keras-u-net-starter.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import random\n",
    "import warnings\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from tqdm import tqdm\n",
    "from itertools import chain\n",
    "from skimage.io import imread, imshow, imread_collection, concatenate_images\n",
    "from skimage.transform import resize\n",
    "from skimage.morphology import label\n",
    "\n",
    "from keras.models import Model, load_model\n",
    "from keras.layers import Input\n",
    "from keras.layers.core import Lambda\n",
    "from keras.layers.convolutional import Conv2D, Conv2DTranspose\n",
    "from keras.layers.pooling import MaxPooling2D\n",
    "from keras.layers.merge import concatenate\n",
    "from keras.callbacks import EarlyStopping, ModelCheckpoint\n",
    "from keras import backend as K\n",
    "\n",
    "import tensorflow as tf\n",
    "\n",
    "# Set some parameters\n",
    "IMG_WIDTH = 128\n",
    "IMG_HEIGHT = 128\n",
    "IMG_CHANNELS = 3\n",
    "TRAIN_PATH = 'data/stage1_train/'\n",
    "TEST_PATH = 'data/stage1_test/'\n",
    "\n",
    "warnings.filterwarnings('ignore', category=UserWarning, module='skimage')\n",
    "seed = 42\n",
    "random.seed = seed\n",
    "np.random.seed = seed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.1.3\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "print(keras.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "670\n",
      "65\n"
     ]
    }
   ],
   "source": [
    "# Get train and test IDs\n",
    "train_ids = next(os.walk(TRAIN_PATH))[1]\n",
    "test_ids = next(os.walk(TEST_PATH))[1]\n",
    "print(len(train_ids))\n",
    "print(len(test_ids))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/chen/workspace/git_examples/kaggle-nuclei-challenge\n"
     ]
    }
   ],
   "source": [
    "current_dir = os.getcwd()\n",
    "print(current_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Getting and resizing train images and masks ... \n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 670/670 [00:50<00:00, 13.25it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Getting and resizing test images ... \n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "100%|██████████| 65/65 [00:00<00:00, 143.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Done!\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Get and resize train images and masks\n",
    "X_train = np.zeros((len(train_ids), IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS), dtype=np.uint8)\n",
    "Y_train = np.zeros((len(train_ids), IMG_HEIGHT, IMG_WIDTH, 1), dtype=np.bool)\n",
    "print('Getting and resizing train images and masks ... ')\n",
    "sys.stdout.flush()\n",
    "for n, id_ in tqdm(enumerate(train_ids), total=len(train_ids)):\n",
    "    path = TRAIN_PATH + id_\n",
    "    img = imread(path + '/images/' + id_ + '.png')[:,:,:IMG_CHANNELS]\n",
    "    img = resize(img, (IMG_HEIGHT, IMG_WIDTH), mode='constant', preserve_range=True)\n",
    "    X_train[n] = img\n",
    "    mask = np.zeros((IMG_HEIGHT, IMG_WIDTH, 1), dtype=np.bool)\n",
    "    for mask_file in next(os.walk(path + '/masks/'))[2]:\n",
    "        mask_ = imread(path + '/masks/' + mask_file)\n",
    "        mask_ = np.expand_dims(resize(mask_, (IMG_HEIGHT, IMG_WIDTH), mode='constant', \n",
    "                                      preserve_range=True), axis=-1)\n",
    "        mask = np.maximum(mask, mask_)\n",
    "    Y_train[n] = mask\n",
    "\n",
    "# Get and resize test images\n",
    "X_test = np.zeros((len(test_ids), IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS), dtype=np.uint8)\n",
    "sizes_test = []\n",
    "print('Getting and resizing test images ... ')\n",
    "sys.stdout.flush()\n",
    "for n, id_ in tqdm(enumerate(test_ids), total=len(test_ids)):\n",
    "    path = TEST_PATH + id_\n",
    "    img = imread(path + '/images/' + id_ + '.png')[:,:,:IMG_CHANNELS]\n",
    "    sizes_test.append([img.shape[0], img.shape[1]])\n",
    "    img = resize(img, (IMG_HEIGHT, IMG_WIDTH), mode='constant', preserve_range=True)\n",
    "    X_test[n] = img\n",
    "\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nqKKiYq74n2rhpTBGHZD6AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9a052bfd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9a02d6ed0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check if training data looks all right\n",
    "ix = random.randint(0, len(train_ids))\n",
    "imshow(X_train[ix])\n",
    "plt.show()\n",
    "imshow(np.squeeze(Y_train[ix]))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define IoU metric\n",
    "def mean_iou(y_true, y_pred):\n",
    "    prec = []\n",
    "    for t in np.arange(0.5, 1.0, 0.05):\n",
    "        y_pred_ = tf.to_int32(y_pred > t)\n",
    "        score, up_opt = tf.metrics.mean_iou(y_true, y_pred_, 2)\n",
    "        K.get_session().run(tf.local_variables_initializer())\n",
    "        with tf.control_dependencies([up_opt]):\n",
    "            score = tf.identity(score)\n",
    "        prec.append(score)\n",
    "    return K.mean(K.stack(prec), axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 128, 128, 3)  0                                            \n",
      "__________________________________________________________________________________________________\n",
      "lambda_1 (Lambda)               (None, 128, 128, 3)  0           input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)               (None, 128, 128, 8)  224         lambda_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)               (None, 128, 128, 8)  584         conv2d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2D)  (None, 64, 64, 8)    0           conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_3 (Conv2D)               (None, 64, 64, 16)   1168        max_pooling2d_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_4 (Conv2D)               (None, 64, 64, 16)   2320        conv2d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2D)  (None, 32, 32, 16)   0           conv2d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_5 (Conv2D)               (None, 32, 32, 32)   4640        max_pooling2d_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)               (None, 32, 32, 32)   9248        conv2d_5[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2D)  (None, 16, 16, 32)   0           conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_7 (Conv2D)               (None, 16, 16, 64)   18496       max_pooling2d_3[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_8 (Conv2D)               (None, 16, 16, 64)   36928       conv2d_7[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)  (None, 8, 8, 64)     0           conv2d_8[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_9 (Conv2D)               (None, 8, 8, 128)    73856       max_pooling2d_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_10 (Conv2D)              (None, 8, 8, 128)    147584      conv2d_9[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_1 (Conv2DTrans (None, 16, 16, 64)   32832       conv2d_10[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 16, 16, 128)  0           conv2d_transpose_1[0][0]         \n",
      "                                                                 conv2d_8[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_11 (Conv2D)              (None, 16, 16, 64)   73792       concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_12 (Conv2D)              (None, 16, 16, 64)   36928       conv2d_11[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_2 (Conv2DTrans (None, 32, 32, 32)   8224        conv2d_12[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, 32, 32, 64)   0           conv2d_transpose_2[0][0]         \n",
