resnet50_basic.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from pathlib import Path\n",
    "import numpy as np\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "from fastai import *\n",
    "from fastai.vision import *\n",
    "\n",
    "from dataset import ImageMulti4Channel\n",
    "from resnet import Resnet4Channel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs = 64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "path = Path('/home/william/data/human-protein-atlas-image-classification/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "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>Id</th>\n",
       "      <th>Target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>00070df0-bbc3-11e8-b2bc-ac1f6b6435d0</td>\n",
       "      <td>16 0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>000a6c98-bb9b-11e8-b2b9-ac1f6b6435d0</td>\n",
       "      <td>7 1 2 0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>000a9596-bbc4-11e8-b2bc-ac1f6b6435d0</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>000c99ba-bba4-11e8-b2b9-ac1f6b6435d0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>001838f8-bbca-11e8-b2bc-ac1f6b6435d0</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                     Id   Target\n",
       "0  00070df0-bbc3-11e8-b2bc-ac1f6b6435d0     16 0\n",
       "1  000a6c98-bb9b-11e8-b2b9-ac1f6b6435d0  7 1 2 0\n",
       "2  000a9596-bbc4-11e8-b2bc-ac1f6b6435d0        5\n",
       "3  000c99ba-bba4-11e8-b2b9-ac1f6b6435d0        1\n",
       "4  001838f8-bbca-11e8-b2bc-ac1f6b6435d0       18"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv(path/'train.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "fns = pd.Series([id + '.png' for id in df.Id])\n",
    "labels = [targ.split(' ') for targ in df.Target]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(42)\n",
    "trn_ds, val_ds = ImageMulti4Channel.from_folder(path, 'train', fns, labels, classes=[str(i) for i in range(28)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_ids = list(sorted({fname.split('_')[0] for fname in os.listdir(path/'test')}))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_ds,_ = ImageMulti4Channel.from_folder(\n",
    "    path, \n",
    "    'test', \n",
    "    pd.Series(test_ids), \n",
    "    [['0'] for _ in range(len(test_ids))],\n",
    "    valid_pct=0,\n",
    "    classes=['0'],\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "trn_tfms,_ = get_transforms(do_flip=True, flip_vert=True, max_rotate=30., max_zoom=1,\n",
    "                      max_lighting=0.05, max_warp=0.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "protein_stats = ([0.08069, 0.05258, 0.05487, 0.08282], [0.13704, 0.10145, 0.15313, 0.13814])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = ImageDataBunch.create(trn_ds, val_ds, test_ds=test_ds, path=path, bs=bs, ds_tfms=(trn_tfms, []), num_workers=8, size=224).normalize(protein_stats)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "resnet50 = Resnet4Channel(encoder_depth=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "f1_score = partial(fbeta, thresh=0.2, beta=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = ClassificationLearner(\n",
    "    data=data,\n",
    "    model=resnet50,\n",
    "    loss_func=F.binary_cross_entropy_with_logits,\n",
    "    path=path,    \n",
    "    metrics=[f1_score], \n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    }
   ],
   "source": [
    "learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "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": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 0.01"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total time: 53:05\n",
      "epoch  train_loss  valid_loss  fbeta   \n",
      "1      0.130200    0.146852    0.388930  (02:41)\n",
      "2      0.121962    0.148459    0.442670  (02:40)\n",
      "3      0.120659    0.125784    0.452106  (02:40)\n",
      "4      0.118589    0.129387    0.428716  (02:39)\n",
      "5      0.115391    0.163984    0.447932  (02:39)\n",
      "6      0.114821    0.138222    0.491987  (02:39)\n",
      "7      0.109604    0.136291    0.525457  (02:39)\n",
      "8      0.107522    0.119311    0.522222  (02:39)\n",
      "9      0.101589    0.131579    0.490353  (02:37)\n",
      "10     0.100474    0.108896    0.571583  (02:39)\n",
      "11     0.097399    0.108986    0.582967  (02:38)\n",
      "12     0.093851    0.106814    0.615101  (02:38)\n",
      "13     0.089337    0.102818    0.620778  (02:38)\n",
      "14     0.088372    0.096513    0.616014  (02:38)\n",
      "15     0.084325    0.089300    0.680281  (02:39)\n",
      "16     0.080916    0.086166    0.680590  (02:38)\n",
      "17     0.078836    0.082739    0.689119  (02:38)\n",
      "18     0.077355    0.080840    0.695626  (02:39)\n",
      "19     0.076507    0.079993    0.702401  (02:39)\n",
      "20     0.074385    0.079752    0.704166  (02:38)\n",
      "\n"
     ]
    }
   ],
   "source": [
    "learn.fit_one_cycle(20, slice(lr))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('resnet50_basic')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/william/anaconda3/envs/fastaiv1/lib/python3.7/site-packages/torch/nn/functional.py:1124: UserWarning: nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.\n",
      "  warnings.warn(\"nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.\")\n"
     ]
    }
   ],
   "source": [
    "preds,_ = learn.get_preds(DatasetType.Test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_labels = [' '.join(list([str(i) for i in np.nonzero(row>0.2)[0]])) for row in np.array(preds)]\n",
    "df = pd.DataFrame({'Id':test_ids,'Predicted':pred_labels})\n",
    "df.to_csv(path/'protein_predictions.csv', header=True, index=False)"
   ]
  }
 ],
 "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}