    " if np.sum(fpred[i][0])<3500:\n",
    "  postpro.append((allblack,fpred[i][1]))\n",
    " else:\n",
    "  postpro.append((fpred[i][0],fpred[i][1]))     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def run_length_enc(label):\n",
    "    '''run length encoder for submission'''\n",
    "    from itertools import chain\n",
    "    x = label.transpose().flatten()\n",
    "    y = np.where(x > 0)[0]\n",
    "    if len(y) < 10:  # consider as empty\n",
    "        return ''\n",
    "    z = np.where(np.diff(y) > 1)[0]\n",
    "    start = np.insert(y[z+1], 0, y[0])\n",
    "    end = np.append(y[z], y[-1])\n",
    "    length = end - start\n",
    "    res = [[s+1, l+1] for s, l in zip(list(start), list(length))]\n",
    "    res = list(chain.from_iterable(res))\n",
    "    return ' '.join([str(r) for r in res])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Done\n"
     ]
    }
   ],
   "source": [
    "encpred=[]\n",
    "for i in xrange(len(fpred)):\n",
    " encpred.append((postpro[i][1],run_length_enc(postpro[i][0])))\n",
    "print('Done')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "with open('submissionUNS.csv','wb') as csvfile:\n",
    " writer=csv.writer(csvfile)\n",
    " length_list=len(encpred)\n",
    " writer.writerow([\"img\",\"pixels\"])\n",
    " for i in xrange(length_list):    \n",
    "    writer.writerow([encpred[i][0],encpred[i][1]])"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [default]",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}