      "                                                                 conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_13 (Conv2D)              (None, 32, 32, 32)   18464       concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_14 (Conv2D)              (None, 32, 32, 32)   9248        conv2d_13[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_3 (Conv2DTrans (None, 64, 64, 16)   2064        conv2d_14[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_3 (Concatenate)     (None, 64, 64, 32)   0           conv2d_transpose_3[0][0]         \n",
      "                                                                 conv2d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_15 (Conv2D)              (None, 64, 64, 16)   4624        concatenate_3[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_16 (Conv2D)              (None, 64, 64, 16)   2320        conv2d_15[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_4 (Conv2DTrans (None, 128, 128, 8)  520         conv2d_16[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_4 (Concatenate)     (None, 128, 128, 16) 0           conv2d_transpose_4[0][0]         \n",
      "                                                                 conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_17 (Conv2D)              (None, 128, 128, 8)  1160        concatenate_4[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_18 (Conv2D)              (None, 128, 128, 8)  584         conv2d_17[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_19 (Conv2D)              (None, 128, 128, 1)  9           conv2d_18[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 485,817\n",
      "Trainable params: 485,817\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Build U-Net model\n",
    "inputs = Input((IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS))\n",
    "s = Lambda(lambda x: x / 255) (inputs)\n",
    "\n",
    "c1 = Conv2D(8, (3, 3), activation='relu', padding='same') (s)\n",
    "c1 = Conv2D(8, (3, 3), activation='relu', padding='same') (c1)\n",
    "p1 = MaxPooling2D((2, 2)) (c1)\n",
    "\n",
    "c2 = Conv2D(16, (3, 3), activation='relu', padding='same') (p1)\n",
    "c2 = Conv2D(16, (3, 3), activation='relu', padding='same') (c2)\n",
    "p2 = MaxPooling2D((2, 2)) (c2)\n",
    "\n",
    "c3 = Conv2D(32, (3, 3), activation='relu', padding='same') (p2)\n",
    "c3 = Conv2D(32, (3, 3), activation='relu', padding='same') (c3)\n",
    "p3 = MaxPooling2D((2, 2)) (c3)\n",
    "\n",
    "c4 = Conv2D(64, (3, 3), activation='relu', padding='same') (p3)\n",
    "c4 = Conv2D(64, (3, 3), activation='relu', padding='same') (c4)\n",
    "p4 = MaxPooling2D(pool_size=(2, 2)) (c4)\n",
    "\n",
    "c5 = Conv2D(128, (3, 3), activation='relu', padding='same') (p4)\n",
    "c5 = Conv2D(128, (3, 3), activation='relu', padding='same') (c5)\n",
    "\n",
    "u6 = Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same') (c5)\n",
    "u6 = concatenate([u6, c4])\n",
    "c6 = Conv2D(64, (3, 3), activation='relu', padding='same') (u6)\n",
    "c6 = Conv2D(64, (3, 3), activation='relu', padding='same') (c6)\n",
    "\n",
    "u7 = Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same') (c6)\n",
    "u7 = concatenate([u7, c3])\n",
    "c7 = Conv2D(32, (3, 3), activation='relu', padding='same') (u7)\n",
    "c7 = Conv2D(32, (3, 3), activation='relu', padding='same') (c7)\n",
    "\n",
    "u8 = Conv2DTranspose(16, (2, 2), strides=(2, 2), padding='same') (c7)\n",
    "u8 = concatenate([u8, c2])\n",
    "c8 = Conv2D(16, (3, 3), activation='relu', padding='same') (u8)\n",
    "c8 = Conv2D(16, (3, 3), activation='relu', padding='same') (c8)\n",
    "\n",
    "u9 = Conv2DTranspose(8, (2, 2), strides=(2, 2), padding='same') (c8)\n",
    "u9 = concatenate([u9, c1], axis=3)\n",
    "c9 = Conv2D(8, (3, 3), activation='relu', padding='same') (u9)\n",
    "c9 = Conv2D(8, (3, 3), activation='relu', padding='same') (c9)\n",
    "\n",
    "outputs = Conv2D(1, (1, 1), activation='sigmoid') (c9)\n",
    "\n",
    "model = Model(inputs=[inputs], outputs=[outputs])\n",
    "model.compile(optimizer='adam', loss='binary_crossentropy', metrics=[mean_iou])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 603 samples, validate on 67 samples\n",
      "Epoch 1/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.5084 - mean_iou: 0.4205\n",
      "Epoch 00001: val_loss improved from inf to 0.41817, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 3s 5ms/step - loss: 0.5062 - mean_iou: 0.4207 - val_loss: 0.4182 - val_mean_iou: 0.4241\n",
      "Epoch 2/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.3050 - mean_iou: 0.4234\n",
      "Epoch 00002: val_loss improved from 0.41817 to 0.25198, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.3032 - mean_iou: 0.4234 - val_loss: 0.2520 - val_mean_iou: 0.4263\n",
      "Epoch 3/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1727 - mean_iou: 0.4600\n",
      "Epoch 00003: val_loss improved from 0.25198 to 0.17279, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1714 - mean_iou: 0.4611 - val_loss: 0.1728 - val_mean_iou: 0.4992\n",
      "Epoch 4/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1394 - mean_iou: 0.5308\n",
      "Epoch 00004: val_loss improved from 0.17279 to 0.16141, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1373 - mean_iou: 0.5316 - val_loss: 0.1614 - val_mean_iou: 0.5569\n",
      "Epoch 5/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1199 - mean_iou: 0.5763\n",
      "Epoch 00005: val_loss improved from 0.16141 to 0.13402, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1190 - mean_iou: 0.5768 - val_loss: 0.1340 - val_mean_iou: 0.5967\n",
      "Epoch 6/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1403 - mean_iou: 0.6096\n",
      "Epoch 00006: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1388 - mean_iou: 0.6100 - val_loss: 0.1452 - val_mean_iou: 0.6219\n",
      "Epoch 7/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1104 - mean_iou: 0.6322\n",
      "Epoch 00007: val_loss improved from 0.13402 to 0.12479, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1107 - mean_iou: 0.6325 - val_loss: 0.1248 - val_mean_iou: 0.6433\n",
      "Epoch 8/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.1008 - mean_iou: 0.6529\n",
      "Epoch 00008: val_loss improved from 0.12479 to 0.11549, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.1007 - mean_iou: 0.6532 - val_loss: 0.1155 - val_mean_iou: 0.6629\n",
      "Epoch 9/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0969 - mean_iou: 0.6713\n",
      "Epoch 00009: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0963 - mean_iou: 0.6715 - val_loss: 0.1163 - val_mean_iou: 0.6795\n",
      "Epoch 10/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0938 - mean_iou: 0.6856\n",
      "Epoch 00010: val_loss improved from 0.11549 to 0.11009, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 2ms/step - loss: 0.0935 - mean_iou: 0.6858 - val_loss: 0.1101 - val_mean_iou: 0.6927\n",
      "Epoch 11/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0897 - mean_iou: 0.6987\n",
      "Epoch 00011: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0907 - mean_iou: 0.6989 - val_loss: 0.1187 - val_mean_iou: 0.7047\n",
      "Epoch 12/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0895 - mean_iou: 0.7097\n",
