|
a |
|
b/Serialized/train_base_models.ipynb |
|
|
1 |
{ |
|
|
2 |
"cells": [ |
|
|
3 |
{ |
|
|
4 |
"cell_type": "code", |
|
|
5 |
"execution_count": 1, |
|
|
6 |
"metadata": {}, |
|
|
7 |
"outputs": [ |
|
|
8 |
{ |
|
|
9 |
"name": "stderr", |
|
|
10 |
"output_type": "stream", |
|
|
11 |
"text": [ |
|
|
12 |
"/home/reina/anaconda3/envs/RSNA/lib/python3.6/importlib/_bootstrap.py:219: ImportWarning: can't resolve package from __spec__ or __package__, falling back on __name__ and __path__\n", |
|
|
13 |
" return f(*args, **kwds)\n", |
|
|
14 |
"/home/reina/anaconda3/envs/RSNA/lib/python3.6/importlib/_bootstrap.py:219: ImportWarning: can't resolve package from __spec__ or __package__, falling back on __name__ and __path__\n", |
|
|
15 |
" return f(*args, **kwds)\n" |
|
|
16 |
] |
|
|
17 |
} |
|
|
18 |
], |
|
|
19 |
"source": [ |
|
|
20 |
"from __future__ import absolute_import\n", |
|
|
21 |
"from __future__ import division\n", |
|
|
22 |
"from __future__ import print_function\n", |
|
|
23 |
"\n", |
|
|
24 |
"\n", |
|
|
25 |
"import numpy as np # linear algebra\n", |
|
|
26 |
"import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n", |
|
|
27 |
"import os\n", |
|
|
28 |
"import datetime\n", |
|
|
29 |
"import seaborn as sns\n", |
|
|
30 |
"import pydicom\n", |
|
|
31 |
"import time\n", |
|
|
32 |
"import gc\n", |
|
|
33 |
"import operator \n", |
|
|
34 |
"from apex import amp \n", |
|
|
35 |
"import matplotlib.pyplot as plt\n", |
|
|
36 |
"import torch\n", |
|
|
37 |
"import torch.nn as nn\n", |
|
|
38 |
"import torch.utils.data as D\n", |
|
|
39 |
"import torch.nn.functional as F\n", |
|
|
40 |
"from sklearn.model_selection import KFold\n", |
|
|
41 |
"from tqdm import tqdm, tqdm_notebook\n", |
|
|
42 |
"from IPython.core.interactiveshell import InteractiveShell\n", |
|
|
43 |
"InteractiveShell.ast_node_interactivity = \"all\"\n", |
|
|
44 |
"import warnings\n", |
|
|
45 |
"warnings.filterwarnings(action='once')\n", |
|
|
46 |
"import pickle\n", |
|
|
47 |
"%load_ext autoreload\n", |
|
|
48 |
"%autoreload 2\n", |
|
|
49 |
"%matplotlib inline\n", |
|
|
50 |
"from skimage.io import imread,imshow\n", |
|
|
51 |
"from helper import *\n", |
|
|
52 |
"from apex import amp\n", |
|
|
53 |
"import helper\n", |
|
|
54 |
"import torchvision.models as models\n", |
|
|
55 |
"import pretrainedmodels\n", |
|
|
56 |
"from torch.optim import Adam\n", |
|
|
57 |
"from functools import partial\n", |
|
|
58 |
"from defenitions import *" |
|
|
59 |
] |
|
|
60 |
}, |
|
|
61 |
{ |
|
|
62 |
"cell_type": "markdown", |
|
|
63 |
"metadata": {}, |
|
|
64 |
"source": [ |
|
|
65 |
"## Set parameters below" |
|
|
66 |
] |
|
|
67 |
}, |
|
|
68 |
{ |
|
|
69 |
"cell_type": "code", |
|
|
70 |
"execution_count": 11, |
|
|
71 |
"metadata": {}, |
|
|
72 |
"outputs": [], |
|
|
73 |
"source": [ |
|
|
74 |
"# here you should set which model parameters you want to choose (see definitions.py) and what GPU to use\n", |
|
|
75 |
"params=parameters['se_resnet101_5'] # se_resnet101_5, se_resnext101_32x4d_3, se_resnext101_32x4d_5\n", |
|
|
76 |
"\n", |
|
|
77 |
"device=device_by_name(\"Tesla\") # RTX , cpu\n", |
|
|
78 |
"torch.cuda.set_device(device)\n", |
|
|
79 |
"sendmeemail=Email_Progress(my_gmail,my_pass,to_email,'{} results'.format(params['model_name']))" |
|
|
80 |
] |
|
|
81 |
}, |
|
|
82 |
{ |
|
|
83 |
"cell_type": "code", |
|
|
84 |
"execution_count": 12, |
|
|
85 |
"metadata": {}, |
|
|
86 |
"outputs": [ |
|
|
87 |
{ |
|
|
88 |
"data": { |
|
|
89 |
"text/plain": [ |
|
|
90 |
"{'model_name': 'se_resnet101',\n", |
|
|
91 |
" 'SEED': 432,\n", |
|
|
92 |
" 'n_splits': 5,\n", |
|
|
93 |
" 'Pre_version': None,\n", |
|
|
94 |
" 'focal': False,\n", |
|
|
95 |
" 'version': 'new_splits',\n", |
|
|
96 |
" 'train_prediction': 'predictions_train_tta',\n", |
|
|
97 |
" 'train_features': 'features_train_tta',\n", |
|
|
98 |
" 'test_prediction': 'predictions_test',\n", |
|
|
99 |
" 'test_features': 'features_test',\n", |
|
|
100 |
" 'num_epochs': 5,\n", |
|
|
101 |
" 'num_pool': 8}" |
|
|
102 |
] |
|
|
103 |
}, |
|
|
104 |
"execution_count": 12, |
|
|
105 |
"metadata": {}, |
|
|
106 |
"output_type": "execute_result" |
|
|
107 |
} |
|
|
108 |
], |
|
|
109 |
"source": [ |
|
|
110 |
"params" |
|
|
111 |
] |
|
|
112 |
}, |
|
|
113 |
{ |
|
|
114 |
"cell_type": "code", |
|
|
115 |
"execution_count": 13, |
|
|
116 |
"metadata": {}, |
|
|
117 |
"outputs": [], |
|
|
118 |
"source": [ |
|
|
119 |
"SEED = params['SEED']\n", |
|
|
120 |
"n_splits=params['n_splits']" |
|
|
121 |
] |
|
|
122 |
}, |
|
|
123 |
{ |
|
|
124 |
"cell_type": "code", |
|
|
125 |
"execution_count": 14, |
|
|
126 |
"metadata": {}, |
|
|
127 |
"outputs": [ |
|
|
128 |
{ |
|
|
129 |
"data": { |
|
|
130 |
