iwildcam-2019-fgvc6_baseline.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "# 多行输出\n",
    "from IPython.core.interactiveshell import InteractiveShell\n",
    "InteractiveShell.ast_node_interactivity = \"all\" "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1 数据准备"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6')"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "'/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6'"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 设置数据路径\n",
    "path = Path('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6')\n",
    "path\n",
    "path.as_posix()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/train.csv'),\n",
       " PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/test.csv'),\n",
       " PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/train'),\n",
       " PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/sample_submission.csv'),\n",
       " PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/test'),\n",
       " PosixPath('/home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6/models')]"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "path.ls()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "name_id = ['empty', 'deer', 'moose', 'squirrel', 'rodent', 'small_mammal', 'elk', 'pronghorn_antelope', 'rabbit,bighorn_sheep', 'fox', 'coyote', 'black_bear', \n",
    "           'raccoon', 'skunk', 'wolf', 'bobcat', 'cat', 'dog', 'opossum', 'bison', 'mountain_goat', 'mountain_lion']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 这里每个图片都会有多个标签"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>category_id</th>\n",
       "      <th>date_captured</th>\n",
       "      <th>file_name</th>\n",
       "      <th>frame_num</th>\n",
       "      <th>id</th>\n",
       "      <th>location</th>\n",
       "      <th>rights_holder</th>\n",
       "      <th>seq_id</th>\n",
       "      <th>seq_num_frames</th>\n",
       "      <th>width</th>\n",
       "      <th>height</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>19</td>\n",
       "      <td>2011-05-13 23:43:18</td>\n",
       "      <td>5998cfa4-23d2-11e8-a6a3-ec086b02610b.jpg</td>\n",
       "      <td>1</td>\n",
       "      <td>5998cfa4-23d2-11e8-a6a3-ec086b02610b</td>\n",
       "      <td>33</td>\n",
       "      <td>Justin Brown</td>\n",
       "      <td>6f084ccc-5567-11e8-bc84-dca9047ef277</td>\n",
       "      <td>3</td>\n",
       "      <td>1024</td>\n",
       "      <td>747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>19</td>\n",
       "      <td>2012-03-17 03:48:44</td>\n",
       "      <td>588a679f-23d2-11e8-a6a3-ec086b02610b.jpg</td>\n",
       "      <td>2</td>\n",
       "      <td>588a679f-23d2-11e8-a6a3-ec086b02610b</td>\n",
       "      <td>115</td>\n",
       "      <td>Justin Brown</td>\n",
       "      <td>6f12067d-5567-11e8-b3c0-dca9047ef277</td>\n",
       "      <td>3</td>\n",
       "      <td>1024</td>\n",
       "      <td>747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>2014-05-11 11:56:46</td>\n",
       "      <td>59279ce3-23d2-11e8-a6a3-ec086b02610b.jpg</td>\n",
       "      <td>1</td>\n",
       "      <td>59279ce3-23d2-11e8-a6a3-ec086b02610b</td>\n",
       "      <td>96</td>\n",
       "      <td>Erin Boydston</td>\n",
       "      <td>6faa92d1-5567-11e8-b1ae-dca9047ef277</td>\n",
       "      <td>1</td>\n",
       "      <td>1024</td>\n",
       "      <td>747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>2013-10-06 02:00:00</td>\n",
       "      <td>5a2af4ab-23d2-11e8-a6a3-ec086b02610b.jpg</td>\n",
       "      <td>1</td>\n",
       "      <td>5a2af4ab-23d2-11e8-a6a3-ec086b02610b</td>\n",
       "      <td>57</td>\n",
       "      <td>Erin Boydston</td>\n",
       "      <td>6f7d4702-5567-11e8-9e03-dca9047ef277</td>\n",
       "      <td>1</td>\n",
       "      <td>1024</td>\n",
       "      <td>747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>2011-07-12 13:11:16</td>\n",
       "      <td>599fbd89-23d2-11e8-a6a3-ec086b02610b.jpg</td>\n",
       "      <td>3</td>\n",
       "      <td>599fbd89-23d2-11e8-a6a3-ec086b02610b</td>\n",
       "      <td>46</td>\n",
       "      <td>Justin Brown</td>\n",
       "      <td>6f1728a1-5567-11e8-9be7-dca9047ef277</td>\n",
       "      <td>3</td>\n",
       "      <td>1024</td>\n",
       "      <td>747</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   category_id        date_captured                                 file_name  \\\n",