      "Epoch 00012: val_loss improved from 0.11009 to 0.10635, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0900 - mean_iou: 0.7099 - val_loss: 0.1063 - val_mean_iou: 0.7147\n",
      "Epoch 13/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0847 - mean_iou: 0.7189\n",
      "Epoch 00013: val_loss improved from 0.10635 to 0.10542, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0840 - mean_iou: 0.7190 - val_loss: 0.1054 - val_mean_iou: 0.7238\n",
      "Epoch 14/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0836 - mean_iou: 0.7279\n",
      "Epoch 00014: val_loss improved from 0.10542 to 0.10171, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0830 - mean_iou: 0.7281 - val_loss: 0.1017 - val_mean_iou: 0.7320\n",
      "Epoch 15/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0837 - mean_iou: 0.7354\n",
      "Epoch 00015: val_loss improved from 0.10171 to 0.10166, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0836 - mean_iou: 0.7355 - val_loss: 0.1017 - val_mean_iou: 0.7390\n",
      "Epoch 16/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0795 - mean_iou: 0.7423\n",
      "Epoch 00016: val_loss improved from 0.10166 to 0.09879, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0794 - mean_iou: 0.7424 - val_loss: 0.0988 - val_mean_iou: 0.7457\n",
      "Epoch 17/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0772 - mean_iou: 0.7486\n",
      "Epoch 00017: val_loss improved from 0.09879 to 0.09876, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0772 - mean_iou: 0.7487 - val_loss: 0.0988 - val_mean_iou: 0.7517\n",
      "Epoch 18/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0771 - mean_iou: 0.7545\n",
      "Epoch 00018: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0771 - mean_iou: 0.7545 - val_loss: 0.1009 - val_mean_iou: 0.7571\n",
      "Epoch 19/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0782 - mean_iou: 0.7596\n",
      "Epoch 00019: val_loss improved from 0.09876 to 0.09450, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0779 - mean_iou: 0.7597 - val_loss: 0.0945 - val_mean_iou: 0.7620\n",
      "Epoch 20/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0756 - mean_iou: 0.7643\n",
      "Epoch 00020: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0753 - mean_iou: 0.7643 - val_loss: 0.1002 - val_mean_iou: 0.7665\n",
      "Epoch 21/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0780 - mean_iou: 0.7684\n",
      "Epoch 00021: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0769 - mean_iou: 0.7685 - val_loss: 0.0949 - val_mean_iou: 0.7704\n",
      "Epoch 22/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0760 - mean_iou: 0.7724\n",
      "Epoch 00022: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0754 - mean_iou: 0.7725 - val_loss: 0.0971 - val_mean_iou: 0.7742\n",
      "Epoch 23/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0742 - mean_iou: 0.7759\n",
      "Epoch 00023: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0743 - mean_iou: 0.7759 - val_loss: 0.1024 - val_mean_iou: 0.7776\n",
      "Epoch 24/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0720 - mean_iou: 0.7791\n",
      "Epoch 00024: val_loss improved from 0.09450 to 0.09207, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0719 - mean_iou: 0.7791 - val_loss: 0.0921 - val_mean_iou: 0.7809\n",
      "Epoch 25/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0723 - mean_iou: 0.7824\n",
      "Epoch 00025: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0722 - mean_iou: 0.7824 - val_loss: 0.0929 - val_mean_iou: 0.7840\n",
      "Epoch 26/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0710 - mean_iou: 0.7854\n",
      "Epoch 00026: val_loss improved from 0.09207 to 0.09192, saving model to model-dsbowl2018-1.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0709 - mean_iou: 0.7854 - val_loss: 0.0919 - val_mean_iou: 0.7869\n",
      "Epoch 27/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0687 - mean_iou: 0.7882\n",
      "Epoch 00027: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0689 - mean_iou: 0.7883 - val_loss: 0.0928 - val_mean_iou: 0.7897\n",
      "Epoch 28/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0716 - mean_iou: 0.7908\n",
      "Epoch 00028: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0709 - mean_iou: 0.7908 - val_loss: 0.0949 - val_mean_iou: 0.7921\n",
      "Epoch 29/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0694 - mean_iou: 0.7934\n",
      "Epoch 00029: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0691 - mean_iou: 0.7934 - val_loss: 0.0938 - val_mean_iou: 0.7946\n",
      "Epoch 30/30\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0709 - mean_iou: 0.7958\n",
      "Epoch 00030: val_loss did not improve\n",
      "603/603 [==============================] - 2s 2ms/step - loss: 0.0705 - mean_iou: 0.7958 - val_loss: 0.0940 - val_mean_iou: 0.7969\n"
     ]
    }
   ],
   "source": [
    "earlystopper = EarlyStopping(patience=5, verbose=1)\n",
    "checkpointer = ModelCheckpoint('model-dsbowl2018-1.h5', verbose=1, save_best_only=True)\n",
    "results = model.fit(X_train, Y_train, validation_split=0.1, batch_size=8, epochs=30, \n",
    "                    callbacks=[earlystopper, checkpointer])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "603/603 [==============================] - 0s 736us/step\n",
      "67/67 [==============================] - 0s 398us/step\n",
      "65/65 [==============================] - 0s 658us/step\n"
     ]
    }
   ],
   "source": [
    "# Predict on train, val and test\n",
    "model = load_model('model-dsbowl2018-1.h5', custom_objects={'mean_iou': mean_iou})\n",
    "preds_train = model.predict(X_train[:int(X_train.shape[0]*0.9)], verbose=1)\n",
    "preds_val = model.predict(X_train[int(X_train.shape[0]*0.9):], verbose=1)\n",
    "preds_test = model.predict(X_test, verbose=1)\n",
    "\n",
    "# Threshold predictions\n",
    "preds_train_t = (preds_train > 0.5).astype(np.uint8)\n",
    "preds_val_t = (preds_val > 0.5).astype(np.uint8)\n",
    "preds_test_t = (preds_test > 0.5).astype(np.uint8)\n",
    "\n",
    "# Create list of upsampled test masks\n",
    "preds_test_upsampled = []\n",
    "for i in range(len(preds_test)):\n",
    "    preds_test_upsampled.append(resize(np.squeeze(preds_test[i]), \n",
    "                                       (sizes_test[i][0], sizes_test[i][1]), \n",
    "                                       mode='constant', preserve_range=True))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nKHwScjgcg8InIYfDMSh8EnI4HIPi/wO6hvz3FD432wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff94105db50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9387540d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUQAAAEYCAYAAAAkpo9KAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff93875ee10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Perform a sanity check on some random training samples\n",
    "ix = random.randint(0, len(preds_train_t))\n",
    "imshow(X_train[ix])\n",
    "plt.show()\n",
    "imshow(np.squeeze(Y_train[ix]))\n",
    "plt.show()\n",