"text/plain": [ |
|
|
131 |
"(674252, 15)" |
|
|
132 |
] |
|
|
133 |
}, |
|
|
134 |
"execution_count": 14, |
|
|
135 |
"metadata": {}, |
|
|
136 |
"output_type": "execute_result" |
|
|
137 |
}, |
|
|
138 |
{ |
|
|
139 |
"data": { |
|
|
140 |
"text/plain": [ |
|
|
141 |
"(674252, 15)" |
|
|
142 |
] |
|
|
143 |
}, |
|
|
144 |
"execution_count": 14, |
|
|
145 |
"metadata": {}, |
|
|
146 |
"output_type": "execute_result" |
|
|
147 |
}, |
|
|
148 |
{ |
|
|
149 |
"data": { |
|
|
150 |
"text/html": [ |
|
|
151 |
"<div>\n", |
|
|
152 |
"<style scoped>\n", |
|
|
153 |
" .dataframe tbody tr th:only-of-type {\n", |
|
|
154 |
" vertical-align: middle;\n", |
|
|
155 |
" }\n", |
|
|
156 |
"\n", |
|
|
157 |
" .dataframe tbody tr th {\n", |
|
|
158 |
" vertical-align: top;\n", |
|
|
159 |
" }\n", |
|
|
160 |
"\n", |
|
|
161 |
" .dataframe thead th {\n", |
|
|
162 |
" text-align: right;\n", |
|
|
163 |
" }\n", |
|
|
164 |
"</style>\n", |
|
|
165 |
"<table border=\"1\" class=\"dataframe\">\n", |
|
|
166 |
" <thead>\n", |
|
|
167 |
" <tr style=\"text-align: right;\">\n", |
|
|
168 |
" <th></th>\n", |
|
|
169 |
" <th>PatientID</th>\n", |
|
|
170 |
" <th>epidural</th>\n", |
|
|
171 |
" <th>intraparenchymal</th>\n", |
|
|
172 |
" <th>intraventricular</th>\n", |
|
|
173 |
" <th>subarachnoid</th>\n", |
|
|
174 |
" <th>subdural</th>\n", |
|
|
175 |
" <th>any</th>\n", |
|
|
176 |
" <th>PID</th>\n", |
|
|
177 |
" <th>StudyI</th>\n", |
|
|
178 |
" <th>SeriesI</th>\n", |
|
|
179 |
" <th>WindowCenter</th>\n", |
|
|
180 |
" <th>WindowWidth</th>\n", |
|
|
181 |
" <th>ImagePositionZ</th>\n", |
|
|
182 |
" <th>ImagePositionX</th>\n", |
|
|
183 |
" <th>ImagePositionY</th>\n", |
|
|
184 |
" </tr>\n", |
|
|
185 |
" </thead>\n", |
|
|
186 |
" <tbody>\n", |
|
|
187 |
" <tr>\n", |
|
|
188 |
" <th>0</th>\n", |
|
|
189 |
" <td>63eb1e259</td>\n", |
|
|
190 |
" <td>0</td>\n", |
|
|
191 |
" <td>0</td>\n", |
|
|
192 |
" <td>0</td>\n", |
|
|
193 |
" <td>0</td>\n", |
|
|
194 |
" <td>0</td>\n", |
|
|
195 |
" <td>0</td>\n", |
|
|
196 |
" <td>a449357f</td>\n", |
|
|
197 |
" <td>62d125e5b2</td>\n", |
|
|
198 |
" <td>0be5c0d1b3</td>\n", |
|
|
199 |
" <td>['00036', '00036']</td>\n", |
|
|
200 |
" <td>['00080', '00080']</td>\n", |
|
|
201 |
" <td>180.199951</td>\n", |
|
|
202 |
" <td>-125.0</td>\n", |
|
|
203 |
" <td>-8.000000</td>\n", |
|
|
204 |
" </tr>\n", |
|
|
205 |
" <tr>\n", |
|
|
206 |
" <th>1</th>\n", |
|
|
207 |
" <td>2669954a7</td>\n", |
|
|
208 |
" <td>0</td>\n", |
|
|
209 |
" <td>0</td>\n", |
|
|
210 |
" <td>0</td>\n", |
|
|
211 |
" <td>0</td>\n", |
|
|
212 |
" <td>0</td>\n", |
|
|
213 |
" <td>0</td>\n", |
|
|
214 |
" <td>363d5865</td>\n", |
|
|
215 |
" <td>a20b80c7bf</td>\n", |
|
|
216 |
" <td>3564d584db</td>\n", |
|
|
217 |
" <td>['00047', '00047']</td>\n", |
|
|
218 |
" <td>['00080', '00080']</td>\n", |
|
|
219 |
" <td>922.530821</td>\n", |
|
|
220 |
" <td>-156.0</td>\n", |
|
|
221 |
" <td>45.572849</td>\n", |
|
|
222 |
" </tr>\n", |
|
|
223 |
" <tr>\n", |
|
|
224 |
" <th>2</th>\n", |
|
|
225 |
" <td>52c9913b1</td>\n", |
|
|
226 |
" <td>0</td>\n", |
|
|
227 |
" <td>0</td>\n", |
|
|
228 |
" <td>0</td>\n", |
|
|
229 |
" <td>0</td>\n", |
|
|
230 |
" <td>0</td>\n", |
|
|
231 |
" <td>0</td>\n", |
|
|
232 |
" <td>9c2b4bd7</td>\n", |
|
|
233 |
" <td>3e3634f8cf</td>\n", |
|
|
234 |
" <td>973274ffc9</td>\n", |
|
|
235 |
" <td>40</td>\n", |
|
|
236 |
" <td>150</td>\n", |
|
|
237 |
" <td>4.455000</td>\n", |
|
|
238 |
" <td>-125.0</td>\n", |
|
|
239 |
" <td>-115.063000</td>\n", |
|
|
240 |
" </tr>\n", |
|
|
241 |
" <tr>\n", |
|
|
242 |
" <th>3</th>\n", |
|
|
243 |
" <td>4e6ff6126</td>\n", |
|
|
244 |
" <td>0</td>\n", |
|
|
245 |
" <td>0</td>\n", |
|
|
246 |
" <td>0</td>\n", |
|
|
247 |
" <td>0</td>\n", |
|
|
248 |
" <td>0</td>\n", |
|
|
249 |
" <td>0</td>\n", |
|
|
250 |
" <td>3ae81c2d</td>\n", |
|
|
251 |
" <td>a1390c15c2</td>\n", |
|
|
252 |
" <td>e5ccad8244</td>\n", |
|
|
253 |
" <td>['00036', '00036']</td>\n", |
|
|
254 |
" <td>['00080', '00080']</td>\n", |
|
|
255 |
" <td>100.000000</td>\n", |
|
|
256 |
" <td>-99.5</td>\n", |
|
|
257 |
" <td>28.500000</td>\n", |
|
|
258 |
" </tr>\n", |
|
|
259 |
" <tr>\n", |
|
|
260 |
" <th>4</th>\n", |
|
|
261 |
" <td>7858edd88</td>\n", |
|
|
262 |
" <td>0</td>\n", |
|
|
263 |
" <td>0</td>\n", |
|
|
264 |
" <td>0</td>\n", |
|
|
265 |
" <td>0</td>\n", |
|
|
266 |
" <td>0</td>\n", |
|
|
267 |
" <td>0</td>\n", |
|
|
268 |
" <td>c1867feb</td>\n", |
|
|
269 |
" <td>c73e81ed3a</td>\n", |
|
|
270 |
" <td>28e0531b3a</td>\n", |
|
|
271 |
" <td>40</td>\n", |
|
|
272 |
" <td>100</td>\n", |
|
|
273 |
" <td>145.793000</td>\n", |
|
|
274 |
" <td>-125.0</td>\n", |
|
|
275 |
" <td>-132.190000</td>\n", |
|
|
276 |
" </tr>\n", |
|
|
277 |
