-
Notifications
You must be signed in to change notification settings - Fork 1
/
seizure_detection.py
332 lines (300 loc) · 19.4 KB
/
seizure_detection.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import json
import os.path
import numpy as np
from common import time
from common.data import CachedDataLoader, makedirs
from common.pipeline import *
from seizure.transforms import *
from seizure.tasks import TaskCore, CrossValidationScoreTask, MakePredictionsTask, TrainClassifierTask, \
TrainingDataTask, LoadTestDataTask
import seizure.tasks
from seizure.scores import get_score_summary, print_results
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier, ExtraTreesClassifier, \
GradientBoostingClassifier
from sklearn.linear_model import LogisticRegression, SGDClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
def run_seizure_detection(build_target, targets=None):
"""
The main entry point for running seizure-detection cross-validation and predictions.
Directories from settings file are configured, classifiers are chosen, pipelines are
chosen, and the chosen build_target ('cv', 'predict', 'train_model') is run across
all combinations of (targets, pipelines, classifiers)
"""
with open('SETTINGS.json') as f:
settings = json.load(f)
data_dir = str(settings['competition-data-dir'])
cache_dir = str(settings['data-cache-dir'])
import seizure.transforms
seizure.transforms.cache_dir = cache_dir
submission_dir = str(settings['submission-dir'])
seizure.tasks.task_predict = str(settings.get('task')) == 'predict'
makedirs(submission_dir)
cached_data_loader = CachedDataLoader(cache_dir)
ts = time.get_millis()
if not targets:
if seizure.tasks.task_predict:
# add leader-board weight to each target. I am using the number of test example as the weight assuming
# all test examples are weighted equally on the leader-board
targets = [
('Dog_1',502),
('Dog_2',1000),
('Dog_3',907),
('Dog_4',990),
('Dog_5',191),
('Patient_1',195),
('Patient_2',150),
]
else:
targets = [
'Dog_1',
'Dog_2',
'Dog_3',
'Dog_4',
'Dog_5',
'Patient_1',
'Patient_2',
'Patient_3',
'Patient_4',
'Patient_5',
'Patient_6',
'Patient_7',
'Patient_8'
]
pipelines = [
# NOTE(mike): you can enable multiple pipelines to run them all and compare results
#Pipeline(gen_ictal=False, pipeline=[FFT(), Slice(1, 48), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[FFT(), Slice(1, 64), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[FFT(), Slice(1, 96), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[FFT(), Slice(1, 128), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[FFT(), Slice(1, 160), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[FFT(), Magnitude(), Log10()]),
# Pipeline(gen_ictal=False, pipeline=[Stats()]),
# Pipeline(gen_ictal=False, pipeline=[DaubWaveletStats(4)]),
# Pipeline(gen_ictal=False, pipeline=[Resample(400), DaubWaveletStats(4)]),
# Pipeline(gen_ictal=False, pipeline=[Resample(400), MFCC()]),
# Pipeline(gen_ictal=False, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'us')]),
# Pipeline(gen_ictal=True, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'us')]),
#Pipeline(gen_ictal=False, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'usf')]), # winning detection submission
# Pipeline(gen_ictal=False, pipeline=[WindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=False, pipeline=[StdWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=False, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
# Pipeline(gen_ictal=True, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
# Pipeline(gen_ictal=2, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
# Pipeline(gen_ictal=4, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=8, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindow1FFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
# Pipeline(gen_ictal=-8, pipeline=[MedianWindowBands1('usf2', 60, p=2)]),
# Pipeline(gen_ictal=-8, pipeline=[MedianWindowBands1('141022-PCA-model', 60, p=2)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindowBands1('141022-ICA-model-1', 60, p=2)]),
# Pipeline(gen_ictal=-8, pipeline=[MedianWindowBands1('ica', 60, p=2, timecorr=True)]),
#Pipeline(gen_ictal=-8.5, pipeline=[MedianWindowBands1('usf', 60, p=2)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindowBands('usf', 10, p=2, window='hammingP2')]),
