isic-task2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ISIC-2018: ResNet50 training\n",
    "Task 2: Lesion Attribute Detection\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import time\n",
    "import cv2\n",
    "import numpy as np\n",
    "import torch\n",
    "import torchvision\n",
    "import torch.optim as optim\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import ResNet50 model and print the layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net(\n",
      "  (resnet): ResNet(\n",
      "    (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "    (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    (relu): ReLU(inplace)\n",
      "    (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n",
      "    (layer1): Sequential(\n",
      "      (0): Bottleneck(\n",
      "        (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "        (downsample): Sequential(\n",
      "          (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "      (1): Bottleneck(\n",
      "        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (2): Bottleneck(\n",
      "        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "    )\n",
      "    (layer2): Sequential(\n",
      "      (0): Bottleneck(\n",
      "        (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "        (downsample): Sequential(\n",
      "          (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "          (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "      (1): Bottleneck(\n",
      "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (2): Bottleneck(\n",
      "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (3): Bottleneck(\n",
      "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "    )\n",
      "    (layer3): Sequential(\n",
      "      (0): Bottleneck(\n",
      "        (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "        (downsample): Sequential(\n",
      "          (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "          (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "      (1): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (2): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (3): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (4): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (5): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "    )\n",
      "    (layer4): Sequential(\n",
      "      (0): Bottleneck(\n",
      "        (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "        (downsample): Sequential(\n",
      "          (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "          (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        )\n",
      "      )\n",
      "      (1): Bottleneck(\n",
      "        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "      (2): Bottleneck(\n",
      "        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (relu): ReLU(inplace)\n",
      "      )\n",
      "    )\n",
      "    (avgpool): AvgPool2d(kernel_size=7, stride=1, padding=0)\n",
      "    (fc): Linear(in_features=2048, out_features=1000, bias=True)\n",
      "  )\n",
      "  (a_convT2d): ConvTranspose2d(2048, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
      "  (b_convT2d): ConvTranspose2d(1280, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
      "  (c_convT2d): ConvTranspose2d(640, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
      "  (convT2d3): ConvTranspose2d(320, 5, kernel_size=(4, 4), stride=(4, 4))\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "from model_resnet import Net\n",
    "\n",
    "net = Net(num_classes=5).cuda()\n",
    "print(net)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load dataset and visualize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training_size: 70, validation_size:18\n"
     ]
    }
   ],
   "source": [
    "import dataset\n",
    "lesionData = dataset.LesionData()\n",
    "train_loader, val_loader = lesionData.getTask2TrainingDataLoaders(val_percent = 20, batch_size = 30, augment=True)\n",
    "\n",
    "print('training_size: %d, validation_size:%d' % (len(train_loader),len(val_loader)) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "def imshow(img):\n",
    "    img = img / 2 + 0.5     # unnormalize\n",
    "    npimg = img.numpy()\n",
    "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
    "    plt.show()\n",
    "\n",
    "# get some random training images\n",
    "dataiter = iter(train_loader)\n",
    "images, labels, fnames = dataiter.next()\n",
    "\n",
    "# show images\n",