       "0           19  2011-05-13 23:43:18  5998cfa4-23d2-11e8-a6a3-ec086b02610b.jpg   \n",
       "1           19  2012-03-17 03:48:44  588a679f-23d2-11e8-a6a3-ec086b02610b.jpg   \n",
       "2            0  2014-05-11 11:56:46  59279ce3-23d2-11e8-a6a3-ec086b02610b.jpg   \n",
       "3            0  2013-10-06 02:00:00  5a2af4ab-23d2-11e8-a6a3-ec086b02610b.jpg   \n",
       "4            0  2011-07-12 13:11:16  599fbd89-23d2-11e8-a6a3-ec086b02610b.jpg   \n",
       "\n",
       "   frame_num                                    id  location  rights_holder  \\\n",
       "0          1  5998cfa4-23d2-11e8-a6a3-ec086b02610b        33   Justin Brown   \n",
       "1          2  588a679f-23d2-11e8-a6a3-ec086b02610b       115   Justin Brown   \n",
       "2          1  59279ce3-23d2-11e8-a6a3-ec086b02610b        96  Erin Boydston   \n",
       "3          1  5a2af4ab-23d2-11e8-a6a3-ec086b02610b        57  Erin Boydston   \n",
       "4          3  599fbd89-23d2-11e8-a6a3-ec086b02610b        46   Justin Brown   \n",
       "\n",
       "                                 seq_id  seq_num_frames  width  height  \n",
       "0  6f084ccc-5567-11e8-bc84-dca9047ef277               3   1024     747  \n",
       "1  6f12067d-5567-11e8-b3c0-dca9047ef277               3   1024     747  \n",
       "2  6faa92d1-5567-11e8-b1ae-dca9047ef277               1   1024     747  \n",
       "3  6f7d4702-5567-11e8-9e03-dca9047ef277               1   1024     747  \n",
       "4  6f1728a1-5567-11e8-9be7-dca9047ef277               3   1024     747  "
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = pd.read_csv(path/'train.csv')\n",
    "train_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>date_captured</th>\n",
       "      <th>file_name</th>\n",
       "      <th>frame_num</th>\n",
       "      <th>id</th>\n",
       "      <th>location</th>\n",
       "      <th>rights_holder</th>\n",
       "      <th>seq_id</th>\n",
       "      <th>seq_num_frames</th>\n",
       "      <th>width</th>\n",
       "      <th>height</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2016-01-03 11:30:56</td>\n",
       "      <td>bce932f6-2bf6-11e9-bcad-06f10d5896c4.jpg</td>\n",
       "      <td>1</td>\n",
       "      <td>bce932f6-2bf6-11e9-bcad-06f10d5896c4</td>\n",
       "      <td>37</td>\n",
       "      <td>Idaho Department of Fish and Game</td>\n",
       "      <td>6e9ac61c-2e32-11e9-90ef-dca9047ef277</td>\n",
       "      <td>5</td>\n",
       "      <td>1024</td>\n",
       "      <td>726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2016-01-03 11:30:57</td>\n",
       "      <td>bce932f7-2bf6-11e9-bcad-06f10d5896c4.jpg</td>\n",
       "      <td>2</td>\n",
       "      <td>bce932f7-2bf6-11e9-bcad-06f10d5896c4</td>\n",
       "      <td>37</td>\n",
       "      <td>Idaho Department of Fish and Game</td>\n",
       "      <td>6e9ac61c-2e32-11e9-90ef-dca9047ef277</td>\n",
       "      <td>5</td>\n",
       "      <td>1024</td>\n",
       "      <td>726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2016-01-03 11:30:58</td>\n",
       "      <td>bce932f8-2bf6-11e9-bcad-06f10d5896c4.jpg</td>\n",
       "      <td>3</td>\n",
       "      <td>bce932f8-2bf6-11e9-bcad-06f10d5896c4</td>\n",
       "      <td>37</td>\n",
       "      <td>Idaho Department of Fish and Game</td>\n",
       "      <td>6e9ac61c-2e32-11e9-90ef-dca9047ef277</td>\n",
       "      <td>5</td>\n",
       "      <td>1024</td>\n",
       "      <td>726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2016-01-03 11:30:59</td>\n",
       "      <td>bce932f9-2bf6-11e9-bcad-06f10d5896c4.jpg</td>\n",
       "      <td>4</td>\n",
       "      <td>bce932f9-2bf6-11e9-bcad-06f10d5896c4</td>\n",
       "      <td>37</td>\n",
       "      <td>Idaho Department of Fish and Game</td>\n",
       "      <td>6e9ac61c-2e32-11e9-90ef-dca9047ef277</td>\n",
       "      <td>5</td>\n",
       "      <td>1024</td>\n",
       "      <td>726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2016-01-03 11:31:00</td>\n",
       "      <td>bce932fa-2bf6-11e9-bcad-06f10d5896c4.jpg</td>\n",
       "      <td>5</td>\n",
       "      <td>bce932fa-2bf6-11e9-bcad-06f10d5896c4</td>\n",
       "      <td>37</td>\n",
       "      <td>Idaho Department of Fish and Game</td>\n",
       "      <td>6e9ac61c-2e32-11e9-90ef-dca9047ef277</td>\n",
       "      <td>5</td>\n",
       "      <td>1024</td>\n",
       "      <td>726</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date_captured                                 file_name  frame_num  \\\n",