" </tbody>\n", |
|
|
278 |
"</table>\n", |
|
|
279 |
"</div>" |
|
|
280 |
], |
|
|
281 |
"text/plain": [ |
|
|
282 |
" PatientID epidural intraparenchymal intraventricular subarachnoid \\\n", |
|
|
283 |
"0 63eb1e259 0 0 0 0 \n", |
|
|
284 |
"1 2669954a7 0 0 0 0 \n", |
|
|
285 |
"2 52c9913b1 0 0 0 0 \n", |
|
|
286 |
"3 4e6ff6126 0 0 0 0 \n", |
|
|
287 |
"4 7858edd88 0 0 0 0 \n", |
|
|
288 |
"\n", |
|
|
289 |
" subdural any PID StudyI SeriesI WindowCenter \\\n", |
|
|
290 |
"0 0 0 a449357f 62d125e5b2 0be5c0d1b3 ['00036', '00036'] \n", |
|
|
291 |
"1 0 0 363d5865 a20b80c7bf 3564d584db ['00047', '00047'] \n", |
|
|
292 |
"2 0 0 9c2b4bd7 3e3634f8cf 973274ffc9 40 \n", |
|
|
293 |
"3 0 0 3ae81c2d a1390c15c2 e5ccad8244 ['00036', '00036'] \n", |
|
|
294 |
"4 0 0 c1867feb c73e81ed3a 28e0531b3a 40 \n", |
|
|
295 |
"\n", |
|
|
296 |
" WindowWidth ImagePositionZ ImagePositionX ImagePositionY \n", |
|
|
297 |
"0 ['00080', '00080'] 180.199951 -125.0 -8.000000 \n", |
|
|
298 |
"1 ['00080', '00080'] 922.530821 -156.0 45.572849 \n", |
|
|
299 |
"2 150 4.455000 -125.0 -115.063000 \n", |
|
|
300 |
"3 ['00080', '00080'] 100.000000 -99.5 28.500000 \n", |
|
|
301 |
"4 100 145.793000 -125.0 -132.190000 " |
|
|
302 |
] |
|
|
303 |
}, |
|
|
304 |
"execution_count": 14, |
|
|
305 |
"metadata": {}, |
|
|
306 |
"output_type": "execute_result" |
|
|
307 |
} |
|
|
308 |
], |
|
|
309 |
"source": [ |
|
|
310 |
"train_df = pd.read_csv(data_dir+'train.csv')\n", |
|
|
311 |
"train_df.shape\n", |
|
|
312 |
"train_df=train_df[~train_df.PatientID.isin(bad_images)].reset_index(drop=True)\n", |
|
|
313 |
"train_df=train_df.drop_duplicates().reset_index(drop=True)\n", |
|
|
314 |
"train_df.shape\n", |
|
|
315 |
"train_df.head()" |
|
|
316 |
] |
|
|
317 |
}, |
|
|
318 |
{ |
|
|
319 |
"cell_type": "code", |
|
|
320 |
"execution_count": 15, |
|
|
321 |
"metadata": {}, |
|
|
322 |
"outputs": [ |
|
|
323 |
{ |
|
|
324 |
"data": { |
|
|
325 |
"text/html": [ |
|
|
326 |
"<div>\n", |
|
|
327 |
"<style scoped>\n", |
|
|
328 |
" .dataframe tbody tr th:only-of-type {\n", |
|
|
329 |
" vertical-align: middle;\n", |
|
|
330 |
" }\n", |
|
|
331 |
"\n", |
|
|
332 |
" .dataframe tbody tr th {\n", |
|
|
333 |
" vertical-align: top;\n", |
|
|
334 |
" }\n", |
|
|
335 |
"\n", |
|
|
336 |
" .dataframe thead th {\n", |
|
|
337 |
" text-align: right;\n", |
|
|
338 |
" }\n", |
|
|
339 |
"</style>\n", |
|
|
340 |
"<table border=\"1\" class=\"dataframe\">\n", |
|
|
341 |
" <thead>\n", |
|
|
342 |
" <tr style=\"text-align: right;\">\n", |
|
|
343 |
" <th></th>\n", |
|
|
344 |
" <th>PatientID</th>\n", |
|
|
345 |
" <th>epidural</th>\n", |
|
|
346 |
" <th>intraparenchymal</th>\n", |
|
|
347 |
" <th>intraventricular</th>\n", |
|
|
348 |
" <th>subarachnoid</th>\n", |
|
|
349 |
" <th>subdural</th>\n", |
|
|
350 |
" <th>any</th>\n", |
|
|
351 |
" <th>SeriesI</th>\n", |
|
|
352 |
" <th>PID</th>\n", |
|
|
353 |
" <th>StudyI</th>\n", |
|
|
354 |
" <th>WindowCenter</th>\n", |
|
|
355 |
" <th>WindowWidth</th>\n", |
|
|
356 |
" <th>ImagePositionZ</th>\n", |
|
|
357 |
" <th>ImagePositionX</th>\n", |
|
|
358 |
" <th>ImagePositionY</th>\n", |
|
|
359 |
" </tr>\n", |
|
|
360 |
" </thead>\n", |
|
|
361 |
" <tbody>\n", |
|
|
362 |
" <tr>\n", |
|
|
363 |
" <th>0</th>\n", |
|
|
364 |
" <td>28fbab7eb</td>\n", |
|
|
365 |
" <td>0.5</td>\n", |
|
|
366 |
" <td>0.5</td>\n", |
|
|
367 |
" <td>0.5</td>\n", |
|
|
368 |
" <td>0.5</td>\n", |
|
|
369 |
" <td>0.5</td>\n", |
|
|
370 |
" <td>0.5</td>\n", |
|
|
371 |
" <td>ebfd7e4506</td>\n", |
|
|
372 |
" <td>cf1b6b11</td>\n", |
|
|
373 |
" <td>93407cadbb</td>\n", |
|
|
374 |
" <td>30</td>\n", |
|
|
375 |
" <td>80</td>\n", |
|
|
376 |
" <td>158.458000</td>\n", |
|
|
377 |
" <td>-125.0</td>\n", |
|
|
378 |
" <td>-135.598000</td>\n", |
|
|
379 |
" </tr>\n", |
|
|
380 |
" <tr>\n", |
|
|
381 |
" <th>1</th>\n", |
|
|
382 |
" <td>877923b8b</td>\n", |
|
|
383 |
" <td>0.5</td>\n", |
|
|
384 |
" <td>0.5</td>\n", |
|
|
385 |
" <td>0.5</td>\n", |
|
|
386 |
" <td>0.5</td>\n", |
|
|
387 |
" <td>0.5</td>\n", |
|
|
388 |
" <td>0.5</td>\n", |
|
|
389 |
" <td>6d95084e15</td>\n", |
|
|
390 |
" <td>ad8ea58f</td>\n", |
|
|
391 |
" <td>a337baa067</td>\n", |
|
|
392 |
" <td>30</td>\n", |
|
|
393 |
" <td>80</td>\n", |
|
|
394 |
" <td>138.729050</td>\n", |
|
|
395 |
" <td>-125.0</td>\n", |
|
|
396 |
" <td>-101.797981</td>\n", |
|
|
397 |
" </tr>\n", |
|
|
398 |
" <tr>\n", |
|
|
399 |
" <th>2</th>\n", |
|
|
400 |
" <td>a591477cb</td>\n", |
|
|
401 |
" <td>0.5</td>\n", |
|
|
402 |
" <td>0.5</td>\n", |
|
|
403 |
" <td>0.5</td>\n", |
|
|
404 |
" <td>0.5</td>\n", |
|
|
405 |
" <td>0.5</td>\n", |
|
|
406 |
" <td>0.5</td>\n", |
|
|
407 |
" <td>8e06b2c9e0</td>\n", |
|
|
408 |
" <td>ecfb278b</td>\n", |