#Pipeline(gen_ictal=-8, pipeline=[AllBands('usf', 60)]),
Pipeline(gen_ictal=-8, pipeline=[AllTimeCorrelation('usf', 60)]),
#Pipeline(gen_ictal=-8, pipeline=[MaxDiff(60)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindowBandsTimeCorrelation('usf', 60)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindowBandsCorrelation('usf', 60)]),
#Pipeline(gen_ictal=-8, pipeline=[MedianWindowTimeCorrelation('usf', 60)]),
#Pipeline(gen_ictal=False, pipeline=[MedianWindow1FFTWithTimeFreqCorrelation(1, 49, 400, 'usf',600)]),
#Pipeline(gen_ictal=8, pipeline=[MedianWindowFFTWithTimeFreqCov2(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=8, pipeline=[CleanMedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, window='hammingP2')]),
#Pipeline(gen_ictal=8, pipeline=[CleanCorMedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, window='hammingP2')]),
#Pipeline(gen_ictal=8, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600,subsample=2)]),
# Pipeline(gen_ictal=16, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
#Pipeline(gen_ictal=16, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 96, 400, 'usf',600, window='hamming')]),
#Pipeline(gen_ictal=16, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, window='hamming2')]),
#Pipeline(gen_ictal=8, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, window='hamming0')]),
# Pipeline(gen_ictal=8, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, window='square0')]),
# Pipeline(gen_ictal=2, pipeline=[Variance(nwindows=600)]),
# UnionPipeline(gen_ictal=2, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600),Variance(nwindows=600)]),
#Pipeline(gen_ictal=True, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600, nunits=4)]),
#Pipeline(gen_ictal=True, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 50, 400, 'usf',600)]),
#Pipeline(gen_ictal=False, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600,[0.5,0.9])]),
# Pipeline(gen_ictal=False, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600,[0.1,0.9])]),
# Pipeline(gen_ictal=False, pipeline=[MedianWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600,[0.05,0.5,0.95])]),
# Pipeline(gen_ictal=False, pipeline=[BoxWindowFFTWithTimeFreqCorrelation(1, 48, 400, 'usf',600)]),
# Pipeline(gen_ictal=True, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'usf')]), # higher score than winning submission
# Pipeline(gen_ictal=False, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'none')]),
# Pipeline(gen_ictal=True, pipeline=[FFTWithTimeFreqCorrelation(1, 48, 400, 'none')]),
# Pipeline(gen_ictal=False, pipeline=[TimeCorrelation(400, 'usf', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[TimeCorrelation(400, 'us', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[TimeCorrelation(400, 'us', with_corr=True, with_eigen=False)]),
# Pipeline(gen_ictal=False, pipeline=[TimeCorrelation(400, 'us', with_corr=False, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[TimeCorrelation(400, 'none', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[FreqCorrelation(1, 48, 'usf', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[FreqCorrelation(1, 48, 'us', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[FreqCorrelation(1, 48, 'us', with_corr=True, with_eigen=False)]),
# Pipeline(gen_ictal=False, pipeline=[FreqCorrelation(1, 48, 'us', with_corr=False, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[FreqCorrelation(1, 48, 'none', with_corr=True, with_eigen=True)]),
# Pipeline(gen_ictal=False, pipeline=[TimeFreqCorrelation(1, 48, 400, 'us')]),
# Pipeline(gen_ictal=False, pipeline=[TimeFreqCorrelation(1, 48, 400, 'usf')]),
# Pipeline(gen_ictal=False, pipeline=[TimeFreqCorrelation(1, 48, 400, 'none')]),
]
