train_segm_unet_vgg16.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import keras\n",
    "from keras import optimizers\n",
    "import re\n",
    "from warnings import warn\n",
    "from unet import construct_unet\n",
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '2'\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !curl -O https://raw.githubusercontent.com/DSLituiev/cardioloader/master/unet.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "fn_meta = '/data/kidney/tables/splits-2019-03-24.csv'\n",
    "meta = pd.read_csv(fn_meta, index_col=0)['set']\n",
    "name = ''\n",
    "roi_unique_labels = []\n",
    "########################################\n",
    "# fix here\n",
    "########################################\n",
    "slide_ids = {}\n",
    "for kk in ['train', 'val', 'test']:\n",
    "    slide_ids[kk] = meta[meta==kk].index.tolist()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y23969-t1+29-r105-tissue+lymph_agg.json\r\n",
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y23969-t1+29-r105-tissue+lymph_agg.png\r\n",
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y24993-t1+29-r114-tissue+lymph_agg.json\r\n",
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y24993-t1+29-r114-tissue+lymph_agg.png\r\n",
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y26017-t1+19+29-r121-tissue+glom+lymph_agg.json\r\n",
      "07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y26017-t1+19+29-r121-tissue+glom+lymph_agg.png\r\n",
      "09eb9369342c45a23a3b52408eaabfe55ded9c7a-x16654-y11395-t2+36-r55-tissue+lymph_agg.json\r\n",
      "09eb9369342c45a23a3b52408eaabfe55ded9c7a-x16654-y11395-t2+36-r55-tissue+lymph_agg.png\r\n",
      "09eb9369342c45a23a3b52408eaabfe55ded9c7a-x36001-y11392-t34+9-r14-tissue+lymph_agg.json\r\n",
      "09eb9369342c45a23a3b52408eaabfe55ded9c7a-x36001-y11392-t34+9-r14-tissue+lymph_agg.png\r\n",
      "ls: write error: Broken pipe\r\n"
     ]
    }
   ],
   "source": [
    "!ls /data/kidney/data/lymphoid_agg/newdata/data_1024/la_split/train/lymph | head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from generators import PngDir, JsonDir, SegmSet, np_collate\n",
    "from torch.utils.data import Dataset, DataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('07272cf63c51cf065e5d7117d2a085491a9dfe9c', '2343', '23969')"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fn_dcm = '07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y23969-t1+29-r105-tissue+lymph_agg.png'\n",
    "# fn_dcm = '07272cf63c51cf065e5d7117d2a085491a9dfe9c-x2343-y23969-.png'\n",
    "matches = re.match('([\\da-f]+)-x([\\d]+)-y([\\d]+)-.*\\.png', fn_dcm)\n",
    "matches.groups()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/exx/.conda/envs/digipath/lib/python3.7/site-packages/ipykernel_launcher.py:13: FutureWarning: Sorting because non-concatenation axis is not aligned. A future version\n",
      "of pandas will change to not sort by default.\n",
      "\n",
      "To accept the future behavior, pass 'sort=False'.\n",
      "\n",
      "To retain the current behavior and silence the warning, pass 'sort=True'.\n",
      "\n",
      "  del sys.path[0]\n"
     ]
    }
   ],
   "source": [
    "batch_size = 8\n",
    "num_workers = 4\n",
    "\n",
    "dfs = {}\n",
    "segmsets = {}\n",
    "dloaders = {}\n",
    "for set_ in ['train', 'val', 'test']:\n",
    "    dfs[set_] = {}\n",
    "    for label in ['norm', 'lymph']:\n",
    "        dirname = f'/data/kidney/data/lymphoid_agg/newdata/data_1024/la_split/{set_}/{label}'\n",
    "        pngdir = PngDir(dirname, verbose=False)\n",
    "        jsondir = JsonDir(dirname, verbose=False)\n",
    "        dfs[set_][label] = pd.concat([pngdir.content, jsondir.content], axis=1)\n",
    "        dfs[set_][label].index.names = ['slide_id', 'x', 'y']\n",
    "        \n",
    "    segmsets[set_] = SegmSet(dfs[set_]['lymph'], shape=(256,256))\n",
    "    dloaders[set_] = DataLoader(segmsets[set_], collate_fn=np_collate,\n",
    "                                num_workers=num_workers,\n",
    "                                batch_size=batch_size,\n",
    "                                shuffle=(set_=='train'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# import json\n",
    "# from pycocotools.mask import decode\n",
    "\n",
    "# def tissue_name_positive(x):\n",
    "#     return ('lymph' in x) or ('infl' in x)\n",
    "\n",
    "# fnjson = dloaders['val'].dataset.filenames.iloc[0]['json']\n",
    "# with open(fnjson) as fh:\n",
    "#     rois = json.load(fh)\n",
    "\n",
    "# any([decode(rr) for rr in rois if tissue_name_positive(rr['name'])])\n",
    "# segmsets[set_][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "import imgaug.augmenters as iaa\n",
    "import imgaug as ia\n",
    "# iaa.seed(2)\n",
    "\n",
    "seq = iaa.Sequential([\n",
    "#     iaa.CoarseDropout(0.1, size_percent=0.2),\n",
    "    iaa.ElasticTransformation(alpha=10, sigma=1),\n",
    "    iaa.Affine(\n",
    "            scale=1.2,\n",
    "            translate_percent=(5),\n",
    "            translate_px=None,\n",
    "            rotate=(-30, 30),\n",
    "            shear=(5.0),\n",
    "            cval = 250,\n",