|
|
409 |
" <td>0cfe838d54</td>\n", |
|
|
410 |
" <td>30</td>\n", |
|
|
411 |
" <td>80</td>\n", |
|
|
412 |
" <td>60.830002</td>\n", |
|
|
413 |
" <td>-125.0</td>\n", |
|
|
414 |
" <td>-133.300003</td>\n", |
|
|
415 |
" </tr>\n", |
|
|
416 |
" <tr>\n", |
|
|
417 |
" <th>3</th>\n", |
|
|
418 |
" <td>42217c898</td>\n", |
|
|
419 |
" <td>0.5</td>\n", |
|
|
420 |
" <td>0.5</td>\n", |
|
|
421 |
" <td>0.5</td>\n", |
|
|
422 |
" <td>0.5</td>\n", |
|
|
423 |
" <td>0.5</td>\n", |
|
|
424 |
" <td>0.5</td>\n", |
|
|
425 |
" <td>e800f419cf</td>\n", |
|
|
426 |
" <td>e96e31f4</td>\n", |
|
|
427 |
" <td>c497ac5bad</td>\n", |
|
|
428 |
" <td>30</td>\n", |
|
|
429 |
" <td>80</td>\n", |
|
|
430 |
" <td>55.388000</td>\n", |
|
|
431 |
" <td>-125.0</td>\n", |
|
|
432 |
" <td>-146.081000</td>\n", |
|
|
433 |
" </tr>\n", |
|
|
434 |
" <tr>\n", |
|
|
435 |
" <th>4</th>\n", |
|
|
436 |
" <td>a130c4d2f</td>\n", |
|
|
437 |
" <td>0.5</td>\n", |
|
|
438 |
" <td>0.5</td>\n", |
|
|
439 |
" <td>0.5</td>\n", |
|
|
440 |
" <td>0.5</td>\n", |
|
|
441 |
" <td>0.5</td>\n", |
|
|
442 |
" <td>0.5</td>\n", |
|
|
443 |
" <td>faeb7454f3</td>\n", |
|
|
444 |
" <td>69affa42</td>\n", |
|
|
445 |
" <td>854e4fbc01</td>\n", |
|
|
446 |
" <td>30</td>\n", |
|
|
447 |
" <td>80</td>\n", |
|
|
448 |
" <td>33.516888</td>\n", |
|
|
449 |
" <td>-125.0</td>\n", |
|
|
450 |
" <td>-118.689819</td>\n", |
|
|
451 |
" </tr>\n", |
|
|
452 |
" </tbody>\n", |
|
|
453 |
"</table>\n", |
|
|
454 |
"</div>" |
|
|
455 |
], |
|
|
456 |
"text/plain": [ |
|
|
457 |
" PatientID epidural intraparenchymal intraventricular subarachnoid \\\n", |
|
|
458 |
"0 28fbab7eb 0.5 0.5 0.5 0.5 \n", |
|
|
459 |
"1 877923b8b 0.5 0.5 0.5 0.5 \n", |
|
|
460 |
"2 a591477cb 0.5 0.5 0.5 0.5 \n", |
|
|
461 |
"3 42217c898 0.5 0.5 0.5 0.5 \n", |
|
|
462 |
"4 a130c4d2f 0.5 0.5 0.5 0.5 \n", |
|
|
463 |
"\n", |
|
|
464 |
" subdural any SeriesI PID StudyI WindowCenter WindowWidth \\\n", |
|
|
465 |
"0 0.5 0.5 ebfd7e4506 cf1b6b11 93407cadbb 30 80 \n", |
|
|
466 |
"1 0.5 0.5 6d95084e15 ad8ea58f a337baa067 30 80 \n", |
|
|
467 |
"2 0.5 0.5 8e06b2c9e0 ecfb278b 0cfe838d54 30 80 \n", |
|
|
468 |
"3 0.5 0.5 e800f419cf e96e31f4 c497ac5bad 30 80 \n", |
|
|
469 |
"4 0.5 0.5 faeb7454f3 69affa42 854e4fbc01 30 80 \n", |
|
|
470 |
"\n", |
|
|
471 |
" ImagePositionZ ImagePositionX ImagePositionY \n", |
|
|
472 |
"0 158.458000 -125.0 -135.598000 \n", |
|
|
473 |
"1 138.729050 -125.0 -101.797981 \n", |
|
|
474 |
"2 60.830002 -125.0 -133.300003 \n", |
|
|
475 |
"3 55.388000 -125.0 -146.081000 \n", |
|
|
476 |
"4 33.516888 -125.0 -118.689819 " |
|
|
477 |
] |
|
|
478 |
}, |
|
|
479 |
"execution_count": 15, |
|
|
480 |
"metadata": {}, |
|
|
481 |
"output_type": "execute_result" |
|
|
482 |
} |
|
|
483 |
], |
|
|
484 |
"source": [ |
|
|
485 |
"test_df = pd.read_csv(data_dir+'test.csv')\n", |
|
|
486 |
"test_df.head()" |
|
|
487 |
] |
|
|
488 |
}, |
|
|
489 |
{ |
|
|
490 |
"cell_type": "code", |
|
|
491 |
"execution_count": 16, |
|
|
492 |
"metadata": {}, |
|
|
493 |
"outputs": [], |
|
|
494 |
"source": [ |
|
|
495 |
"split_sid = train_df.PID.unique()\n", |
|
|
496 |
"splits=list(KFold(n_splits=n_splits,shuffle=True, random_state=SEED).split(split_sid))\n" |
|
|
497 |
] |
|
|
498 |
}, |
|
|
499 |
{ |
|
|
500 |
"cell_type": "code", |
|
|
501 |
"execution_count": 17, |
|
|
502 |
"metadata": {}, |
|
|
503 |
"outputs": [], |
|
|
504 |
"source": [ |
|
|
505 |
"pickle_file=open(outputs_dir+\"PID_splits_{}.pkl\".format(n_splits),'wb')\n", |
|
|
506 |
"pickle.dump((split_sid,splits),pickle_file,protocol=4)\n", |
|
|
507 |
"pickle_file.close()\n" |
|
|
508 |
] |
|
|
509 |
}, |
|
|
510 |
{ |
|
|
511 |
"cell_type": "code", |
|
|
512 |
"execution_count": 10, |
|
|
513 |
"metadata": {}, |
|
|
514 |
"outputs": [], |
|
|
515 |
"source": [ |
|
|
516 |
"def my_loss(y_pred,y_true,weights):\n", |
|
|
517 |
" if len(y_pred.shape)==len(y_true.shape): \n", |
|
|
518 |
" # Normal loss\n", |
|
|
519 |
" loss = F.binary_cross_entropy_with_logits(y_pred,y_true,weights.expand_as(y_pred))\n", |
|
|
520 |
" else:\n", |
|
|
521 |
" # Mixup loss (not used here)\n", |
|
|
522 |
" loss0 = F.binary_cross_entropy_with_logits(y_pred,y_true[...,0],weights.repeat(y_pred.shape[0],1),reduction='none')\n", |
|
|
523 |
" loss1 = F.binary_cross_entropy_with_logits(y_pred,y_true[...,1],weights.repeat(y_pred.shape[0],1),reduction='none')\n", |
|
|
524 |
" loss = (y_true[...,2]*loss0+(1.0-y_true[...,2])*loss1).mean() \n", |
|
|
525 |
" return loss" |
|
|
526 |
] |
|
|
527 |
}, |
|
|
528 |
{ |
|
|
529 |
"cell_type": "code", |
|
|
530 |
"execution_count": 11, |
|
|
531 |
"metadata": {}, |
|
|
532 |
"outputs": [], |
|
|
533 |
"source": [ |
|
|
534 |
"class FocalLoss(nn.Module):\n", |