    "imshow(np.squeeze(preds_train_t[ix]))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n8gG1vd4AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9a02c5790>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASEAAAEYCAYAAAATaEB+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nVv8HqYiYkc3r94AAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9384925d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUQAAAEYCAYAAAAkpo9KAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAF81JREFUeJzt3X+s1fV9x/Hny4vCtFVEOofAKktZO+paNQQ1LisTG9AZ\n6LLGQLsWLRtZoq2t7VqYi3YuTera1bYJtb2rVNo4mVJbiaOlSjVmy6Rcp6ECUu9wCojiD7Smpgrc\n9/74fg58ud57z/fc8+N7zj2vR/IN5/s93/M97/uF++bz+6uIwMzM4LiyAzAzaxdOiGZmiROimVni\nhGhmljghmpklTohmZokTopl1JEmrJe2X9Pgw70vSNyX1S9oq6dxq12xaQpS0QNLOFMyKZn2PmXWt\n24AFI7x/CTAzbcuBW6pdsCkJUVIPsCoFNAtYImlWM77LzLpTRDwEvDzCKYuA70fmYWCipCkjXXNc\nIwPMmQP0R8QuAElrU3Dbhzr5BI2PCZzUpFDMrAyvceDFiHgHwPw/Oyleevlw4c8+svWNbcBvc4d6\nI6K3xhCmArtz+3vSsX3DfaBZCXGoQM7LnyBpOVkxlgmcyHma16RQzKwM98e6pyuvX3r5ML/Y+PuF\nP9sz5cnfRsTspgQ2gmYlxKpStu8FOFmTPKHabAwLYICBVn/tXmB6bn9aOjasZnWq1ByImY1lweEY\nKLw1yHrg46m3+Xzg1YgYtroMzSshbgFmSppBlggXAx9p0neZWZvLSoiNrQhKugOYC0yWtAe4ATge\nICK+DWwALgX6gdeBK6tdsykJMSIOSboa2Aj0AKsjYlszvsvMOkOjq8wRsaTK+wFcVcs1m9aGGBEb\nyDK0mXW5IDjcAWuvltapYmbdpdFV5mZwQjSzpgvgsBOimVnGJUQzM7IS4kG3IZqZpU4VlxDNzICA\nw+2fD50Qzaz5soHZ7c8J0cxaQBxGZQdRlROimTVdAAOuMpuZZVxCNDOjMjDbCdHMDICBcEI0M3MJ\n0cysIhCHO+Cpx06IZtYSrjKbmeEqs5lZjjgcrjKbmaWpe06IZmaAq8xmZgBEuMpsZnbEgEuIZmaV\nXmaXEM3McC+zmVniXmYzs5zDnqliZua5zGZmR2SPIW3/dNP+EZpZxwvkKrOZWYU7VczMgAg6YtjN\nqCOUNF3SA5K2S9om6Zp0fJKk+yQ9mf48tXHhmllnEgM1bGWpJ2UfAj4bEbOA84GrJM0CVgCbImIm\nsCntm1kXC7ISYtGtLKOuMkfEPmBfev2apB3AVGARMDedtgZ4EPhCXVGaWcfrmmE3ks4EzgE2A6en\nZAnwHHD6MJ9ZDiwHmMCJjQjDzNpUoO54hICktwE/BD4dEb+Wjv7QERGSYqjPRUQv0AtwsiYNeY6Z\njR1jvoQo6XiyZHh7RNydDj8vaUpE7JM0Bdhfb5Bm1tkCGBjjvcwCbgV2RMTXcm+tB5am10uBe0Yf\nnpmNDeJwDVtZ6ikhXgh8DPilpMfSsb8HvgzcKWkZ8DRweX0hmlmn65QSYj29zP8Jw6byeaO9rpmN\nTZ3wTJX2T9lm1vEixEAcV3grQtICSTsl9Ut6y3hnSb+fJo88KmmrpEurXdNT98ysJRo54FpSD7AK\n+CCwB9giaX1EbM+d9g/AnRFxS5o0sgE4c6TruoRoZk2XrZjd0Kl7c4D+iNgVEW8Ca8kmhQz+2pPT\n61OAZ6td1CVEM2uBmp+pMllSX26/N41drpgK7M7t7wHOG3SNLwI/k/RJ4CTg4mpf6oRoZk2X9TLX\n1KnyYkTMrvNrlwC3RcS/SLoA+IGksyJiYLgPOCGaWUs0eKbKXmB6bn9aOpa3DFgAEBH/LWkCMJkR\nJou4DdHMmq4yl7noVsAWYKakGZJOABaTTQrJe4Y0BFDSHwETgBdGuqhLiGbWEo1cMTsiDkm6GtgI\n9ACrI2KbpBuBvohYD3wW+FdJnyGrtV8RESOum+CEaGZNl62Y3diB2RGxgWwoTf7Y9bnX28lm1BXm\nhGhmLdEVy3+ZmVUTiIPRU3YYVTkhmlnTjWLYTSmcELvAxmcfG/H9+Wec3aJIrHtpbK92Y2ZWizKf\npleUE+IYVq1kOPg8lxStWZrRy9wMTohjVNFkaNYqrjKbmdFFT90zMyvCbYhmZnjYjZnZMdyGaGYG\nUHwVm1I5IY5R8884u3BPc9HhNkNdz0N1rIjKIwTanROimbWES4hWqkaV3kYqaXpQtxXhThUzsxwn\nROtonu1ijeKB2WZmOe5UsZbLl+rcrmdtI1xlNjMDuqhTRVIP0AfsjYjLJM0A1gKnAY8AH4uIN+v9\nnm5VTzve4M+6xGhl6oSE2Ii5NNcAO3L7NwE3R8S7gANkD4s2sy7WhOcyN0VdCVHSNODPge+mfQEX\nAevSKWuAD9XzHd1q47OPuZfXxpQIFd7KUm+V+evA54G3p/3TgFci4lDa3wNMHeqDkpYDywEmcGKd\nYZhZuxvTvcySLgP2R8QjkubW+vmI6AV6AU7WpBhtHN2u0i5YpDRZ66ySatd2m6QVFV3Qy3whsFDS\npcAE4GTgG8BESeNSKXEasLf+MM2ss4nDA2N4+a+IWAmsBEglxM9FxEcl3QV8mKyneSlwTwPitEEG\nl84G7zdyZRqXBK0RymwbLKoZKfsLwLWS+snaFG9twneYWQepjENs917mhgzMjogHgQfT613AnEZc\nt5sN135Xa/ufWVuIrB2x3Xmmipm1xJjuZbbWaERJz2sWWtmCzmhDdEIcwwZXt50YrTxe/svM7Ai3\nIVphtU7Tq6eU55KilcFVZjMzstKhE6JVNdoFHKqV8rwwhLUbtyGamSWd0IbY/pMLbVTcPmjtphuW\n/zIzqyooN9EV5YTYhVx6tDJ0QI3ZVWYza4FofJVZ0gJJOyX1S1oxzDmXS9ouaZukf6t2TZcQS1Zk\n2a6RjHS+S4LWVhpYREwPt1sFfJBsZf4tktZHxPbcOTPJlii8MCIOSPrdatd1CdHMWqLBJcQ5QH9E\n7EpP9VwLLBp0zt8AqyLiQPb9sb/aRV1CbDPVSnW1lCA9I8XaSYOH3UwFduf29wDnDTrnDwEk/RfQ\nA3wxIn460kWdEM2s6Uax2s1kSX25/d70HKZajANmAnPJHmfykKQ/johXRvqAmVlzBVBbQnwxImaP\n8P5eYHpuf6jnN+0BNkfEQeApSb8iS5Bbhruo2xDNrCUiim8FbAFmSpoh6QRgMbB+0Dk/JisdImky\nWRV610gXdULsMPPPONttgtaZooat2qWyp3peDWwEdgB3RsQ2STdKWphO2wi8JGk78ADwdxHx0kjX\ndZXZzFpAxEBjZ6pExAZgw6Bj1+deB3Bt2gpxQuxQRR5Q75KktQ0v/2VmltMBc/ecEDucS4HWOVxC\nNDPLuIRoZpY4IZqZMZqB2aVwQjSzluiERwg4IZpZazghmpklHVBlrmvqnqSJktZJekLSDkkXSJok\n6T5JT6Y/T21UsGbWuRTFt7LUO5f5G8BPI+I9wPvJ5hSuADZFxExgU9o3s25WyzzmTkyIkk4B/hS4\nFSAi3kzrjC0C1qTT1gAfqjdIM+t0yqrMRbeS1FNCnAG8AHxP0qOSvivpJOD0iNiXznkOOH2oD0ta\nLqlPUt9B3qgjDDPrCGO5hEjWIXMucEtEnAP8hkHV47TaxJA/XkT0RsTsiJh9POPrCMPMOsIYT4h7\ngD0RsTntryNLkM9LmgKQ/qz6YBcz6wJjOSFGxHPAbknvTofmAdvJVq1dmo4tBe6pK0Iz63yVmSpt\n3oZY7zjETwK3pyW8dwFXkiXZOyUtA54GLq/zO8xsDChzOE1RdSXEiHgMGOpBMPPqua6ZjUEdkBD9\nTBUzs8RT98ysJcZ8ldnMrLAOmMvshGhmzVfycJqinBDNrCU0UHYE1TkhmllruIRoZpY4IZqZlb/O\nYVFOiGbWGu5lNjNLXEI0M8u4ytzGNj772DH78884u6RIzLqEE6KZGeBOlfY0uGQ4+LhLimZN4oTY\neZwYzZrECdHMLNMJVWavh2hmlriEaGat0QElRCdEM2s+9zKbmeU4IbafSu/xcMNvzKxJnBDNzEB0\nRpW563qZNz77mEuHZmWIGrYCJC2QtFNSv6QVI5z3l5JC0lCPTD5G1yVEMytBHF0TschWjaQeYBVw\nCTALWCJp1hDnvR24BthcJEwnRDNrjcaWEOcA/RGxKyLeBNYCi4Y475+Am4DfFrmoE6KZtUZjE+JU\nYHduf086doSkc4HpEfEfRUN0p4qZtUSNnSqTJfXl9nsjorfwd0nHAV8DrqjlS50Qzaz5AqjtMaQv\nRsRInSB7gem5/WnpWMXbgbOAByUB/B6wXtLCiMgn2mM4IZpZSzR42M0WYKakGWSJcDHwkcqbEfEq\nMPnId0sPAp8bKRmC2xDNrFUa2IYYEYeAq4GNwA7gzojYJulGSQtHG2JdJURJnwH+muxH+CVwJTCF\nrMfnNOAR4GOpF8jMulijB2ZHxAZgw6Bj1w9z7twi1xx1CVHSVOBTwOyIOAvoISu23gTcHBHvAg4A\ny0b7HWY2hjR4YHYz1NuGOA74HUkHgROBfcBFHK3LrwG+CNxS5/c0zHBzmb1CtlkTlZzoihp1CTEi\n9gJfBZ4hS4SvklWRX0n1exhibFCFpOWS+iT1HeSN0YZhZh1ANW5lGXUJUdKpZCPDZwCvAHcBC4p+\nPo0p6gU4WZNa/n+HS4RmLdYBJcR6qswXA09FxAsAku4GLgQmShqXSomDxwaZWZca66vdPAOcL+lE\nZSMf5wHbgQeAD6dzlgL31BeimY0JHdCpUk8b4mZgHfA/ZENujiOrAn8BuFZSP9nQm1sbEKeZdboO\nSIh19TJHxA3ADYMO7yJbicLMLONnqpiZ5TghmpllXEI0M6twQjQzy7iEaGZdpzIttmdK7mCHTN1z\nQjSz1nBCNLNuc3RabP+RY53yXGYnRDNrDSdEM7OMov0zohOimTWfO1XMzI5yG6KZWaLaHkNaCidE\nM2sNlxDNzPBqN2Zmx3BCNDPzwGwzs2N5HKKZWcYlRDMz8MBsM7M8j0M0M6twCdHMLOM2RDMzSG2I\n7Z8RnRDNrCVcQjQzq3BCNDPzTBUzs6Mi3IZoZlbhEqKZWUUHJMTjqp0gabWk/ZIezx2bJOk+SU+m\nP09NxyXpm5L6JW2VdG4zgzezzqEovpWlakIEbgMWDDq2AtgUETOBTWkf4BJgZtqWA7c0Jkwz62gB\nDETxrSRVE2JEPAS8POjwImBNer0G+FDu+Pcj8zAwUdKURgVrZh0sathKUqSEOJTTI2Jfev0ccHp6\nPRXYnTtvTzr2FpKWS+qT1HeQN0YZhpl1irFSZR5RRIwqp0dEb0TMjojZxzO+3jDMrN1Vht4U2QqQ\ntEDSztRnsWKI96+VtD31Z2yS9M5q1xxtQny+UhVOf+5Px/cC03PnTUvHzKybRbb8V9GtGkk9wCqy\nfotZwBJJswad9igwOyLeB6wD/rnadUebENcDS9PrpcA9ueMfT73N5wOv5qrWZtalspkqUXgrYA7Q\nHxG7IuJNYC1ZH8YREfFARLyedh8mK6CNqOo4REl3AHOByZL2ADcAXwbulLQMeBq4PJ2+AbgU6Ade\nB66s/nOZWVeobYHYyZL6cvu9EdGb2x+qv+K8Ea63DPhJtS+tmhAjYskwb80b4twArqp2TTPrPgVL\nfhUvRsTshnyv9FfAbOAD1c71TBUza77GD6cp1F8h6WLgOuADEVF1OEvdvcxmZtXV0MNcrCS5BZgp\naYakE4DFZH0YR0g6B/gOsDAi9g9xjbdwCdHMWqKR4wsj4pCkq4GNQA+wOiK2SboR6IuI9cBXgLcB\nd0kCeCYiFo50XSdEM2uNBi//FREbyDpy88euz72+uNZrOiGaWfOFH0NqZnaUF4g1M0vaPx86IZpZ\na9Q4DrEUTohm1hpOiGZmpAViyw6iOidEM2s6UXjRhlI5IZpZazghmpklTohmZrgN0cwsz22IZmYV\nTohmZnBk+a8254RoZs0XOCGamR3hThUzs4wG2j8jOiGaWfMFMOAqs5kZ7lQxM8tzQjQzS5wQzcxw\nG6KZ2VEB4V5mM7OMq8xmZrjKbGZ2jA4oIR5X7QRJqyXtl/R47thXJD0haaukH0mamHtvpaR+STsl\nzW9W4GbWYSKKbyWpmhCB24AFg47dB5wVEe8DfgWsBJA0C1gMvDd95luSehoWrZl1qBqSYTsnxIh4\nCHh50LGfRcShtPswMC29XgSsjYg3IuIpoB+Y08B4zawTBTAwUHwrSZESYjWfAH6SXk8Fdufe25OO\nvYWk5ZL6JPUd5I0GhGFmba0DSoh1dapIug44BNxe62cjohfoBThZk9q/tdXM6tMBnSqjToiSrgAu\nA+ZFHPlJ9wLTc6dNS8fMrKtFRwy7GVWVWdIC4PPAwoh4PffWemCxpPGSZgAzgV/UH6aZdbSAiIHC\nW1mqlhAl3QHMBSZL2gPcQNarPB64TxLAwxHxtxGxTdKdwHayqvRVEXG4WcGbWQfpgBJi1YQYEUuG\nOHzrCOd/CfhSPUGZ2Rg0ltsQzcwKiyh1OE1RTohm1houIZqZZcIlRDMz8DNVzMwqvPyXmVkmgDjc\n/iPwGjGX2cxsZJEeIVB0K0DSgrTMYL+kFUO8P17Sv6f3N0s6s9o1nRDNrCViIApv1aRlBVcBlwCz\ngCVp+cG8ZcCBiHgXcDNwU7XrOiGaWWs0toQ4B+iPiF0R8Sawlmz5wbxFwJr0eh0wT2lq3XDaog3x\nNQ68eH+s+w3wYtmxDGMy7Rmb46pdu8Y2FuN6Z+XFaxzYeH+sm1zDZydI6svt96YVsiqGWmrwvEHX\nOHJORByS9CpwGiP8PG2RECPiHZL6ImJ22bEMpV1jc1y1a9fYxnpcETF41f225CqzmXWiIksNHjlH\n0jjgFOClkS7qhGhmnWgLMFPSDEknkD3Laf2gc9YDS9PrDwM/z63dOqS2qDInvdVPKU27xua4ateu\nsTmuGqQ2wauBjUAPsDotP3gj0BcR68lW5fqBpH6y50ItrnZdVUmYZmZdw1VmM7PECdHMLGmLhFht\nCk4L45gu6QFJ2yVtk3RNOj5J0n2Snkx/nlpSfD2SHpV0b9qfkaYk9acpSieUFNdESeskPSFph6QL\n2uGeSfpM+nt8XNIdkiaUdc8krZa0X9LjuWND3iNlvpli3Crp3BbH9ZX0d7lV0o8kTcy9tzLFtVPS\n/GbFVZbSE2LBKTitcgj4bETMAs4HrkqxrAA2RcRMYFPaL8M1wI7c/k3AzWlq0gGyqUpl+Abw04h4\nD/B+shhLvWeSpgKfAmZHxFlkDe+LKe+e3QYMHos33D26hOwBbTOB5cAtLY7rPuCsiHgf8CuyZyiR\nfhcWA+9Nn/lW+v0dOyKi1A24ANiY218JrCw7rhTLPcAHgZ3AlHRsCrCzhFimkf3SXATcC4hsxP24\noe5jC+M6BXiK1EGXO17qPePoLIVJZKMp7gXml3nPgDOBx6vdI+A7wJKhzmtFXIPe+wvg9vT6mN9N\nsh7eC1r9b66ZW+klRIaegjO1pFiOSCtjnANsBk6PiH3preeA00sI6etkj36tTPQ8DXglIg6l/bLu\n2wzgBeB7qTr/XUknUfI9i4i9wFeBZ4B9wKvAI7THPasY7h610+/EJ4CfpNftFFdTtENCbDuS3gb8\nEPh0RPw6/15k/zW2dKySpMuA/RHxSCu/t6BxwLnALRFxDvAbBlWPS7pnp5JN7p8BnAGcxFurhm2j\njHtUjaTryJqRbi87llZph4RYZApOy0g6niwZ3h4Rd6fDz0uakt6fAuxvcVgXAgsl/R/Zqh4XkbXb\nTUxTkqC8+7YH2BMRm9P+OrIEWfY9uxh4KiJeiIiDwN1k97Ed7lnFcPeo9N8JSVcAlwEfTcm6LeJq\ntnZIiEWm4LREWhroVmBHRHwt91Z+CtBSsrbFlomIlRExLSLOJLs/P4+IjwIPkE1JKiWuFNtzwG5J\n706H5gHbKfmekVWVz5d0Yvp7rcRV+j3LGe4erQc+nnqbzwdezVWtm07SArLmmYUR8fqgeBcrW3h1\nBlmnzy9aFVdLlN2Imf7zuZSsN+t/getKjONPyKotW4HH0nYpWXvdJuBJ4H5gUokxzgXuTa//gOwf\nZD9wFzC+pJjOBvrSffsxcGo73DPgH4EngMeBHwDjy7pnwB1kbZkHyUrVy4a7R2QdZqvS78MvyXrK\nWxlXP1lbYeV34Nu5869Lce0ELinj31szN0/dMzNL2qHKbGbWFpwQzcwSJ0Qzs8QJ0cwscUI0M0uc\nEM3MEidEM7Pk/wFN1Ov2tjycWwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9384b9d50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Perform a sanity check on some random validation samples\n",