    "imshow(torchvision.utils.make_grid(images[0:4], nrow=4, padding=20))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loss functions and metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def dice_coef(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return (2. * intersection + 1.) / (torch.sum(y_true) + torch.sum(y_pred) + 1.)\n",
    "\n",
    "def dice_coef_loss(y_true, y_pred):\n",
    "    return -dice_coef(y_true, y_pred)\n",
    "\n",
    "def iou(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return (intersection + 1.) / (torch.sum(y_true) + torch.sum(y_pred) - intersection + 1.)\n",
    "\n",
    "def falsepos(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return torch.sum(y_pred) - intersection\n",
    "\n",
    "def falseneg(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return torch.sum(y_true) - intersection\n",
    "\n",
    "def precision(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return intersection / (torch.sum(y_pred) + 1.)\n",
    "\n",
    "def recall(y_true, y_pred):\n",
    "    intersection = torch.sum(y_true * y_pred)\n",
    "    return intersection / (torch.sum(y_true) + 1.)\n",
    "\n",
    "def fscore(y_true, y_pred):\n",
    "    presci = precision(y_true, y_pred)\n",
    "    rec = recall(y_true, y_pred)\n",
    "    return 2*(presci * rec)/(presci + rec)\n",
    "\n",
    "def weighted_fscore_loss(prew=1, recw=1):\n",
    "    def fscore_loss(y_true, y_pred):\n",
    "        presci = precision(y_true, y_pred)\n",
    "        rec = recall(y_true, y_pred)\n",
    "        return -(prew+recw)*(presci * rec)/(prew*presci + recw*rec)\n",
    "    return fscore_loss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualizing losses and metrics with visdom\n",
    "Run visdom server before executing this"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from visualization import Visualization\n",
    "\n",
    "learning_rate = 1e-3\n",
    "\n",
    "optimizer = optim.Adam(net.parameters(), lr=learning_rate)\n",
    "lrscheduler = None #optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "iou_vis = Visualization('IOU / Dice Coeff.')\n",
    "fpfn_vis = Visualization('False Positive / False Negative')\n",
    "fscore_vis = Visualization('Precision, Recall, F-Score')\n",
    "loss_vis = Visualization('Mean loss')\n",
    "\n",
    "measures = {\n",
    "    'dice_coeff': (dice_coef, iou_vis), \n",
    "    'iou': (iou, iou_vis), \n",
    "    'fp': (falsepos, fpfn_vis), \n",
    "    'fn': (falseneg, fpfn_vis), \n",
    "    'precision': (precision, fscore_vis), \n",
    "    'recall': (recall, fscore_vis), \n",
    "    'fscore': (fscore, fscore_vis) }\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training the model\n",
    "Vary number of frozen layers by naming layers to un-freeze in ```trainable```\n",
    "\n",
    "Weight the loss function to emphasize precsion or recall by modifying ```fscores``` array"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 0:  dice_coeff:0.11218, iou:0.06016, fp:244671.36849, fn:130659.15332, precision:0.08941, recall:0.15717, fscore:0.11218, train_loss:-0.23036, val_loss:-0.11218,\n",
      "Epoch: 1:  dice_coeff:0.22818, iou:0.12968, fp:138454.05512, fn:763857.56076, precision:0.49442, recall:0.14990, fscore:0.22818, train_loss:-0.27273, val_loss:-0.22818,\n",
      "Epoch: 2:  dice_coeff:0.26751, iou:0.15770, fp:203958.22114, fn:153274.80968, precision:0.24537, recall:0.31456, fscore:0.26751, train_loss:-0.31464, val_loss:-0.26751,\n",
      "Epoch: 3:  dice_coeff:0.29822, iou:0.17936, fp:191323.50477, fn:172666.69878, precision:0.29482, recall:0.31250, fscore:0.29821, train_loss:-0.33292, val_loss:-0.29821,\n",
      "Epoch: 4:  dice_coeff:0.28410, iou:0.16961, fp:203016.84462, fn:123011.03559, precision:0.24745, recall:0.34578, fscore:0.28410, train_loss:-0.35203, val_loss:-0.28410,\n",
      "Epoch: 5:  dice_coeff:0.34437, iou:0.21187, fp:151325.37977, fn:299405.34245, precision:0.44112, recall:0.28967, fscore:0.34437, train_loss:-0.38517, val_loss:-0.34437,\n",
      "Epoch: 6:  dice_coeff:0.35040, iou:0.21687, fp:165505.55859, fn:218008.39019, precision:0.39089, recall:0.32471, fscore:0.35039, train_loss:-0.38874, val_loss:-0.35039,\n",
      "Epoch: 7:  dice_coeff:0.24173, iou:0.14619, fp:218622.08333, fn:67438.74599, precision:0.18668, recall:0.38159, fscore:0.24173, train_loss:-0.41062, val_loss:-0.24173,\n",
      "Epoch: 8:  dice_coeff:0.33921, iou:0.20959, fp:172423.66016, fn:198887.42622, precision:0.35794, recall:0.33227, fscore:0.33921, train_loss:-0.40828, val_loss:-0.33921,\n",
      "Epoch: 9:  dice_coeff:0.35474, iou:0.21890, fp:143859.81641, fn:303712.26693, precision:0.46994, recall:0.28930, fscore:0.35474, train_loss:-0.41306, val_loss:-0.35474,\n",
      "Epoch: 10:  dice_coeff:0.35300, iou:0.21884, fp:140472.13238, fn:337389.84115, precision:0.48065, recall:0.28418, fscore:0.35300, train_loss:-0.39520, val_loss:-0.35300,\n",