    "        ),\n",
    "     iaa.Sometimes(0.3, iaa.AddToHueAndSaturation((-15, 15))),\n",
    "     iaa.Sometimes(0.3, iaa.GaussianBlur(sigma=(0.0, 3.0))),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "png     /data/kidney/data/lymphoid_agg/newdata/data_10...\n",
       "json    /data/kidney/data/lymphoid_agg/newdata/data_10...\n",
       "Name: (07272cf63c51cf065e5d7117d2a085491a9dfe9c, 2343, 23969), dtype: object"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfs['train']['lymph'].iloc[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# sample = SegmSet.parse_sample(dfs['train']['lymph'].iloc[0], \n",
    "#                               positive=False,\n",
    "#                              )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def data_gen(x, augment=False):\n",
    "    while True:\n",
    "        for sample in x:\n",
    "            #sample['image'], sample['mask'][...,np.newaxis]\n",
    "            if augment:\n",
    "                bg = seq(images=sample['image'],#.astype(np.float16), \n",
    "                         segmentation_maps=sample['mask'],\n",
    "                        return_batch=True,\n",
    "                    )\n",
    "                yield bg.images_aug, bg.segmentation_maps_aug[..., np.newaxis].astype(np.float32)\n",
    "            else:\n",
    "                yield (keras.applications.vgg16.preprocess_input(sample['image']),\n",
    "                        sample['mask'][..., np.newaxis].astype(np.float32))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img, msk in data_gen(dloaders['val']):\n",
    "    break\n",
    "# batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fe99c355e80>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(img[4])\n",
    "# plt.imshow(msk[2,...,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# plt.imshow(dloaders['val'].dataset[4]['image'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8, 256, 256, 3)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/exx/.conda/envs/digipath/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "width, height = 256, 256\n",
    "unet = construct_unet(width, height, encoder = 'vgg16', freeze_encoder=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 256, 256, 3)  0                                            \n",
      "__________________________________________________________________________________________________\n",
      "block1_conv1 (Conv2D)           (None, 256, 256, 64) 1792        input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "block1_conv2 (Conv2D)           (None, 256, 256, 64) 36928       block1_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block1_pool (MaxPooling2D)      (None, 128, 128, 64) 0           block1_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block2_conv1 (Conv2D)           (None, 128, 128, 128 73856       block1_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "block2_conv2 (Conv2D)           (None, 128, 128, 128 147584      block2_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block2_pool (MaxPooling2D)      (None, 64, 64, 128)  0           block2_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv1 (Conv2D)           (None, 64, 64, 256)  295168      block2_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv2 (Conv2D)           (None, 64, 64, 256)  590080      block3_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv3 (Conv2D)           (None, 64, 64, 256)  590080      block3_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_pool (MaxPooling2D)      (None, 32, 32, 256)  0           block3_conv3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block4_conv1 (Conv2D)           (None, 32, 32, 512)  1180160     block3_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "block4_conv2 (Conv2D)           (None, 32, 32, 512)  2359808     block4_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block4_conv3 (Conv2D)           (None, 32, 32, 512)  2359808     block4_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block4_pool (MaxPooling2D)      (None, 16, 16, 512)  0           block4_conv3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2D)  (None, 32, 32, 512)  0           block4_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 32, 32, 768)  0           up_sampling2d_1[0][0]            \n",
      "                                                                 block3_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)               (None, 32, 32, 512)  3539456     concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_1 (BatchNor (None, 32, 32, 512)  2048        conv2d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_1 (Activation)       (None, 32, 32, 512)  0           batch_normalization_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)               (None, 32, 32, 256)  1179904     activation_1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_2 (BatchNor (None, 32, 32, 256)  1024        conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_2 (Activation)       (None, 32, 32, 256)  0           batch_normalization_2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2D)  (None, 64, 64, 256)  0           activation_2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, 64, 64, 384)  0           up_sampling2d_2[0][0]            \n",