    "ix = random.randint(0, len(preds_val_t))\n",
    "imshow(X_train[int(X_train.shape[0]*0.9):][ix])\n",
    "plt.show()\n",
    "imshow(np.squeeze(Y_train[int(Y_train.shape[0]*0.9):][ix]))\n",
    "plt.show()\n",
    "imshow(np.squeeze(preds_val_t[ix]))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Run-length encoding stolen from https://www.kaggle.com/rakhlin/fast-run-length-encoding-python\n",
    "def rle_encoding(x):\n",
    "    dots = np.where(x.T.flatten() == 1)[0]\n",
    "    run_lengths = []\n",
    "    prev = -2\n",
    "    for b in dots:\n",
    "        if (b>prev+1): run_lengths.extend((b + 1, 0))\n",
    "        run_lengths[-1] += 1\n",
    "        prev = b\n",
    "    return run_lengths\n",
    "\n",
    "def prob_to_rles(x, cutoff=0.5):\n",
    "    lab_img = label(x > cutoff)\n",
    "    for i in range(1, lab_img.max() + 1):\n",
    "        yield rle_encoding(lab_img == i)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "new_test_ids = []\n",
    "rles = []\n",
    "for n, id_ in enumerate(test_ids):\n",
    "    rle = list(prob_to_rles(preds_test_upsampled[n]))\n",
    "    rles.extend(rle)\n",
    "    new_test_ids.extend([id_] * len(rle))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create submission DataFrame\n",
    "sub = pd.DataFrame()\n",
    "sub['ImageId'] = new_test_ids\n",
    "sub['EncodedPixels'] = pd.Series(rles).apply(lambda x: ' '.join(str(y) for y in x))\n",
    "sub.to_csv('sub-dsbowl2018-1.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 603 samples, validate on 67 samples\n",
      "Epoch 1/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0728 - mean_iou: 0.8476\n",
      "Epoch 00001: val_loss improved from inf to 0.09326, saving model to model-dsbowl2018-2.h5\n",
      "603/603 [==============================] - 3s 4ms/step - loss: 0.0728 - mean_iou: 0.8480 - val_loss: 0.0933 - val_mean_iou: 0.8592\n",
      "Epoch 2/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0703 - mean_iou: 0.8581\n",
      "Epoch 00002: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0700 - mean_iou: 0.8582 - val_loss: 0.0970 - val_mean_iou: 0.8600\n",
      "Epoch 3/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0696 - mean_iou: 0.8610\n",
      "Epoch 00003: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0691 - mean_iou: 0.8610 - val_loss: 0.0988 - val_mean_iou: 0.8617\n",
      "Epoch 4/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0694 - mean_iou: 0.8610\n",
      "Epoch 00004: val_loss improved from 0.09326 to 0.09073, saving model to model-dsbowl2018-2.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0704 - mean_iou: 0.8611 - val_loss: 0.0907 - val_mean_iou: 0.8616\n",
      "Epoch 5/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0691 - mean_iou: 0.8616\n",
      "Epoch 00005: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0690 - mean_iou: 0.8617 - val_loss: 0.0926 - val_mean_iou: 0.8618\n",
      "Epoch 6/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0724 - mean_iou: 0.8610\n",
      "Epoch 00006: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0721 - mean_iou: 0.8611 - val_loss: 0.0997 - val_mean_iou: 0.8615\n",
      "Epoch 7/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0702 - mean_iou: 0.8621\n",
      "Epoch 00007: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0704 - mean_iou: 0.8621 - val_loss: 0.0960 - val_mean_iou: 0.8620\n",
      "Epoch 8/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0677 - mean_iou: 0.8617\n",
      "Epoch 00008: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0681 - mean_iou: 0.8617 - val_loss: 0.1091 - val_mean_iou: 0.8618\n",
      "Epoch 9/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0695 - mean_iou: 0.8611\n",
      "Epoch 00009: val_loss improved from 0.09073 to 0.08894, saving model to model-dsbowl2018-2.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0693 - mean_iou: 0.8611 - val_loss: 0.0889 - val_mean_iou: 0.8613\n",
      "Epoch 10/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0678 - mean_iou: 0.8617\n",
      "Epoch 00010: val_loss improved from 0.08894 to 0.08889, saving model to model-dsbowl2018-2.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0670 - mean_iou: 0.8617 - val_loss: 0.0889 - val_mean_iou: 0.8619\n",
      "Epoch 11/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0650 - mean_iou: 0.8623\n",
      "Epoch 00011: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0657 - mean_iou: 0.8623 - val_loss: 0.0892 - val_mean_iou: 0.8625\n",
      "Epoch 12/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0645 - mean_iou: 0.8626\n",
      "Epoch 00012: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0645 - mean_iou: 0.8626 - val_loss: 0.0919 - val_mean_iou: 0.8632\n",
      "Epoch 13/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0657 - mean_iou: 0.8635\n",
      "Epoch 00013: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0655 - mean_iou: 0.8635 - val_loss: 0.0912 - val_mean_iou: 0.8637\n",
      "Epoch 14/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0645 - mean_iou: 0.8638\n",
      "Epoch 00014: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0643 - mean_iou: 0.8639 - val_loss: 0.1016 - val_mean_iou: 0.8643\n",
      "Epoch 15/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0652 - mean_iou: 0.8645\n",
      "Epoch 00015: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0658 - mean_iou: 0.8645 - val_loss: 0.0941 - val_mean_iou: 0.8648\n",
      "Epoch 16/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0643 - mean_iou: 0.8648\n",
      "Epoch 00016: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0637 - mean_iou: 0.8648 - val_loss: 0.0922 - val_mean_iou: 0.8651\n",
      "Epoch 17/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0624 - mean_iou: 0.8652\n",
      "Epoch 00017: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0634 - mean_iou: 0.8652 - val_loss: 0.0910 - val_mean_iou: 0.8655\n",
      "Epoch 18/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0631 - mean_iou: 0.8656\n",
      "Epoch 00018: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0624 - mean_iou: 0.8656 - val_loss: 0.0914 - val_mean_iou: 0.8659\n",
      "Epoch 19/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0629 - mean_iou: 0.8661\n",
      "Epoch 00019: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0630 - mean_iou: 0.8661 - val_loss: 0.0903 - val_mean_iou: 0.8664\n",
      "Epoch 20/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0661 - mean_iou: 0.8664\n",
      "Epoch 00020: val_loss did not improve\n",
      "603/603 [==============================] - 2s 2ms/step - loss: 0.0656 - mean_iou: 0.8664 - val_loss: 0.0939 - val_mean_iou: 0.8666\n",