|
|
535 |
" def __init__(self, alpha=1, gamma=2, logits=True, reduce=True):\n", |
|
|
536 |
" super(FocalLoss, self).__init__()\n", |
|
|
537 |
" self.alpha = alpha\n", |
|
|
538 |
" self.gamma = gamma\n", |
|
|
539 |
" self.logits = logits\n", |
|
|
540 |
" self.reduce = reduce\n", |
|
|
541 |
"\n", |
|
|
542 |
" def forward(self, y_pred,y_true,weights):\n", |
|
|
543 |
" if self.logits:\n", |
|
|
544 |
" BCE_loss = F.binary_cross_entropy_with_logits(y_pred,y_true,weights.expand_as(y_pred), reduction='none')\n", |
|
|
545 |
" else:\n", |
|
|
546 |
" BCE_loss = F.binary_cross_entropy(y_pred,y_true,weights.expand_as(y_pred), reduction='none')\n", |
|
|
547 |
" pt = torch.exp(-BCE_loss)\n", |
|
|
548 |
" F_loss = self.alpha * (1-pt)**self.gamma * BCE_loss\n", |
|
|
549 |
"\n", |
|
|
550 |
" if self.reduce:\n", |
|
|
551 |
" return torch.mean(F_loss)\n", |
|
|
552 |
" else:\n", |
|
|
553 |
" return F_loss" |
|
|
554 |
] |
|
|
555 |
}, |
|
|
556 |
{ |
|
|
557 |
"cell_type": "code", |
|
|
558 |
"execution_count": 12, |
|
|
559 |
"metadata": {}, |
|
|
560 |
"outputs": [], |
|
|
561 |
"source": [ |
|
|
562 |
"class parameter_scheduler():\n", |
|
|
563 |
" def __init__(self,model,do_first=['classifier'],num_epoch=1):\n", |
|
|
564 |
" self.model=model\n", |
|
|
565 |
" self.do_first = do_first\n", |
|
|
566 |
" self.num_epoch=num_epoch\n", |
|
|
567 |
" def __call__(self,epoch):\n", |
|
|
568 |
" if epoch>=self.num_epoch:\n", |
|
|
569 |
" for n,p in self.model.named_parameters():\n", |
|
|
570 |
" p.requires_grad=True\n", |
|
|
571 |
" else:\n", |
|
|
572 |
" for n,p in self.model.named_parameters():\n", |
|
|
573 |
" p.requires_grad= any(nd in n for nd in self.do_first)\n" |
|
|
574 |
] |
|
|
575 |
}, |
|
|
576 |
{ |
|
|
577 |
"cell_type": "code", |
|
|
578 |
"execution_count": 13, |
|
|
579 |
"metadata": {}, |
|
|
580 |
"outputs": [], |
|
|
581 |
"source": [ |
|
|
582 |
"def get_model(model_name):\n", |
|
|
583 |
" if params['model_name'].startswith('se'):\n", |
|
|
584 |
" return MySENet, pretrainedmodels.__dict__[params['model_name']](num_classes=1000, pretrained='imagenet')\n", |
|
|
585 |
" elif 'Densenet161' in params['model_name']:\n", |
|
|
586 |
" return partial(MyDenseNet, strategy='none'),models.densenet161(pretrained=True)\n", |
|
|
587 |
" elif 'Densenet169' in params['model_name']:\n", |
|
|
588 |
" return partial(MyDenseNet, strategy='none'),models.densenet169(pretrained=True)\n", |
|
|
589 |
" else:\n", |
|
|
590 |
" raise" |
|
|
591 |
] |
|
|
592 |
}, |
|
|
593 |
{ |
|
|
594 |
"cell_type": "code", |
|
|
595 |
"execution_count": null, |
|
|
596 |
"metadata": { |
|
|
597 |
"scrolled": false |
|
|
598 |
}, |
|
|
599 |
"outputs": [], |
|
|
600 |
"source": [ |
|
|
601 |
"%matplotlib nbagg\n", |
|
|
602 |
"for num_split in range(params['n_splits']):\n", |
|
|
603 |
" np.random.seed(SEED+num_split)\n", |
|
|
604 |
" torch.manual_seed(SEED+num_split)\n", |
|
|
605 |
" torch.cuda.manual_seed(SEED+num_split)\n", |
|
|
606 |
" #torch.backends.cudnn.deterministic = True\n", |
|
|
607 |
" idx_train = train_df[train_df.PID.isin(set(split_sid[splits[num_split][0]]))].index.values\n", |
|
|
608 |
" idx_validate = train_df[train_df.PID.isin(set(split_sid[splits[num_split][1]]))].index.values\n", |
|
|
609 |
" idx_train.shape\n", |
|
|
610 |
" idx_validate.shape\n", |
|
|
611 |
"\n", |
|
|
612 |
" klr=1\n", |
|
|
613 |
" batch_size=32\n", |
|
|
614 |
" num_workers=12\n", |
|
|
615 |
" num_epochs=params['num_epochs']\n", |
|
|
616 |
" model_name,version = params['model_name'] , params['version']\n", |
|
|
617 |
" new_model,base_model=get_model(params['model_name'])\n", |
|
|
618 |
" model = new_model(base_model,\n", |
|
|
619 |
" len(hemorrhage_types),\n", |
|
|
620 |
" num_channels=3,\n", |
|
|
621 |
" dropout=0.2,\n", |
|
|
622 |
" wso=((40,80),(80,200),(40,400)),\n", |
|
|
623 |
" dont_do_grad=[],\n", |
|
|
624 |
" extra_pool=params['num_pool'],\n", |
|
|
625 |
" )\n", |
|
|
626 |
" if params['Pre_version'] is not None:\n", |
|
|
627 |
" model.load_state_dict(torch.load(models_dir+models_format.format(model_name,params['Pre_version'],\n", |
|
|
628 |
" num_split),map_location=torch.device(device)))\n", |
|
|
629 |
"\n", |
|
|
630 |
" _=model.to(device)\n", |
|
|
631 |
" weights = torch.tensor([1.,1.,1.,1.,1.,2.],device=device)\n", |
|
|
632 |
" loss_func=my_loss if not params['focal'] else FocalLoss()\n", |
|
|
633 |
" targets_dataset=D.TensorDataset(torch.tensor(train_df[hemorrhage_types].values,dtype=torch.float))\n", |
|
|
634 |
" transform=MyTransform(mean_change=15,\n", |
|
|
635 |
" std_change=0,\n", |
|
|
636 |
" flip=True,\n", |
|
|
637 |
" zoom=(0.2,0.2),\n", |
|
|
638 |
" rotate=30,\n", |
|
|
639 |