       "0  2016-01-03 11:30:56  bce932f6-2bf6-11e9-bcad-06f10d5896c4.jpg          1   \n",
       "1  2016-01-03 11:30:57  bce932f7-2bf6-11e9-bcad-06f10d5896c4.jpg          2   \n",
       "2  2016-01-03 11:30:58  bce932f8-2bf6-11e9-bcad-06f10d5896c4.jpg          3   \n",
       "3  2016-01-03 11:30:59  bce932f9-2bf6-11e9-bcad-06f10d5896c4.jpg          4   \n",
       "4  2016-01-03 11:31:00  bce932fa-2bf6-11e9-bcad-06f10d5896c4.jpg          5   \n",
       "\n",
       "                                     id  location  \\\n",
       "0  bce932f6-2bf6-11e9-bcad-06f10d5896c4        37   \n",
       "1  bce932f7-2bf6-11e9-bcad-06f10d5896c4        37   \n",
       "2  bce932f8-2bf6-11e9-bcad-06f10d5896c4        37   \n",
       "3  bce932f9-2bf6-11e9-bcad-06f10d5896c4        37   \n",
       "4  bce932fa-2bf6-11e9-bcad-06f10d5896c4        37   \n",
       "\n",
       "                       rights_holder                                seq_id  \\\n",
       "0  Idaho Department of Fish and Game  6e9ac61c-2e32-11e9-90ef-dca9047ef277   \n",
       "1  Idaho Department of Fish and Game  6e9ac61c-2e32-11e9-90ef-dca9047ef277   \n",
       "2  Idaho Department of Fish and Game  6e9ac61c-2e32-11e9-90ef-dca9047ef277   \n",
       "3  Idaho Department of Fish and Game  6e9ac61c-2e32-11e9-90ef-dca9047ef277   \n",
       "4  Idaho Department of Fish and Game  6e9ac61c-2e32-11e9-90ef-dca9047ef277   \n",
       "\n",
       "   seq_num_frames  width  height  \n",
       "0               5   1024     726  \n",
       "1               5   1024     726  \n",
       "2               5   1024     726  \n",
       "3               5   1024     726  \n",
       "4               5   1024     726  "
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_csv(path/'test.csv')\n",
    "test_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 图片变换\n",
    "tfms = get_transforms(do_flip=True, max_rotate=20, max_zoom=1.3, max_lighting=0.4,\n",
    "                      max_warp=0.4, p_affine=1., p_lighting=1.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_set = ImageList.from_df(test_df, path=path, cols='id', folder='test', suffix='.jpg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 构建数据集\n",
    "np.random.seed(42)\n",
    "# 使用 ImageList 是因为图像是多标签的\n",
    "src = (ImageList.from_df(train_df, path=path, folder='train', cols='id', suffix='.jpg')\n",
    "       .split_by_rand_pct(0.1)\n",
    "       .label_from_df(cols='category_id')\n",
    "       .add_test(test_set)\n",
    "      )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 图像大小设置为 32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_size=32\n",
    "bs=64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (src.transform(tfms, size=img_size)\n",
    "        .databunch().normalize(imagenet_stats))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ImageDataBunch;\n",
       "\n",
       "Train: LabelList (176670 items)\n",
       "x: ImageList\n",
       "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
       "y: CategoryList\n",
       "19,19,0,0,3\n",
       "Path: /home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6;\n",
       "\n",
       "Valid: LabelList (19629 items)\n",
       "x: ImageList\n",
       "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
       "y: CategoryList\n",
       "0,0,0,19,0\n",
       "Path: /home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6;\n",
       "\n",
       "Test: LabelList (153730 items)\n",
       "x: ImageList\n",
       "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
       "y: EmptyLabelList\n",
       ",,,,\n",
       "Path: /home/lyc/Datasets/kaggle/competitions/iwildcam-2019-fgvc6"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "14"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.c"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x648 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 查看部分数据\n",
    "data.show_batch(rows=3, figsize=(12,9))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 使用 densenet169 作为模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "arch = models.densenet169"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = cnn_learner(data, arch, metrics=[fbeta, accuracy])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 寻找学习率"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "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()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 查看学习率情况并作出选择"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "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": "markdown",