      "Epoch 21/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0636 - mean_iou: 0.8668\n",
      "Epoch 00021: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0641 - mean_iou: 0.8668 - val_loss: 0.0898 - val_mean_iou: 0.8669\n",
      "Epoch 22/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0602 - mean_iou: 0.8672\n",
      "Epoch 00022: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0604 - mean_iou: 0.8672 - val_loss: 0.0896 - val_mean_iou: 0.8674\n",
      "Epoch 23/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0605 - mean_iou: 0.8677\n",
      "Epoch 00023: val_loss improved from 0.08889 to 0.08832, saving model to model-dsbowl2018-2.h5\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0602 - mean_iou: 0.8677 - val_loss: 0.0883 - val_mean_iou: 0.8679\n",
      "Epoch 24/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0595 - mean_iou: 0.8681\n",
      "Epoch 00024: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0592 - mean_iou: 0.8681 - val_loss: 0.0895 - val_mean_iou: 0.8684\n",
      "Epoch 25/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0579 - mean_iou: 0.8686\n",
      "Epoch 00025: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0576 - mean_iou: 0.8686 - val_loss: 0.0884 - val_mean_iou: 0.8689\n",
      "Epoch 26/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0579 - mean_iou: 0.8691\n",
      "Epoch 00026: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0581 - mean_iou: 0.8691 - val_loss: 0.0888 - val_mean_iou: 0.8694\n",
      "Epoch 27/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0561 - mean_iou: 0.8697\n",
      "Epoch 00027: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0560 - mean_iou: 0.8697 - val_loss: 0.0915 - val_mean_iou: 0.8700\n",
      "Epoch 28/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0539 - mean_iou: 0.8702\n",
      "Epoch 00028: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0547 - mean_iou: 0.8702 - val_loss: 0.0920 - val_mean_iou: 0.8705\n",
      "Epoch 29/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0541 - mean_iou: 0.8708\n",
      "Epoch 00029: val_loss did not improve\n",
      "603/603 [==============================] - 2s 2ms/step - loss: 0.0542 - mean_iou: 0.8708 - val_loss: 0.0902 - val_mean_iou: 0.8711\n",
      "Epoch 30/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0558 - mean_iou: 0.8714\n",
      "Epoch 00030: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0556 - mean_iou: 0.8715 - val_loss: 0.0932 - val_mean_iou: 0.8716\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 31/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0519 - mean_iou: 0.8719\n",
      "Epoch 00031: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0519 - mean_iou: 0.8719 - val_loss: 0.0951 - val_mean_iou: 0.8723\n",
      "Epoch 32/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0512 - mean_iou: 0.8726\n",
      "Epoch 00032: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0513 - mean_iou: 0.8727 - val_loss: 0.0912 - val_mean_iou: 0.8729\n",
      "Epoch 33/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0521 - mean_iou: 0.8732\n",
      "Epoch 00033: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0518 - mean_iou: 0.8732 - val_loss: 0.0979 - val_mean_iou: 0.8735\n",
      "Epoch 34/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0502 - mean_iou: 0.8738\n",
      "Epoch 00034: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0503 - mean_iou: 0.8738 - val_loss: 0.0953 - val_mean_iou: 0.8742\n",
      "Epoch 35/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0492 - mean_iou: 0.8745\n",
      "Epoch 00035: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0498 - mean_iou: 0.8745 - val_loss: 0.0930 - val_mean_iou: 0.8748\n",
      "Epoch 36/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0478 - mean_iou: 0.8751\n",
      "Epoch 00036: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0479 - mean_iou: 0.8751 - val_loss: 0.0971 - val_mean_iou: 0.8755\n",
      "Epoch 37/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0458 - mean_iou: 0.8758\n",
      "Epoch 00037: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0467 - mean_iou: 0.8758 - val_loss: 0.1008 - val_mean_iou: 0.8762\n",
      "Epoch 38/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0497 - mean_iou: 0.8763\n",
      "Epoch 00038: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0494 - mean_iou: 0.8763 - val_loss: 0.0993 - val_mean_iou: 0.8766\n",
      "Epoch 39/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0457 - mean_iou: 0.8769\n",
      "Epoch 00039: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0459 - mean_iou: 0.8769 - val_loss: 0.1004 - val_mean_iou: 0.8773\n",
      "Epoch 40/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0438 - mean_iou: 0.8776\n",
      "Epoch 00040: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0440 - mean_iou: 0.8776 - val_loss: 0.1030 - val_mean_iou: 0.8780\n",
      "Epoch 41/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0438 - mean_iou: 0.8783\n",
      "Epoch 00041: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0440 - mean_iou: 0.8783 - val_loss: 0.0989 - val_mean_iou: 0.8787\n",
      "Epoch 42/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0430 - mean_iou: 0.8790\n",
      "Epoch 00042: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0435 - mean_iou: 0.8790 - val_loss: 0.1034 - val_mean_iou: 0.8793\n",
      "Epoch 43/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0427 - mean_iou: 0.8796\n",
      "Epoch 00043: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0426 - mean_iou: 0.8796 - val_loss: 0.1026 - val_mean_iou: 0.8800\n",
      "Epoch 44/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0419 - mean_iou: 0.8803\n",
      "Epoch 00044: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0418 - mean_iou: 0.8803 - val_loss: 0.1093 - val_mean_iou: 0.8806\n",
      "Epoch 45/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0435 - mean_iou: 0.8809\n",
      "Epoch 00045: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0437 - mean_iou: 0.8809 - val_loss: 0.1014 - val_mean_iou: 0.8812\n",
      "Epoch 46/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0413 - mean_iou: 0.8815\n",
      "Epoch 00046: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0412 - mean_iou: 0.8815 - val_loss: 0.1028 - val_mean_iou: 0.8819\n",
      "Epoch 47/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0392 - mean_iou: 0.8822\n",
      "Epoch 00047: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0393 - mean_iou: 0.8822 - val_loss: 0.1059 - val_mean_iou: 0.8826\n",
      "Epoch 48/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0383 - mean_iou: 0.8829\n",
      "Epoch 00048: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0383 - mean_iou: 0.8829 - val_loss: 0.1063 - val_mean_iou: 0.8833\n",
      "Epoch 49/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0382 - mean_iou: 0.8836\n",
      "Epoch 00049: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0384 - mean_iou: 0.8836 - val_loss: 0.1097 - val_mean_iou: 0.8839\n",
      "Epoch 50/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0378 - mean_iou: 0.8842\n",
      "Epoch 00050: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0385 - mean_iou: 0.8842 - val_loss: 0.1097 - val_mean_iou: 0.8846\n",
      "Epoch 51/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0373 - mean_iou: 0.8849\n",
      "Epoch 00051: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0373 - mean_iou: 0.8849 - val_loss: 0.1093 - val_mean_iou: 0.8852\n",