      "                                                                 block2_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_3 (Conv2D)               (None, 64, 64, 384)  1327488     concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_3 (BatchNor (None, 64, 64, 384)  1536        conv2d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_3 (Activation)       (None, 64, 64, 384)  0           batch_normalization_3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_4 (Conv2D)               (None, 64, 64, 192)  663744      activation_3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_4 (BatchNor (None, 64, 64, 192)  768         conv2d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_4 (Activation)       (None, 64, 64, 192)  0           batch_normalization_4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "up_sampling2d_3 (UpSampling2D)  (None, 128, 128, 192 0           activation_4[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_3 (Concatenate)     (None, 128, 128, 256 0           up_sampling2d_3[0][0]            \n",
      "                                                                 block1_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_5 (Conv2D)               (None, 128, 128, 256 590080      concatenate_3[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_5 (BatchNor (None, 128, 128, 256 1024        conv2d_5[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_5 (Activation)       (None, 128, 128, 256 0           batch_normalization_5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)               (None, 128, 128, 128 295040      activation_5[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_6 (BatchNor (None, 128, 128, 128 512         conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_6 (Activation)       (None, 128, 128, 128 0           batch_normalization_6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "up_sampling2d_4 (UpSampling2D)  (None, 256, 256, 128 0           activation_6[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_4 (Concatenate)     (None, 256, 256, 192 0           up_sampling2d_4[0][0]            \n",
      "                                                                 block1_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_7 (Conv2D)               (None, 256, 256, 128 221312      concatenate_4[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_7 (BatchNor (None, 256, 256, 128 512         conv2d_7[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_7 (Activation)       (None, 256, 256, 128 0           batch_normalization_7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_8 (Conv2D)               (None, 256, 256, 64) 73792       activation_7[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_8 (BatchNor (None, 256, 256, 64) 256         conv2d_8[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_8 (Activation)       (None, 256, 256, 64) 0           batch_normalization_8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_9 (Conv2D)               (None, 256, 256, 1)  65          activation_8[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_9 (Activation)       (None, 256, 256, 1)  0           conv2d_9[0][0]                   \n",
      "==================================================================================================\n",
      "Total params: 15,533,825\n",
      "Trainable params: 7,894,721\n",
      "Non-trainable params: 7,639,104\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "unet.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "unet.compile(optimizer=optimizers.Adam(lr=1e-5), loss='binary_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "23/23 [==============================] - 11s 483ms/step - loss: 0.2168 - acc: 0.9261 - val_loss: 0.2399 - val_acc: 0.9149\n",
      "Epoch 2/20\n",
      "23/23 [==============================] - 8s 335ms/step - loss: 0.2269 - acc: 0.9217 - val_loss: 0.2421 - val_acc: 0.9151\n",
      "Epoch 3/20\n",
      "23/23 [==============================] - 8s 349ms/step - loss: 0.2269 - acc: 0.9224 - val_loss: 0.2499 - val_acc: 0.9153\n",
      "Epoch 4/20\n",
      "23/23 [==============================] - 8s 348ms/step - loss: 0.2290 - acc: 0.9208 - val_loss: 0.2426 - val_acc: 0.9155\n",
      "Epoch 5/20\n",
      "23/23 [==============================] - 8s 351ms/step - loss: 0.2116 - acc: 0.9283 - val_loss: 0.2463 - val_acc: 0.9159\n",
      "Epoch 6/20\n",
      "23/23 [==============================] - 8s 343ms/step - loss: 0.2196 - acc: 0.9247 - val_loss: 0.2439 - val_acc: 0.9163\n",
      "Epoch 7/20\n",
      "23/23 [==============================] - 8s 351ms/step - loss: 0.2126 - acc: 0.9277 - val_loss: 0.2710 - val_acc: 0.9104\n",
      "Epoch 8/20\n",
      "23/23 [==============================] - 8s 349ms/step - loss: 0.2180 - acc: 0.9258 - val_loss: 0.2501 - val_acc: 0.9157\n",
      "Epoch 9/20\n",
      "23/23 [==============================] - 8s 346ms/step - loss: 0.2264 - acc: 0.9224 - val_loss: 0.2438 - val_acc: 0.9163\n",
      "Epoch 10/20\n",