      "Epoch: 11:  dice_coeff:0.36098, iou:0.22485, fp:133044.63281, fn:355697.50868, precision:0.51472, recall:0.28151, fscore:0.36098, train_loss:-0.41187, val_loss:-0.36098,\n",
      "Epoch: 12:  dice_coeff:0.31741, iou:0.19698, fp:194668.90582, fn:104944.85265, precision:0.27599, recall:0.39265, fscore:0.31740, train_loss:-0.41618, val_loss:-0.31740,\n",
      "Epoch: 13:  dice_coeff:0.34154, iou:0.21386, fp:180340.75738, fn:150303.00781, precision:0.33277, recall:0.36094, fscore:0.34154, train_loss:-0.40168, val_loss:-0.34154,\n",
      "Epoch: 14:  dice_coeff:0.35917, iou:0.22510, fp:158885.40148, fn:230439.57812, precision:0.40992, recall:0.32857, fscore:0.35917, train_loss:-0.42526, val_loss:-0.35917,\n",
      "Epoch: 15:  dice_coeff:0.35871, iou:0.22471, fp:154156.28928, fn:249258.35916, precision:0.42802, recall:0.31671, fscore:0.35870, train_loss:-0.42294, val_loss:-0.35870,\n",
      "Epoch: 16:  dice_coeff:0.33771, iou:0.20579, fp:118047.27821, fn:480014.76172, precision:0.56946, recall:0.24225, fscore:0.33771, train_loss:-0.39734, val_loss:-0.33771,\n",
      "Epoch: 17:  dice_coeff:0.34720, iou:0.21416, fp:134762.31901, fn:368483.31380, precision:0.50545, recall:0.26960, fscore:0.34719, train_loss:-0.42483, val_loss:-0.34719,\n",
      "Epoch: 18:  dice_coeff:0.34784, iou:0.21738, fp:174451.60547, fn:168973.02951, precision:0.35519, recall:0.35098, fscore:0.34783, train_loss:-0.41118, val_loss:-0.34783,\n",
      "Epoch: 19:  dice_coeff:0.36101, iou:0.22502, fp:131530.15365, fn:362723.33767, precision:0.52023, recall:0.28018, fscore:0.36101, train_loss:-0.41610, val_loss:-0.36101,\n",
      "Epoch: 20:  dice_coeff:0.37740, iou:0.23827, fp:151657.96701, fn:232391.74913, precision:0.44475, recall:0.33514, fscore:0.37740, train_loss:-0.43082, val_loss:-0.37740,\n",
      "Epoch: 21:  dice_coeff:0.37229, iou:0.23588, fp:166301.04167, fn:169779.33377, precision:0.38712, recall:0.37271, fscore:0.37229, train_loss:-0.43947, val_loss:-0.37229,\n",
      "Epoch: 22:  dice_coeff:0.34805, iou:0.21881, fp:178632.88064, fn:140160.18316, precision:0.33723, recall:0.37087, fscore:0.34805, train_loss:-0.42532, val_loss:-0.34805,\n",
      "Epoch: 23:  dice_coeff:0.36789, iou:0.23166, fp:151964.88368, fn:243415.75087, precision:0.43568, recall:0.32422, fscore:0.36788, train_loss:-0.41330, val_loss:-0.36788,\n",
      "Epoch: 24:  dice_coeff:0.36717, iou:0.23124, fp:159893.97179, fn:208582.35851, precision:0.41459, recall:0.33772, fscore:0.36717, train_loss:-0.42940, val_loss:-0.36717,\n"
     ]
    }
   ],
   "source": [
    "import train\n",
    "\n",
    "measurement_log = []\n",
    "fscores = [(1,1)] * 25\n",
    "trainable = [['layer2', 'layer3', 'layer4']]*25 # [['layer4']]*10 + [['layer3','layer4']]*15\n",
    "# fscores = [(2,1)] * 5 + [(1.5,1)]*5 + [(1,1)]*5 + [(1,1.5)]*5 + [(1,2)]*5\n",
    "# frozen = [9] * 25\n",
    "\n",
    "for epoch in range(25):\n",
    "    net.setTrainable(trainable[epoch])\n",
    "    measurements = train.fit(net, train_loader, val_loader, weighted_fscore_loss(*(fscores[epoch])), optimizer, lrscheduler, measures, epoch, loss_vis)\n",
    "    measurement_log.append(measurements)\n",
    "    print(\"Epoch: %d: \" % epoch, end='')\n",
    "    for k,v in measurements.items():\n",
    "        print(\" {}:{:.5f}\".format(k,v), end=',')\n",
    "    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save trained model and measurements"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import csv\n",
    "\n",
    "timestamp = time.strftime('%m%d%H%M')\n",
    "outputfolder = './output/resnet-%s'%timestamp\n",
    "os.mkdir(outputfolder)\n",
    "\n",
    "# write out weights\n",
    "net.save(os.path.join(outputfolder,'lesions.pth'))\n",
    "\n",
    "# write out log\n",
    "with open(os.path.join(outputfolder,'info.txt'), 'w') as f:\n",
    "    f.write('%s\\n' % time.strftime('%m-%d %H:%M'))\n",
    "    f.write('optimizer: %s\\n' % type(optimizer).__name__)\n",
    "    f.write('scheduler: %s\\n' % type(lrscheduler).__name__)\n",
    "    f.write('learning_rate: %.6f\\n' % learning_rate)\n",
    "    f.write('\\n')\n",
    "    \n",
    "with open('./output/resnet-%s/measurements.txt' % timestamp, 'w') as f:\n",
    "    wr = csv.writer(f)\n",
    "    wr.writerow([\"Epoch\"] + list(measurements.keys()))\n",
    "    for epoch,measurement in enumerate(measurement_log):\n",
    "        wr.writerow([epoch] + list(measurement.values()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sample some images and show results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "ename": "RuntimeError",