" out_size=512,\n", |
|
|
640 |
" shift=10,\n", |
|
|
641 |
" normal=False)\n", |
|
|
642 |
" imagedataset = ImageDataset(train_df,transform=transform.random,base_path=train_images_dir,\n", |
|
|
643 |
" window_eq=False,equalize=False,rescale=True)\n", |
|
|
644 |
" transform_val=MyTransform(out_size=512)\n", |
|
|
645 |
" imagedataset_val = ImageDataset(train_df,transform=transform_val.random,base_path=train_images_dir,\n", |
|
|
646 |
" window_eq=False,equalize=False,rescale=True)\n", |
|
|
647 |
" combined_dataset=DatasetCat([imagedataset,targets_dataset])\n", |
|
|
648 |
" combined_dataset_val=DatasetCat([imagedataset_val,targets_dataset])\n", |
|
|
649 |
" optimizer_grouped_parameters=model.get_optimizer_parameters(klr)\n", |
|
|
650 |
" sampling=sampler(train_df[hemorrhage_types].values[idx_train],0.5,[0,0,0,0,0,1])\n", |
|
|
651 |
" sample_ratio=1.02*float(sampling().shape[0])/idx_train.shape[0]\n", |
|
|
652 |
" train_dataset=D.Subset(combined_dataset,idx_train)\n", |
|
|
653 |
" validate_dataset=D.Subset(combined_dataset_val,idx_validate)\n", |
|
|
654 |
" num_train_optimization_steps = num_epochs*(sample_ratio*len(train_dataset)//batch_size+int(len(train_dataset)%batch_size>0))\n", |
|
|
655 |
" fig,ax = plt.subplots(figsize=(10,7))\n", |
|
|
656 |
" gr=loss_graph(fig,ax,num_epochs,int(num_train_optimization_steps/num_epochs)+1,limits=(0.05,0.2))\n", |
|
|
657 |
" sched=WarmupExpCosineWithWarmupRestartsSchedule( t_total=num_train_optimization_steps, cycles=num_epochs,tau=1)\n", |
|
|
658 |
" optimizer = BertAdam(optimizer_grouped_parameters,lr=klr*1e-3,schedule=sched)\n", |
|
|
659 |
" model, optimizer = amp.initialize(model, optimizer, opt_level=\"O1\",verbosity=0)\n", |
|
|
660 |
" history,best_model= model_train(model,\n", |
|
|
661 |
" optimizer,\n", |
|
|
662 |
" train_dataset,\n", |
|
|
663 |
" batch_size,\n", |
|
|
664 |
" num_epochs,\n", |
|
|
665 |
" loss_func,\n", |
|
|
666 |
" weights=weights,\n", |
|
|
667 |
" do_apex=False,\n", |
|
|
668 |
" model_apexed=True,\n", |
|
|
669 |
" validate_dataset=validate_dataset,\n", |
|
|
670 |
" param_schedualer=None,\n", |
|
|
671 |
" weights_data=None,\n", |
|
|
672 |
" metric=None,\n", |
|
|
673 |
" return_model=True,\n", |
|
|
674 |
" num_workers=num_workers,\n", |
|
|
675 |
" sampler=None,\n", |
|
|
676 |
" pre_process = None,\n", |
|
|
677 |
" graph=gr,\n", |
|
|
678 |
" call_progress=sendmeemail)\n", |
|
|
679 |
"\n", |
|
|
680 |
" torch.save(best_model.state_dict(), models_dir+models_format.format(model_name,version,num_split))" |
|
|
681 |
] |
|
|
682 |
}, |
|
|
683 |
{ |
|
|
684 |
"cell_type": "code", |
|
|
685 |
"execution_count": null, |
|
|
686 |
"metadata": {}, |
|
|
687 |
"outputs": [], |
|
|
688 |
"source": [ |
|
|
689 |
"for num_split in range(params['n_splits']):\n", |
|
|
690 |
" idx_validate = train_df[train_df.PID.isin(set(split_sid[splits[num_split][1]]))].index.values\n", |
|
|
691 |
" model_name,version =params['model_name'] , params['version']\n", |
|
|
692 |
" new_model,base_model=get_model(params['model_name'])\n", |
|
|
693 |
" model = new_model(base_model,\n", |
|
|
694 |
" len(hemorrhage_types),\n", |
|
|
695 |
" num_channels=3,\n", |
|
|
696 |
" dropout=0.2,\n", |
|
|
697 |
" wso=((40,80),(80,200),(40,400)),\n", |
|
|
698 |
" dont_do_grad=[],\n", |
|
|
699 |
" extra_pool=params['num_pool'],\n", |
|
|
700 |
" )\n", |
|
|
701 |
" model.load_state_dict(torch.load(models_dir+models_format.format(model_name,version,num_split),map_location=torch.device(device)))\n", |
|
|
702 |
" _=model.to(device)\n", |
|
|
703 |
" transform=MyTransform(mean_change=15,\n", |
|
|
704 |
" std_change=0,\n", |
|
|
705 |
" flip=True,\n", |
|
|
706 |
" zoom=(0.2,0.2),\n", |
|
|
707 |
" rotate=30,\n", |
|
|
708 |
" out_size=512,\n", |
|
|
709 |
" shift=0,\n", |
|
|
710 |
" normal=False)\n", |
|
|
711 |
" indexes=np.arange(train_df.shape[0]).repeat(4)\n", |
|
|
712 |
" train_dataset=D.Subset(ImageDataset(train_df,transform=transform.random,base_path=train_images_dir,\n", |
|
|
713 |
" window_eq=False,equalize=False,rescale=True),indexes)\n", |
|
|
714 |
" pred,features = model_run(model,train_dataset,do_apex=True,batch_size=96,num_workers=14)\n", |
|
|
715 |
"\n", |
|
|
716 |
" pickle_file=open(outputs_dir+outputs_format.format(model_name,version,params['train_features'],num_split),'wb')\n", |
|
|
717 |
" pickle.dump(features,pickle_file,protocol=4)\n", |
|
|
718 |
" pickle_file.close()\n", |
|
|
719 |
"\n", |
|
|
720 |
" pickle_file=open(outputs_dir+outputs_format.format(model_name,version,params['train_prediction'],num_split),'wb')\n", |
|
|
721 |
" pickle.dump(pred,pickle_file,protocol=4)\n", |
|
|
722 |
" pickle_file.close()\n", |
|
|
723 |
"\n", |
|