      "Epoch 52/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0377 - mean_iou: 0.8855\n",
      "Epoch 00052: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0372 - mean_iou: 0.8855 - val_loss: 0.1104 - val_mean_iou: 0.8859\n",
      "Epoch 53/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0363 - mean_iou: 0.8861\n",
      "Epoch 00053: val_loss did not improve\n",
      "603/603 [==============================] - 2s 3ms/step - loss: 0.0366 - mean_iou: 0.8862 - val_loss: 0.1136 - val_mean_iou: 0.8865\n",
      "Epoch 54/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0362 - mean_iou: 0.8867\n",
      "Epoch 00054: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0362 - mean_iou: 0.8867 - val_loss: 0.1184 - val_mean_iou: 0.8871\n",
      "Epoch 55/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0360 - mean_iou: 0.8873\n",
      "Epoch 00055: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0358 - mean_iou: 0.8874 - val_loss: 0.1117 - val_mean_iou: 0.8877\n",
      "Epoch 56/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0352 - mean_iou: 0.8879\n",
      "Epoch 00056: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0351 - mean_iou: 0.8879 - val_loss: 0.1131 - val_mean_iou: 0.8883\n",
      "Epoch 57/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0353 - mean_iou: 0.8885\n",
      "Epoch 00057: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0352 - mean_iou: 0.8885 - val_loss: 0.1114 - val_mean_iou: 0.8888\n",
      "Epoch 58/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0348 - mean_iou: 0.8891\n",
      "Epoch 00058: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0352 - mean_iou: 0.8891 - val_loss: 0.1115 - val_mean_iou: 0.8894\n",
      "Epoch 59/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0333 - mean_iou: 0.8897\n",
      "Epoch 00059: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0334 - mean_iou: 0.8897 - val_loss: 0.1187 - val_mean_iou: 0.8900\n",
      "Epoch 60/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0340 - mean_iou: 0.8902\n",
      "Epoch 00060: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0339 - mean_iou: 0.8903 - val_loss: 0.1240 - val_mean_iou: 0.8906\n",
      "Epoch 61/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0327 - mean_iou: 0.8908\n",
      "Epoch 00061: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0327 - mean_iou: 0.8908 - val_loss: 0.1183 - val_mean_iou: 0.8911\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 62/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0314 - mean_iou: 0.8914\n",
      "Epoch 00062: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0318 - mean_iou: 0.8914 - val_loss: 0.1213 - val_mean_iou: 0.8917\n",
      "Epoch 63/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0324 - mean_iou: 0.8920\n",
      "Epoch 00063: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0325 - mean_iou: 0.8920 - val_loss: 0.1237 - val_mean_iou: 0.8922\n",
      "Epoch 64/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0332 - mean_iou: 0.8925\n",
      "Epoch 00064: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0336 - mean_iou: 0.8925 - val_loss: 0.1192 - val_mean_iou: 0.8927\n",
      "Epoch 65/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0341 - mean_iou: 0.8929\n",
      "Epoch 00065: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0341 - mean_iou: 0.8929 - val_loss: 0.1169 - val_mean_iou: 0.8932\n",
      "Epoch 66/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0315 - mean_iou: 0.8935\n",
      "Epoch 00066: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0315 - mean_iou: 0.8935 - val_loss: 0.1220 - val_mean_iou: 0.8937\n",
      "Epoch 67/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0311 - mean_iou: 0.8940\n",
      "Epoch 00067: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0308 - mean_iou: 0.8940 - val_loss: 0.1216 - val_mean_iou: 0.8943\n",
      "Epoch 68/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0299 - mean_iou: 0.8945\n",
      "Epoch 00068: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0299 - mean_iou: 0.8945 - val_loss: 0.1221 - val_mean_iou: 0.8948\n",
      "Epoch 69/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0299 - mean_iou: 0.8950\n",
      "Epoch 00069: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0297 - mean_iou: 0.8950 - val_loss: 0.1285 - val_mean_iou: 0.8953\n",
      "Epoch 70/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0311 - mean_iou: 0.8955\n",
      "Epoch 00070: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0311 - mean_iou: 0.8956 - val_loss: 0.1226 - val_mean_iou: 0.8958\n",
      "Epoch 71/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0301 - mean_iou: 0.8960\n",
      "Epoch 00071: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0301 - mean_iou: 0.8960 - val_loss: 0.1252 - val_mean_iou: 0.8963\n",
      "Epoch 72/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0287 - mean_iou: 0.8965\n",
      "Epoch 00072: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0288 - mean_iou: 0.8965 - val_loss: 0.1305 - val_mean_iou: 0.8968\n",
      "Epoch 73/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0284 - mean_iou: 0.8970\n",
      "Epoch 00073: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0283 - mean_iou: 0.8970 - val_loss: 0.1255 - val_mean_iou: 0.8973\n",
      "Epoch 74/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0323 - mean_iou: 0.8975\n",
      "Epoch 00074: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0322 - mean_iou: 0.8975 - val_loss: 0.1247 - val_mean_iou: 0.8977\n",
      "Epoch 75/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0316 - mean_iou: 0.8979\n",
      "Epoch 00075: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0315 - mean_iou: 0.8979 - val_loss: 0.1243 - val_mean_iou: 0.8981\n",
      "Epoch 76/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0289 - mean_iou: 0.8983\n",
      "Epoch 00076: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0292 - mean_iou: 0.8983 - val_loss: 0.1288 - val_mean_iou: 0.8985\n",
      "Epoch 77/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0302 - mean_iou: 0.8987\n",
      "Epoch 00077: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0301 - mean_iou: 0.8987 - val_loss: 0.1270 - val_mean_iou: 0.8990\n",
      "Epoch 78/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0283 - mean_iou: 0.8992\n",
      "Epoch 00078: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0282 - mean_iou: 0.8992 - val_loss: 0.1306 - val_mean_iou: 0.8994\n",
      "Epoch 79/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0280 - mean_iou: 0.8996\n",
      "Epoch 00079: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0282 - mean_iou: 0.8996 - val_loss: 0.1320 - val_mean_iou: 0.8998\n",
      "Epoch 80/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0277 - mean_iou: 0.9000\n",
      "Epoch 00080: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0276 - mean_iou: 0.9000 - val_loss: 0.1384 - val_mean_iou: 0.9003\n",
      "Epoch 81/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0290 - mean_iou: 0.9005\n",
      "Epoch 00081: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0290 - mean_iou: 0.9005 - val_loss: 0.1299 - val_mean_iou: 0.9007\n",
      "Epoch 82/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0271 - mean_iou: 0.9009\n",
      "Epoch 00082: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0270 - mean_iou: 0.9009 - val_loss: 0.1295 - val_mean_iou: 0.9011\n",
      "Epoch 83/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0259 - mean_iou: 0.9013\n",
