Fast.ai/(Fast.ai) MNIST CNN, Skip-connection (U-NET).ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load Data\n",
    "path = untar_data(URLs.MNIST)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WindowsPath('C:/Users/bokhy/.fastai/data/mnist_png')"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 216x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 1. Create imagelist\n",
    "il = ImageList.from_folder(path, convert_mode='L')\n",
    "il\n",
    "il[0].show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "defaults.cmap='binary'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ItemLists;\n",
       "\n",
       "Train: ImageList (60000 items)\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Valid: ImageList (10000 items)\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Test: None"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sd = il.split_by_folder(train='training', valid='testing')\n",
    "sd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LabelLists;\n",
       "\n",
       "Train: LabelList (60000 items)\n",
       "x: ImageList\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "y: CategoryList\n",
       "0,0,0,0,0\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Valid: LabelList (10000 items)\n",
       "x: ImageList\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "y: CategoryList\n",
       "0,0,0,0,0\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Test: None"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 2. take imagelist and label it (by folder in this case)\n",
    "ll = sd.label_from_folder()\n",
    "ll"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "x,y = ll.train[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([RandTransform(tfm=TfmPixel (pad), kwargs={'padding': 3, 'mode': 'zeros'}, p=1.0, resolved={}, do_run=True, is_random=True, use_on_y=True),\n",
       "  RandTransform(tfm=TfmPixel (crop), kwargs={'size': 28, 'row_pct': (0, 1), 'col_pct': (0, 1)}, p=1.0, resolved={}, do_run=True, is_random=True, use_on_y=True)],\n",
       " [])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tfms = ([*rand_pad(padding=3, size=28, mode='zeros')], [])\n",
    "tfms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 3. Transform\n",
    "ll = ll.transform(tfms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "#batch size\n",
    "bs = 32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ImageDataBunch;\n",
       "\n",
       "Train: LabelList (60000 items)\n",
       "x: ImageList\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "y: CategoryList\n",
       "0,0,0,0,0\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Valid: LabelList (10000 items)\n",
       "x: ImageList\n",
       "Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28),Image (1, 28, 28)\n",
       "y: CategoryList\n",
       "0,0,0,0,0\n",
       "Path: C:\\Users\\bokhy\\.fastai\\data\\mnist_png;\n",
       "\n",
       "Test: None"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 4.databunch \n",
    "# not using imagenet_stats because not using pretrained model\n",
    "data = ll.databunch(bs=bs).normalize()\n",
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "x,y = data.train_ds[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 216x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x.show()\n",
    "print(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(torch.Size([28, 1, 28, 28]), torch.Size([28]))"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xb,yb = data.one_batch()\n",
    "xb.shape,yb.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "data.show_batch(rows=3, figsize=(5,5))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Basic CNN with batchnorm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conv(ni,nf): return nn.Conv2d(ni, nf, kernel_size=3, stride=2, padding=1)\n",
    "model = nn.Sequential(\n",
    "    conv(1, 8), # 14\n",
    "    nn.BatchNorm2d(8),\n",
    "    nn.ReLU(),\n",
    "    conv(8, 16), # 7\n",
    "    nn.BatchNorm2d(16),\n",
    "    nn.ReLU(),\n",
    "    conv(16, 32), # 4\n",
    "    nn.BatchNorm2d(32),\n",
    "    nn.ReLU(),\n",
    "    conv(32, 16), # 2\n",
    "    nn.BatchNorm2d(16),\n",
    "    nn.ReLU(),\n",
    "    conv(16, 10), # 1\n",
    "    nn.BatchNorm2d(10),\n",
    "    Flatten()     # remove (1,1) grid\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = Learner(data, model, loss_func = nn.CrossEntropyLoss(), metrics=accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequential\n",
      "======================================================================\n",
      "Layer (type)         Output Shape         Param #    Trainable \n",
      "======================================================================\n",
      "Conv2d               [8, 14, 14]          80         True      \n",
      "______________________________________________________________________\n",
      "BatchNorm2d          [8, 14, 14]          16         True      \n",
      "______________________________________________________________________\n",
      "ReLU                 [8, 14, 14]          0          False     \n",
      "______________________________________________________________________\n",
      "Conv2d               [16, 7, 7]           1,168      True      \n",
      "______________________________________________________________________\n",
      "BatchNorm2d          [16, 7, 7]           32         True      \n",
      "______________________________________________________________________\n",
      "ReLU                 [16, 7, 7]           0          False     \n",
      "______________________________________________________________________\n",
      "Conv2d               [32, 4, 4]           4,640      True      \n",
      "______________________________________________________________________\n",
      "BatchNorm2d          [32, 4, 4]           64         True      \n",
      "______________________________________________________________________\n",
      "ReLU                 [32, 4, 4]           0          False     \n",
      "______________________________________________________________________\n",
      "Conv2d               [16, 2, 2]           4,624      True      \n",
      "______________________________________________________________________\n",
      "BatchNorm2d          [16, 2, 2]           32         True      \n",
      "______________________________________________________________________\n",
      "ReLU                 [16, 2, 2]           0          False     \n",
      "______________________________________________________________________\n",
      "Conv2d               [10, 1, 1]           1,450      True      \n",
      "______________________________________________________________________\n",
      "BatchNorm2d          [10, 1, 1]           20         True      \n",
      "______________________________________________________________________\n",
      "Flatten              [10]                 0          False     \n",
      "______________________________________________________________________\n",
      "\n",
      "Total params: 12,126\n",
      "Total trainable params: 12,126\n",
      "Total non-trainable params: 0\n",
      "Optimized with 'torch.optim.adam.Adam', betas=(0.9, 0.99)\n",