|
724 |
"\n", |
|
|
725 |
" my_loss(pred[(idx_validate*4+np.arange(4)[:,None]).transpose(1,0)].mean(1),\n", |
|
|
726 |
" torch.tensor(train_df[hemorrhage_types].values[idx_validate],dtype=torch.float),\n", |
|
|
727 |
" torch.tensor([1.,1.,1.,1.,1.,2.]))" |
|
|
728 |
] |
|
|
729 |
}, |
|
|
730 |
{ |
|
|
731 |
"cell_type": "code", |
|
|
732 |
"execution_count": null, |
|
|
733 |
"metadata": {}, |
|
|
734 |
"outputs": [], |
|
|
735 |
"source": [ |
|
|
736 |
"for num_split in range(params['n_splits']):\n", |
|
|
737 |
" idx_validate = train_df[train_df.PID.isin(set(split_sid[splits[num_split][1]]))].index.values\n", |
|
|
738 |
" model_name,version =params['model_name'] , params['version']\n", |
|
|
739 |
" new_model,base_model=get_model(params['model_name'])\n", |
|
|
740 |
" model = new_model(base_model,\n", |
|
|
741 |
" len(hemorrhage_types),\n", |
|
|
742 |
" num_channels=3,\n", |
|
|
743 |
" dropout=0.2,\n", |
|
|
744 |
" wso=((40,80),(80,200),(40,400)),\n", |
|
|
745 |
" dont_do_grad=[],\n", |
|
|
746 |
" extra_pool=params['num_pool'],\n", |
|
|
747 |
" )\n", |
|
|
748 |
" model.load_state_dict(torch.load(models_dir+models_format.format(model_name,version,num_split),map_location=torch.device(device)))\n", |
|
|
749 |
" _=model.to(device)\n", |
|
|
750 |
" transform=MyTransform(mean_change=15,\n", |
|
|
751 |
" std_change=0,\n", |
|
|
752 |
" flip=True,\n", |
|
|
753 |
" zoom=(0.2,0.2),\n", |
|
|
754 |
" rotate=30,\n", |
|
|
755 |
" out_size=512,\n", |
|
|
756 |
" shift=0,\n", |
|
|
757 |
" normal=False)\n", |
|
|
758 |
" indexes=np.arange(test_df.shape[0]).repeat(8)\n", |
|
|
759 |
" imagedataset_test=D.Subset(ImageDataset(test_df,transform=transform.random,base_path=test_images_dir,\n", |
|
|
760 |
" window_eq=False,equalize=False,rescale=True),indexes)\n", |
|
|
761 |
" pred,features = model_run(model,imagedataset_test,do_apex=True,batch_size=96,num_workers=18)\n", |
|
|
762 |
" pickle_file=open(outputs_dir+outputs_format.format(model_name,version,params['test_features'],num_split),'wb')\n", |
|
|
763 |
" pickle.dump(features,pickle_file,protocol=4)\n", |
|
|
764 |
" pickle_file.close()\n", |
|
|
765 |
"\n", |
|
|
766 |
" pickle_file=open(outputs_dir+outputs_format.format(model_name,version,params['test_prediction'],num_split),'wb')\n", |
|
|
767 |
" pickle.dump(pred,pickle_file,protocol=4)\n", |
|
|
768 |
" pickle_file.close()\n" |
|
|
769 |
] |
|
|
770 |
}, |
|
|
771 |
{ |
|
|
772 |
"cell_type": "markdown", |
|
|
773 |
"metadata": {}, |
|
|
774 |
"source": [ |
|
|
775 |
"## create submission file - for reference" |
|
|
776 |
] |
|
|
777 |
}, |
|
|
778 |
{ |
|
|
779 |
"cell_type": "code", |
|
|
780 |
"execution_count": 33, |
|
|
781 |
"metadata": {}, |
|
|
782 |
"outputs": [ |
|
|
783 |
{ |
|
|
784 |
"data": { |
|
|
785 |
"application/vnd.jupyter.widget-view+json": { |
|
|
786 |
"model_id": "3ae2e5d2ec4d4919b47ab1bf3482807c", |
|
|
787 |
"version_major": 2, |
|
|
788 |
"version_minor": 0 |
|
|
789 |
}, |
|
|
790 |
"text/plain": [ |
|
|
791 |
"HBox(children=(IntProgress(value=0, max=3), HTML(value='')))" |
|
|
792 |
] |
|
|
793 |
}, |
|
|
794 |
"metadata": {}, |
|
|
795 |
"output_type": "display_data" |
|
|
796 |
}, |
|
|
797 |
{ |
|
|
798 |
"data": { |
|
|
799 |
"text/plain": [ |
|
|
800 |
"torch.Size([78545, 24, 6])" |
|
|
801 |
] |
|
|
802 |
}, |
|
|
803 |
"execution_count": 33, |
|
|
804 |
"metadata": {}, |
|
|
805 |
"output_type": "execute_result" |
|
|
806 |
} |
|
|
807 |
], |
|
|
808 |
"source": [ |
|
|
809 |
"preds=[]\n", |
|
|
810 |
"for i in tqdm_notebook(range(params['n_splits'])):\n", |
|
|
811 |
" model_name,version, num_split = params['model_name'] , params['version'],i\n", |
|
|
812 |
" pickle_file=open(outputs_dir+outputs_format.format(model_name,version,params['test_prediction'],num_split),'rb')\n", |
|
|
813 |
" pred=pickle.load(pickle_file)\n", |
|
|
814 |
" pickle_file.close()\n", |
|
|
815 |
" preds.append(pred[(np.arange(pred.shape[0]).reshape(pred.shape[0]//8,8))])\n", |
|
|
816 |
"predss = torch.cat(preds,1)\n", |
|
|
817 |
"predss.shape" |
|
|
818 |
] |
|
|
819 |
}, |
|
|
820 |
{ |
|
|
821 |
"cell_type": "code", |
|
|
822 |
"execution_count": 36, |
|
|
823 |
"metadata": {}, |
|
|
824 |
"outputs": [ |
|
|
825 |
{ |
|
|
826 |
"data": { |
|
|
827 |
"text/html": [ |
|
|
828 |
"<div>\n", |
|
|
829 |
"<style scoped>\n", |
|
|
830 |
" .dataframe tbody tr th:only-of-type {\n", |
|
|
831 |
" vertical-align: middle;\n", |
|
|
832 |
" }\n", |
|
|
833 |
"\n", |
|
|
834 |
" .dataframe tbody tr th {\n", |
|
|
835 |
" vertical-align: top;\n", |
|
|
836 |
" }\n", |
|
|
837 |
"\n", |
|
|
838 |
" .dataframe thead th {\n", |
|
|
839 |
" text-align: right;\n", |
|
|
840 |
" }\n", |
|
|
841 |
"</style>\n", |