      "Epoch 00083: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0260 - mean_iou: 0.9013 - val_loss: 0.1402 - val_mean_iou: 0.9015\n",
      "Epoch 84/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0291 - mean_iou: 0.9017\n",
      "Epoch 00084: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0292 - mean_iou: 0.9017 - val_loss: 0.1244 - val_mean_iou: 0.9019\n",
      "Epoch 85/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0372 - mean_iou: 0.9020\n",
      "Epoch 00085: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0372 - mean_iou: 0.9020 - val_loss: 0.1247 - val_mean_iou: 0.9021\n",
      "Epoch 86/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0291 - mean_iou: 0.9023\n",
      "Epoch 00086: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0292 - mean_iou: 0.9023 - val_loss: 0.1256 - val_mean_iou: 0.9025\n",
      "Epoch 87/100\n",
      "576/603 [===========================>..] - ETA: 0s - loss: 0.0259 - mean_iou: 0.9026\n",
      "Epoch 00087: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0261 - mean_iou: 0.9026 - val_loss: 0.1368 - val_mean_iou: 0.9028\n",
      "Epoch 88/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0274 - mean_iou: 0.9030\n",
      "Epoch 00088: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0273 - mean_iou: 0.9030 - val_loss: 0.1351 - val_mean_iou: 0.9032\n",
      "Epoch 89/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0266 - mean_iou: 0.9034\n",
      "Epoch 00089: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0265 - mean_iou: 0.9034 - val_loss: 0.1377 - val_mean_iou: 0.9036\n",
      "Epoch 90/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0252 - mean_iou: 0.9038\n",
      "Epoch 00090: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0252 - mean_iou: 0.9038 - val_loss: 0.1351 - val_mean_iou: 0.9040\n",
      "Epoch 91/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0241 - mean_iou: 0.9042\n",
      "Epoch 00091: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0241 - mean_iou: 0.9042 - val_loss: 0.1453 - val_mean_iou: 0.9044\n",
      "Epoch 92/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0236 - mean_iou: 0.9046\n",
      "Epoch 00092: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0235 - mean_iou: 0.9046 - val_loss: 0.1415 - val_mean_iou: 0.9048\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 93/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0234 - mean_iou: 0.9050\n",
      "Epoch 00093: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0234 - mean_iou: 0.9050 - val_loss: 0.1470 - val_mean_iou: 0.9052\n",
      "Epoch 94/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0235 - mean_iou: 0.9054\n",
      "Epoch 00094: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0238 - mean_iou: 0.9054 - val_loss: 0.1423 - val_mean_iou: 0.9056\n",
      "Epoch 95/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0245 - mean_iou: 0.9057\n",
      "Epoch 00095: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0245 - mean_iou: 0.9057 - val_loss: 0.1438 - val_mean_iou: 0.9059\n",
      "Epoch 96/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0244 - mean_iou: 0.9061\n",
      "Epoch 00096: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0244 - mean_iou: 0.9061 - val_loss: 0.1411 - val_mean_iou: 0.9063\n",
      "Epoch 97/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0247 - mean_iou: 0.9065\n",
      "Epoch 00097: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0247 - mean_iou: 0.9065 - val_loss: 0.1438 - val_mean_iou: 0.9066\n",
      "Epoch 98/100\n",
      "592/603 [============================>.] - ETA: 0s - loss: 0.0247 - mean_iou: 0.9068\n",
      "Epoch 00098: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0246 - mean_iou: 0.9068 - val_loss: 0.1475 - val_mean_iou: 0.9070\n",
      "Epoch 99/100\n",
      "600/603 [============================>.] - ETA: 0s - loss: 0.0237 - mean_iou: 0.9072\n",
      "Epoch 00099: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0237 - mean_iou: 0.9072 - val_loss: 0.1453 - val_mean_iou: 0.9073\n",
      "Epoch 100/100\n",
      "584/603 [============================>.] - ETA: 0s - loss: 0.0253 - mean_iou: 0.9075\n",
      "Epoch 00100: val_loss did not improve\n",
      "603/603 [==============================] - 1s 2ms/step - loss: 0.0252 - mean_iou: 0.9075 - val_loss: 0.1455 - val_mean_iou: 0.9076\n"
     ]
    }
   ],
   "source": [
    "earlystopper = EarlyStopping(patience=100, verbose=1)\n",
    "checkpointer = ModelCheckpoint('model-dsbowl2018-2.h5', verbose=1, save_best_only=True)\n",
    "results = model.fit(X_train, Y_train, validation_split=0.1, batch_size=8, epochs=100, \n",
    "                    callbacks=[earlystopper, checkpointer])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "603/603 [==============================] - 0s 803us/step\n",
      "67/67 [==============================] - 0s 416us/step\n",
      "65/65 [==============================] - 0s 421us/step\n"
     ]
    }
   ],
   "source": [
    "# Predict on train, val and test\n",
    "model = load_model('model-dsbowl2018-2.h5', custom_objects={'mean_iou': mean_iou})\n",
    "preds_train = model.predict(X_train[:int(X_train.shape[0]*0.9)], verbose=1)\n",
    "preds_val = model.predict(X_train[int(X_train.shape[0]*0.9):], verbose=1)\n",
    "preds_test = model.predict(X_test, verbose=1)\n",
    "\n",
    "# Threshold predictions\n",
    "preds_train_t = (preds_train > 0.5).astype(np.uint8)\n",
    "preds_val_t = (preds_val > 0.5).astype(np.uint8)\n",
    "preds_test_t = (preds_test > 0.5).astype(np.uint8)\n",
    "\n",
    "# Create list of upsampled test masks\n",
    "preds_test_upsampled = []\n",
    "for i in range(len(preds_test)):\n",
    "    preds_test_upsampled.append(resize(np.squeeze(preds_test[i]), \n",
    "                                       (sizes_test[i][0], sizes_test[i][1]), \n",
    "                                       mode='constant', preserve_range=True))\n",
    "    \n",
    "new_test_ids = []\n",
    "rles = []\n",
    "for n, id_ in enumerate(test_ids):\n",
    "    rle = list(prob_to_rles(preds_test_upsampled[n]))\n",
    "    rles.extend(rle)\n",
    "    new_test_ids.extend([id_] * len(rle))\n",
    "# Create submission DataFrame\n",
    "sub = pd.DataFrame()\n",
    "sub['ImageId'] = new_test_ids\n",
    "sub['EncodedPixels'] = pd.Series(rles).apply(lambda x: ' '.join(str(y) for y in x))\n",
    "sub.to_csv('sub-dsbowl2018-2.csv', index=False)    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 2",
   "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": "2.7.14"
  },
  "nav_menu": {},
  "nbpresent": {
   "slides": {
    "28b43202-5690-4169-9aca-6b9dabfeb3ec": {
     "id": "28b43202-5690-4169-9aca-6b9dabfeb3ec",
     "prev": null,
     "regions": {
      "3bba644a-cf4d-4a49-9fbd-e2554428cf9f": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "f3d3a388-7e2a-4151-9b50-c20498fceacc",
        "part": "whole"
       },
       "id": "3bba644a-cf4d-4a49-9fbd-e2554428cf9f"
      }
     }
    },
    "8104def2-4b68-44a0-8f1b-b03bf3b2a079": {
     "id": "8104def2-4b68-44a0-8f1b-b03bf3b2a079",
     "prev": "28b43202-5690-4169-9aca-6b9dabfeb3ec",
     "regions": {
      "7dded777-1ddf-4100-99ae-25cf1c15b575": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "fe47bd48-3414-4657-92e7-8b8d6cb0df00",
        "part": "whole"
       },
       "id": "7dded777-1ddf-4100-99ae-25cf1c15b575"
      }
     }
    }
   },
   "themes": {}
  },
  "toc": {
   "navigate_menu": true,
   "number_sections": true,
   "sideBar": true,
   "threshold": 6,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}