classifiers = [
# NOTE(mike): you can enable multiple classifiers to run them all and compare results
# (RandomForestClassifier(n_estimators=50, min_samples_split=1, bootstrap=False, n_jobs=4, random_state=0), 'rf50mss1Bfrs0'),
# (RandomForestClassifier(n_estimators=150, min_samples_split=1, bootstrap=False, n_jobs=4, random_state=0), 'rf150mss1Bfrs0'),
# (RandomForestClassifier(n_estimators=300, min_samples_split=1, bootstrap=False, n_jobs=4, random_state=0), 'rf300mss1Bfrs0'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, bootstrap=False, n_jobs=4, random_state=0), 'rf3000mss1Bfrs0'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=10, bootstrap=True, n_jobs=-1, random_state=0), 'rf3000mss1md10Bt'),
# (RandomForestClassifier(n_estimators=1000, min_samples_split=1, max_depth=10, bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md10Bf'),
(RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=10, bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md10Bf'),
# (RandomForestClassifier(n_estimators=10000, min_samples_split=1, max_depth=10, bootstrap=False, n_jobs=-1, random_state=0), 'rf10000mss1md10Bf'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=3, bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md3Bf'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=30, bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md30Bf'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=10, max_features='log2', bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md10BfmfL2'),
# (RandomForestClassifier(n_estimators=3000, min_samples_split=1, max_depth=10, max_features=200, bootstrap=False, n_jobs=-1, random_state=0), 'rf3000mss1md10Bfmf200'),
# (GradientBoostingClassifier(n_estimators=500,min_samples_split=1,),'gbc500mss1'),
# (GradientBoostingClassifier(n_estimators=1000,min_samples_split=1, random_state=0),'gbc1000mss1'),
# (GradientBoostingClassifier(n_estimators=1000,min_samples_split=1, random_state=0, learning_rate=0.03),'gbc1000mss1lr03'),
# (GradientBoostingClassifier(n_estimators=1000,min_samples_split=1, random_state=0, learning_rate=0.01),'gbc1000mss1lr01'),
# (GradientBoostingClassifier(n_estimators=1000,min_samples_split=1, random_state=0, learning_rate=0.01, max_depth=1000),'gbc1000mss1lr01md1000'),
]
cv_ratio = 0.5
def should_normalize(classifier):
clazzes = [LogisticRegression]
return np.any(np.array([isinstance(classifier, clazz) for clazz in clazzes]) == True)
def train_full_model(make_predictions):
for pipeline in pipelines:
for (classifier, classifier_name) in classifiers:
print 'Using pipeline %s with classifier %s' % (pipeline.get_name(), classifier_name)
if seizure.tasks.task_predict:
guesses = ['clip,preictal']
else:
guesses = ['clip,seizure,early']
classifier_filenames = []
for target in targets:
if isinstance(target,tuple):
target, leaderboard_weight = target
else:
leaderboard_weight = 1
task_core = TaskCore(cached_data_loader=cached_data_loader, data_dir=data_dir,
target=target, pipeline=pipeline,
classifier_name=classifier_name, classifier=classifier,
normalize=should_normalize(classifier), gen_ictal=pipeline.gen_ictal,
cv_ratio=cv_ratio)
if make_predictions:
predictions = MakePredictionsTask(task_core).run()
guesses.append(predictions.data)
else:
task = TrainClassifierTask(task_core)
task.run()
classifier_filenames.append(task.filename())
if make_predictions:
filename = 'submission%d-%s_%s.csv' % (ts, classifier_name, pipeline.get_name())
filename = os.path.join(submission_dir, filename)
with open(filename, 'w') as f:
print >> f, '\n'.join(guesses)
print 'wrote', filename
else:
print 'Trained classifiers ready in %s' % cache_dir
for filename in classifier_filenames:
print os.path.join(cache_dir, filename + '.pickle')
def do_cross_validation():
summaries = []
for pipeline in pipelines:
for (classifier, classifier_name) in classifiers:
print 'Using pipeline %s with classifier %s' % (pipeline.get_name(), classifier_name)
scores = []
S_scores = []
E_scores = []
leaderboard_weights = []
for target in targets:
if isinstance(target,tuple):
target, leaderboard_weight = target
else:
leaderboard_weight = 1
leaderboard_weights.append(leaderboard_weight)
print 'Processing %s (classifier %s)' % (target, classifier_name)
task_core = TaskCore(cached_data_loader=cached_data_loader, data_dir=data_dir,
target=target, pipeline=pipeline,
classifier_name=classifier_name, classifier=classifier,
normalize=should_normalize(classifier), gen_ictal=pipeline.gen_ictal,
cv_ratio=cv_ratio)
data = CrossValidationScoreTask(task_core).run()
score = data.score
scores.append(score)
print '%.3f' % score, 'S=%.4f' % data.S_auc, 'E=%.4f' % data.E_auc
S_scores.append(data.S_auc)
E_scores.append(data.E_auc)
if len(scores) > 0:
name = pipeline.get_name() + '_' + classifier_name
weighted_average = np.average(scores, weights=leaderboard_weights)
summary = get_score_summary(name, scores, weighted_average)
summaries.append((summary, weighted_average))
print summary
if len(S_scores) > 0:
name = pipeline.get_name() + '_' + classifier_name
summary = get_score_summary(name, S_scores, np.mean(S_scores))
print 'S', summary
if len(E_scores) > 0:
name = pipeline.get_name() + '_' + classifier_name
summary = get_score_summary(name, E_scores, np.mean(E_scores))
print 'E', summary
print_results(summaries)
def do_train_data():
for pipeline in pipelines:
print 'Using pipeline %s' % (pipeline.get_name())
for target in targets:
if isinstance(target,tuple):
target, leaderboard_weight = target
task_core = TaskCore(cached_data_loader=cached_data_loader, data_dir=data_dir,
target=target, pipeline=pipeline,
classifier_name=None, classifier=None,
normalize=None, gen_ictal=pipeline.gen_ictal,
cv_ratio=None)
# call the load data tasks for positive and negative examples (ignore the merge of the two.)
TrainingDataTask(task_core).run()
def do_test_data():
for pipeline in pipelines:
print 'Using pipeline %s' % (pipeline.get_name())
for target in targets:
if isinstance(target,tuple):
target, leaderboard_weight = target
task_core = TaskCore(cached_data_loader=cached_data_loader, data_dir=data_dir,
target=target, pipeline=pipeline,
classifier_name=None, classifier=None,
normalize=None, gen_ictal=pipeline.gen_ictal,
cv_ratio=None)
LoadTestDataTask(task_core).run()
if build_target == 'train_data':
do_train_data()
elif build_target == 'test_data':
do_test_data()
elif build_target == 'cv':
do_cross_validation()
elif build_target == 'train_model':
train_full_model(make_predictions=False)
elif build_target == 'make_predictions':
train_full_model(make_predictions=True)
else:
raise Exception("unknown build target %s" % build_target)
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description='Seizure prediction or detection')
parser.add_argument('-b','--build',
help='select what we want to build: train_data, test_data, cv, train_model, make_predictions')
parser.add_argument('-t', '--targets', nargs='+',
help='Target file(s) we want to process')
args = parser.parse_args()
if args.build == 'predict':
args.build = 'make_predictions'
elif args.build == 'train':
args.build = 'train_model'
elif args.build == 'td':
args.build = 'train_data'
elif args.build == 'tt':
args.build = 'test_data'
assert args.build in ['data', 'train_data', 'test_data', 'cv', 'train_model', 'make_predictions'], \
'Illegal/missing build command %s'%args.build
assert len(args.targets), 'No target(s) were given'
if args.build == 'data':
run_seizure_detection('train_data', targets=args.targets)
run_seizure_detection('test_data', targets=args.targets)
else:
run_seizure_detection(args.build, targets=args.targets)