|
|
842 |
"<table border=\"1\" class=\"dataframe\">\n", |
|
|
843 |
" <thead>\n", |
|
|
844 |
" <tr style=\"text-align: right;\">\n", |
|
|
845 |
" <th></th>\n", |
|
|
846 |
" <th>ID</th>\n", |
|
|
847 |
" <th>Label</th>\n", |
|
|
848 |
" </tr>\n", |
|
|
849 |
" </thead>\n", |
|
|
850 |
" <tbody>\n", |
|
|
851 |
" <tr>\n", |
|
|
852 |
" <th>0</th>\n", |
|
|
853 |
" <td>ID_000012eaf_any</td>\n", |
|
|
854 |
" <td>0.012404</td>\n", |
|
|
855 |
" </tr>\n", |
|
|
856 |
" <tr>\n", |
|
|
857 |
" <th>1</th>\n", |
|
|
858 |
" <td>ID_000012eaf_epidural</td>\n", |
|
|
859 |
" <td>0.000464</td>\n", |
|
|
860 |
" </tr>\n", |
|
|
861 |
" <tr>\n", |
|
|
862 |
" <th>2</th>\n", |
|
|
863 |
" <td>ID_000012eaf_intraparenchymal</td>\n", |
|
|
864 |
" <td>0.001818</td>\n", |
|
|
865 |
" </tr>\n", |
|
|
866 |
" <tr>\n", |
|
|
867 |
" <th>3</th>\n", |
|
|
868 |
" <td>ID_000012eaf_intraventricular</td>\n", |
|
|
869 |
" <td>0.000576</td>\n", |
|
|
870 |
" </tr>\n", |
|
|
871 |
" <tr>\n", |
|
|
872 |
" <th>4</th>\n", |
|
|
873 |
" <td>ID_000012eaf_subarachnoid</td>\n", |
|
|
874 |
" <td>0.001655</td>\n", |
|
|
875 |
" </tr>\n", |
|
|
876 |
" <tr>\n", |
|
|
877 |
" <th>5</th>\n", |
|
|
878 |
" <td>ID_000012eaf_subdural</td>\n", |
|
|
879 |
" <td>0.010707</td>\n", |
|
|
880 |
" </tr>\n", |
|
|
881 |
" <tr>\n", |
|
|
882 |
" <th>6</th>\n", |
|
|
883 |
" <td>ID_0000ca2f6_any</td>\n", |
|
|
884 |
" <td>0.002507</td>\n", |
|
|
885 |
" </tr>\n", |
|
|
886 |
" <tr>\n", |
|
|
887 |
" <th>7</th>\n", |
|
|
888 |
" <td>ID_0000ca2f6_epidural</td>\n", |
|
|
889 |
" <td>0.000038</td>\n", |
|
|
890 |
" </tr>\n", |
|
|
891 |
" <tr>\n", |
|
|
892 |
" <th>8</th>\n", |
|
|
893 |
" <td>ID_0000ca2f6_intraparenchymal</td>\n", |
|
|
894 |
" <td>0.000540</td>\n", |
|
|
895 |
" </tr>\n", |
|
|
896 |
" <tr>\n", |
|
|
897 |
" <th>9</th>\n", |
|
|
898 |
" <td>ID_0000ca2f6_intraventricular</td>\n", |
|
|
899 |
" <td>0.000080</td>\n", |
|
|
900 |
" </tr>\n", |
|
|
901 |
" <tr>\n", |
|
|
902 |
" <th>10</th>\n", |
|
|
903 |
" <td>ID_0000ca2f6_subarachnoid</td>\n", |
|
|
904 |
" <td>0.000490</td>\n", |
|
|
905 |
" </tr>\n", |
|
|
906 |
" <tr>\n", |
|
|
907 |
" <th>11</th>\n", |
|
|
908 |
" <td>ID_0000ca2f6_subdural</td>\n", |
|
|
909 |
" <td>0.001157</td>\n", |
|
|
910 |
" </tr>\n", |
|
|
911 |
" </tbody>\n", |
|
|
912 |
"</table>\n", |
|
|
913 |
"</div>" |
|
|
914 |
], |
|
|
915 |
"text/plain": [ |
|
|
916 |
" ID Label\n", |
|
|
917 |
"0 ID_000012eaf_any 0.012404\n", |
|
|
918 |
"1 ID_000012eaf_epidural 0.000464\n", |
|
|
919 |
"2 ID_000012eaf_intraparenchymal 0.001818\n", |
|
|
920 |
"3 ID_000012eaf_intraventricular 0.000576\n", |
|
|
921 |
"4 ID_000012eaf_subarachnoid 0.001655\n", |
|
|
922 |
"5 ID_000012eaf_subdural 0.010707\n", |
|
|
923 |
"6 ID_0000ca2f6_any 0.002507\n", |
|
|
924 |
"7 ID_0000ca2f6_epidural 0.000038\n", |
|
|
925 |
"8 ID_0000ca2f6_intraparenchymal 0.000540\n", |
|
|
926 |
"9 ID_0000ca2f6_intraventricular 0.000080\n", |
|
|
927 |
"10 ID_0000ca2f6_subarachnoid 0.000490\n", |
|
|
928 |
"11 ID_0000ca2f6_subdural 0.001157" |
|
|
929 |
] |
|
|
930 |
}, |
|
|
931 |
"execution_count": 36, |
|
|
932 |
"metadata": {}, |
|
|
933 |
"output_type": "execute_result" |
|
|
934 |
}, |
|
|
935 |
{ |
|
|
936 |
"data": { |
|
|
937 |
"text/plain": [ |
|
|
938 |
"(471270, 2)" |
|
|
939 |
] |
|
|
940 |
}, |
|
|
941 |
"execution_count": 36, |
|
|
942 |
"metadata": {}, |
|
|
943 |
"output_type": "execute_result" |
|
|
944 |
} |
|
|
945 |
], |
|
|
946 |
"source": [ |
|
|
947 |
"submission_df=get_submission(test_df,torch.sigmoid(predss).mean(1),False)\n", |
|
|
948 |
"submission_df.head(12)\n", |
|
|
949 |
"submission_df.shape\n", |
|
|
950 |
"sub_num=999\n", |
|
|
951 |
"submission_df.to_csv('/media/hd/notebooks/data/RSNA/submissions/submission{}.csv'.format(sub_num),\n", |
|
|
952 |
" index=False, columns=['ID','Label'])\n" |
|
|
953 |
] |
|
|
954 |
}, |
|
|
955 |
{ |
|
|
956 |
"cell_type": "code", |
|
|
957 |
"execution_count": null, |
|
|
958 |
"metadata": {}, |
|
|
959 |
"outputs": [], |
|
|
960 |
"source": [] |
|
|
961 |
} |
|
|
962 |
], |
|
|
963 |
"metadata": { |
|
|
964 |
"kernelspec": { |
|
|
965 |
"display_name": "Python 3", |
|
|
966 |
"language": "python", |
|
|
967 |
"name": "python3" |
|
|
968 |
}, |
|
|
969 |
"language_info": { |
|
|
970 |
"codemirror_mode": { |
|
|
971 |
"name": "ipython", |
|
|
972 |
"version": 3 |
|
|
973 |
}, |
|
|
974 |
"file_extension": ".py", |
|
|
975 |
"mimetype": "text/x-python", |
|
|
976 |
"name": "python", |
|
|
977 |
"nbconvert_exporter": "python", |
|
|
978 |
"pygments_lexer": "ipython3", |
|
|
979 |
"version": "3.6.6" |
|
|
980 |
} |
|
|
981 |
}, |
|
|
982 |
"nbformat": 4, |
|
|
983 |
"nbformat_minor": 2 |
|
|
984 |
} |