     "evalue": "Error(s) in loading state_dict for Net:\n\tsize mismatch for convT2d3.weight: copying a param with shape torch.Size([320, 5, 4, 4]) from checkpoint, the shape in current model is torch.Size([320, 1, 4, 4]).\n\tsize mismatch for convT2d3.bias: copying a param with shape torch.Size([5]) from checkpoint, the shape in current model is torch.Size([1]).",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-9-555b296cf697>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0meval\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mevaluator\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0meval\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mEvaluator\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moutputfolder\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m'lesions.pth'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moutputfolder\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0msamples\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mevaluator\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msample\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/code/isic-2018/eval.py\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, model_filepath, outputDir)\u001b[0m\n\u001b[1;32m     15\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputDir\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moutputDir\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     16\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnet\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mNet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 17\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnet\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel_filepath\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     18\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnet\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0meval\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     19\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlesionData\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdataset\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mLesionData\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/code/isic-2018/model_resnet.py\u001b[0m in \u001b[0;36mload\u001b[0;34m(self, filename)\u001b[0m\n\u001b[1;32m     66\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     67\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfilename\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 68\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_state_dict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfilename\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     69\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     70\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0m__name__\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;34m\"__main__\"\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/torch/lib/python3.7/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36mload_state_dict\u001b[0;34m(self, state_dict, strict)\u001b[0m\n\u001b[1;32m    767\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0merror_msgs\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    768\u001b[0m             raise RuntimeError('Error(s) in loading state_dict for {}:\\n\\t{}'.format(\n\u001b[0;32m--> 769\u001b[0;31m                                self.__class__.__name__, \"\\n\\t\".join(error_msgs)))\n\u001b[0m\u001b[1;32m    770\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    771\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_named_members\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mget_members_fn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mprefix\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m''\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrecurse\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mRuntimeError\u001b[0m: Error(s) in loading state_dict for Net:\n\tsize mismatch for convT2d3.weight: copying a param with shape torch.Size([320, 5, 4, 4]) from checkpoint, the shape in current model is torch.Size([320, 1, 4, 4]).\n\tsize mismatch for convT2d3.bias: copying a param with shape torch.Size([5]) from checkpoint, the shape in current model is torch.Size([1])."
     ]
    }
   ],
   "source": [
    "import eval\n",
    "\n",
    "evaluator = eval.Evaluator(os.path.join(outputfolder,'lesions.pth'), outputfolder)\n",
    "samples = evaluator.sample()\n",
    "\n",
    "for idx in range(len(samples)):\n",
    "    plt.figure()\n",
    "    plt.axis('off')\n",
    "    plt.imshow(samples[idx])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (torch-1.0)",
   "language": "python",
   "name": "myenv"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}