   "metadata": {},
   "source": [
    "- 根据上述图像，我们选择学习率为 0.01 Find the thing with the steepest slope "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 0.01"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 训练模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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='5', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      0.00% [0/5 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>fbeta</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='717' class='' max='2760', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      25.98% [717/2760 17:28<49:46 0.8969]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(5, lr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 保存第一阶段的结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('stage-1')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- unfreeze 并训练浅层网络的权重，微调过程"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 再次查找最佳学习率"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 根据上述图，选择 lr1=1e-4\n",
    "- 再次训练浅层网络并保存结果\n",
    "- 使用判别式学习率"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr1 = 1e-4\n",
    "learn.fit_one_cycle(10, slice(lr1, lr/5))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('stage-2')\n",
    "learn.export()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 图像大小设置为 128"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 为了避免内存超出，这里将上述训练好的模型再次导入，并清理内存\n",
    "- 仍旧使用之前的学习器，只是简单地更换数据"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('stage-2');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs=32\n",
    "img_sz=128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (src.transform(tfms, size=img_sz)\n",
    "        .databunch(bs=bs).normalize(imagenet_stats))\n",
    "\n",
    "learn.data = data\n",
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data.show_batch(rows=3, figsize=(12,9))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 我们仍然使用之前定义的模型\n",
    "- 再次训练之前，我们需要将浅层结构进行 freeze，不训练浅层结构的权重"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 寻找最佳学习率并再次训练"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr2=1e-3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit_one_cycle(10, slice(lr2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('stage-1-128')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 继续进行第二阶段的微调"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr3 = 1e-4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit_one_cycle(10, slice(lr3, lr2/5))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 查看训练损失"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 保存结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('stage-2-128')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.export()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 继续增大图片尺寸"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('stage-2-128');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_sz=128\n",
    "bs=16"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (src.transform(tfms, size=img_sz)\n",
    "        .databunch(bs=bs).normalize(imagenet_stats))\n",
    "\n",
    "learn.data = data\n",
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data.show_batch(rows=3, figsize=(12,9))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 测试集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test = ImageList.from_folder(path/'test-jpg').add(ImageList.from_folder(path/'test-jpg-additional'))\n",
    "len(test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- 重新加载学习器"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = load_learner(path, test=test)\n",
    "preds, _ = learn.get_preds(ds_type=DatasetType.Test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "thresh = 0.2\n",
    "labelled_preds = [' '.join([learn.data.classes[i] for i,p in enumerate(pred) if p > thresh]) for pred in preds]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "labelled_preds[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fnames = [f.name[:-4] for f in learn.data.test_ds.items]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame({'image_name':fnames, 'tags':labelled_preds}, columns=['image_name', 'tags'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.to_csv(path/'submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 提交结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!kaggle competitions submit -c iwildcam-2019-fgvc6 -f {path}/submission.csv -m \"submit01\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}