      "Using true weight decay as discussed in https://www.fast.ai/2018/07/02/adam-weight-decay/ \n",
      "Loss function : CrossEntropyLoss\n",
      "======================================================================\n",
      "Callbacks functions applied \n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(learn.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([28, 10])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xb = xb.cuda()\n",
    "model(xb).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='0' class='' max='1', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      0.00% [0/1 00:00<00:00]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='99' class='' max='2142', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      4.62% [99/2142 00:35<12:03 3.6399]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    }
   ],
   "source": [
    "learn.lr_find(end_lr=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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": [
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.358661</td>\n",
       "      <td>0.303763</td>\n",
       "      <td>0.902700</td>\n",
       "      <td>01:55</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.173309</td>\n",
       "      <td>0.122332</td>\n",
       "      <td>0.961500</td>\n",
       "      <td>01:50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.095996</td>\n",
       "      <td>0.047142</td>\n",
       "      <td>0.985600</td>\n",
       "      <td>01:52</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(3, max_lr=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Using fastai version"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conv2(ni,nf): return conv_layer(ni,nf,stride=2)\n",
    "model = nn.Sequential(\n",
    "    conv2(1, 8),   # 14\n",
    "    conv2(8, 16),  # 7\n",
    "    conv2(16, 32), # 4\n",
    "    conv2(32, 16), # 2\n",
    "    conv2(16, 10), # 1\n",
    "    Flatten()      # remove (1,1) grid\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = Learner(data, model, loss_func = nn.CrossEntropyLoss(), metrics=accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='0' class='' max='1', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      0.00% [0/1 00:00<00:00]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='99' class='' max='2142', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      4.62% [99/2142 00:34<11:49 3.6789]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    },
    {
     "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": [
    "learn.lr_find(end_lr=100)\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.323802</td>\n",
       "      <td>0.154137</td>\n",
       "      <td>0.954100</td>\n",
       "      <td>01:43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.265099</td>\n",
       "      <td>0.160194</td>\n",
       "      <td>0.949700</td>\n",
       "      <td>01:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.290638</td>\n",
       "      <td>0.313892</td>\n",
       "      <td>0.904300</td>\n",
       "      <td>01:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.285804</td>\n",
       "      <td>0.596903</td>\n",
       "      <td>0.801100</td>\n",
       "      <td>01:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.300262</td>\n",
       "      <td>0.325268</td>\n",
       "      <td>0.898900</td>\n",
       "      <td>01:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.229259</td>\n",
       "      <td>0.146741</td>\n",
       "      <td>0.957200</td>\n",
       "      <td>02:01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.183178</td>\n",
       "      <td>0.090863</td>\n",
       "      <td>0.974000</td>\n",
       "      <td>01:52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.106404</td>\n",
       "      <td>0.051761</td>\n",
       "      <td>0.984200</td>\n",
       "      <td>01:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.091911</td>\n",
       "      <td>0.039558</td>\n",
       "      <td>0.989100</td>\n",
       "      <td>01:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.083305</td>\n",
       "      <td>0.033296</td>\n",
       "      <td>0.989200</td>\n",
       "      <td>01:50</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(10, max_lr=0.1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resnet-ish --> #skip connection."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = nn.Sequential(\n",
    "    conv2(1, 8),\n",
    "    res_block(8),\n",
    "    conv2(8, 16),\n",
    "    res_block(16),\n",
    "    conv2(16, 32),\n",
    "    res_block(32),\n",
    "    conv2(32, 16),\n",
    "    res_block(16),\n",
    "    conv2(16, 10),\n",
    "    Flatten()\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = Learner(data, model, loss_func = nn.CrossEntropyLoss(), metrics=accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='0' class='' max='1', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      0.00% [0/1 00:00<00:00]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='99' class='' max='2142', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      4.62% [99/2142 00:39<13:29 4.1272]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    },
    {
     "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": [
    "learn.lr_find(end_lr=100)\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.228893</td>\n",
       "      <td>0.136044</td>\n",
       "      <td>0.965100</td>\n",
       "      <td>02:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.170586</td>\n",
       "      <td>0.104724</td>\n",
       "      <td>0.969300</td>\n",
       "      <td>02:25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.163270</td>\n",
       "      <td>0.165926</td>\n",
       "      <td>0.964100</td>\n",
       "      <td>02:25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.148630</td>\n",
       "      <td>3.654489</td>\n",
       "      <td>0.951700</td>\n",
       "      <td>02:25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.137074</td>\n",
       "      <td>0.066778</td>\n",
       "      <td>0.980700</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.132727</td>\n",
       "      <td>0.081799</td>\n",
       "      <td>0.975500</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.103128</td>\n",
       "      <td>0.078956</td>\n",
       "      <td>0.977900</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.104520</td>\n",
       "      <td>0.075536</td>\n",
       "      <td>0.978000</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.061379</td>\n",
       "      <td>0.040652</td>\n",
       "      <td>0.987500</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.037959</td>\n",
       "      <td>0.023798</td>\n",
       "      <td>0.992800</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.032771</td>\n",
       "      <td>0.016457</td>\n",
       "      <td>0.995500</td>\n",
       "      <td>02:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.026977</td>\n",
       "      <td>0.015921</td>\n",
       "      <td>0.994900</td>\n",
       "      <td>02:31</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(12, max_lr=0.05)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}