      "23/23 [==============================] - 8s 356ms/step - loss: 0.2211 - acc: 0.9236 - val_loss: 0.2449 - val_acc: 0.9162\n",
      "Epoch 11/20\n",
      "23/23 [==============================] - 8s 349ms/step - loss: 0.2057 - acc: 0.9309 - val_loss: 0.2588 - val_acc: 0.9137\n",
      "Epoch 12/20\n",
      "23/23 [==============================] - 8s 346ms/step - loss: 0.2076 - acc: 0.9289 - val_loss: 0.2619 - val_acc: 0.9133\n",
      "Epoch 13/20\n",
      "23/23 [==============================] - 8s 342ms/step - loss: 0.2110 - acc: 0.9276 - val_loss: 0.2695 - val_acc: 0.9111\n",
      "Epoch 14/20\n",
      "23/23 [==============================] - 8s 354ms/step - loss: 0.2171 - acc: 0.9256 - val_loss: 0.2614 - val_acc: 0.9134\n",
      "Epoch 15/20\n",
      "23/23 [==============================] - 8s 347ms/step - loss: 0.2068 - acc: 0.9296 - val_loss: 0.2675 - val_acc: 0.9119\n",
      "Epoch 16/20\n",
      "23/23 [==============================] - 8s 348ms/step - loss: 0.2088 - acc: 0.9293 - val_loss: 0.2802 - val_acc: 0.9080\n",
      "Epoch 17/20\n",
      "23/23 [==============================] - 8s 345ms/step - loss: 0.2183 - acc: 0.9235 - val_loss: 0.2582 - val_acc: 0.9144\n",
      "Epoch 18/20\n",
      "23/23 [==============================] - 8s 349ms/step - loss: 0.2039 - acc: 0.9299 - val_loss: 0.2643 - val_acc: 0.9126\n",
      "Epoch 19/20\n",
      "23/23 [==============================] - 8s 348ms/step - loss: 0.2067 - acc: 0.9293 - val_loss: 0.2469 - val_acc: 0.9164\n",
      "Epoch 20/20\n",
      "23/23 [==============================] - 8s 352ms/step - loss: 0.2025 - acc: 0.9306 - val_loss: 0.2483 - val_acc: 0.9161\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fe98c253b70>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unet.fit_generator(data_gen(dloaders['train']), steps_per_epoch=len(dloaders['train']),\n",
    "                   validation_data=data_gen(dloaders['val']), validation_steps=len(dloaders['val']),\n",
    "                   epochs=20,\n",
    "                  )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [],
   "source": [
    "# unet.save('checkpoints/unet_dlituiev_2019-05-17.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "val_predicted = unet.predict_generator(data_gen(dloaders['val']), steps=len(dloaders['val']),)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_predicted = unet.predict_generator(data_gen(dloaders['test']), steps=len(dloaders['test']),)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Int64Index([ 1809,  2833,  2833, 10001, 10001, 11025, 11025, 12049, 12049,\n",
       "            12049, 14346, 15370, 16145, 16394, 17169, 17169, 17169, 17418,\n",
       "            17418, 18193, 18193, 18442, 18442, 18442, 19466, 19466, 20490,\n",
       "            21265, 21265, 21514, 21514, 22289, 22289, 22538, 23313, 23313,\n",
       "            23562, 24337, 25361, 25361, 25361, 26385, 26385, 27658, 27658,\n",
       "            28682, 28682, 33553, 34577, 36874, 40721, 40721, 41745, 41745,\n",
       "            41745, 42769, 42769, 42769, 44042, 44042, 45066, 45066, 45066,\n",
       "            45066, 45841, 46865, 46865, 46865, 47889, 49937, 49937, 50961,\n",
       "            50961, 51985, 54033, 54033, 54033, 55057, 55057, 55057, 55057,\n",
       "            56081, 56081, 56081, 56081, 57105, 57105, 57105, 58129],\n",
       "           dtype='int64', name='x')"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dloaders['test'].dataset.filenames.loc['f7f931a5cf3185a385e9aa34e6e9a566fc88000c', :].index.get_level_values('x')#.index\n",
    "# dloaders['test'].dataset.filenames"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "155"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "index_test = dloaders['test'].dataset.filenames.index.to_frame().reset_index(drop=True)\n",
    "# index_test = [x.tolist() for _, x in index_test.iterrows()]\n",
    "('c886827fe8c10b4699a0f1616331e36b46a05617', 44677, 6426)\n",
    "len(index_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['70bb3032750d09e7549928c0dbf79afc30d7cb68',\n",
       "       'a1fc67fbb21f43b9e8904b9b46bd94f83493b37a',\n",
       "       'c886827fe8c10b4699a0f1616331e36b46a05617',\n",
       "       'dfe3ee768f72bddd289c7d5bb88b15cbb89be7e6',\n",
       "       'f7f931a5cf3185a385e9aa34e6e9a566fc88000c'], dtype=object)"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "index_test.slide_id.unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([118, 119, 120]),)"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.where((index_test.slide_id == 'c886827fe8c10b4699a0f1616331e36b46a05617') &\\\n",
    "            (index_test.x == 44677) & \\\n",
    "            (index_test.y == 6426)\n",
    "        )\n",
    "\n",
    "np.where((index_test.slide_id == 'f7f931a5cf3185a385e9aa34e6e9a566fc88000c') &\\\n",
    "            (index_test.x == 41745) #& \\\n",
    "#             (index_test.y == 20967)\n",
    "        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "# [tuple(x) for x in index_test.values]\n",
    "# index_test"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "py3.7: digipath",
   "language": "python",
   "name": "digipath"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}