/
eICU_tstr_evaluation.py
245 lines (205 loc) · 8.07 KB
/
eICU_tstr_evaluation.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
import data_utils
import pandas as pd
import numpy as np
import tensorflow as tf
import math, random, itertools
import pickle
import time
import json
import os
import math
import data_utils
import pickle
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score, precision_score, recall_score, roc_curve, auc, precision_recall_curve
import copy
from scipy.stats import sem
print ("Starting TSTR experiment.")
print ("loading data...")
samples, labels = data_utils.eICU_task()
train_seqs = samples['train'].reshape(-1,16,4)
vali_seqs = samples['vali'].reshape(-1,16,4)
test_seqs = samples['test'].reshape(-1,16,4)
train_targets = labels['train']
vali_targets = labels['vali']
test_targets = labels['test']
train_seqs, vali_seqs, test_seqs = data_utils.scale_data(train_seqs, vali_seqs, test_seqs)
print ("data loaded.")
# iterate over all dataset versions generated after running the GAN for 5 times
aurocs_all_runs = []
auprcs_all_runs = []
for oo in range(5):
print (oo)
# find the best "dataset epoch", meaning the GAN epoch that generated the dataset
# validation is only done in some of the tasks, and the others are considered unknown
# (use validation set to pick best GAN epoch, then get result on test set)
vali_seqs_r = vali_seqs.reshape((vali_seqs.shape[0], -1))
test_seqs_r = test_seqs.reshape((test_seqs.shape[0], -1))
all_aurocs_exp = []
all_auprcs_exp = []
for nn in np.arange(50,1050,50):
with open('./synthetic_eICU_datasets/samples_eICU_cdgan_synthetic_dataset_r' + str(oo) + '_' + str(nn) + '.pk', 'rb') as f:
synth_data = pickle.load(file=f)
with open('./synthetic_eICU_datasets/labels_eICU_cdgan_synthetic_dataset_r' + str(oo) + '_' + str(nn) + '.pk', 'rb') as f:
synth_labels = pickle.load(file=f)
train_seqs = synth_data
train_targets = synth_labels
train_seqs_r = train_seqs.reshape((train_seqs.shape[0], -1))
all_aurocs = []
all_auprcs = []
# in case we want to train each random forest multiple times with each dataset
for exp_num in range(1):
accuracies = []
precisions = []
recalls = []
aurocs = []
auprcs = []
for col_num in range(train_targets.shape[1]):
estimator = RandomForestClassifier(n_estimators=100)
estimator.fit(train_seqs_r, train_targets[:,col_num])
accuracies.append(estimator.score(vali_seqs_r, vali_targets[:,col_num]))
preds = estimator.predict(vali_seqs_r)
precisions.append(precision_score(y_pred=preds, y_true=vali_targets[:,col_num]))
recalls.append(recall_score(y_pred=preds, y_true=vali_targets[:,col_num]))
preds = estimator.predict_proba(vali_seqs_r)
fpr, tpr, thresholds = roc_curve(vali_targets[:,col_num], preds[:,1])
aurocs.append(auc(fpr, tpr))
precision, recall, thresholds = precision_recall_curve(vali_targets[:,col_num], preds[:,1])
auprcs.append(auc(recall, precision))
all_aurocs.append(aurocs)
all_auprcs.append(auprcs)
all_aurocs_exp.append(all_aurocs)
all_auprcs_exp.append(all_auprcs)
#with open('all_aurocs_exp_r' + str(oo) + '.pk', 'wb') as f:
# pickle.dump(file=f, obj=all_aurocs_exp)
#with open('all_auprcs_exp_r' + str(oo) + '.pk', 'wb') as f:
# pickle.dump(file=f, obj=all_auprcs_exp)
best_idx = np.argmax(np.array(all_aurocs_exp).sum(axis=1)[:,[0,2,4]].sum(axis=1) + np.array(all_auprcs_exp).sum(axis=1)[:,[0,2,4]].sum(axis=1))
best = np.arange(50,1050,50)[best_idx]
with open('./synthetic_eICU_datasets/samples_eICU_cdgan_synthetic_dataset_r' + str(oo) + '_' + str(best) + '.pk', 'rb') as f:
synth_data = pickle.load(file=f)
with open('./synthetic_eICU_datasets/labels_eICU_cdgan_synthetic_dataset_r' + str(oo) + '_' + str(best) + '.pk', 'rb') as f:
synth_labels = pickle.load(file=f)
train_seqs = synth_data
train_targets = synth_labels
train_seqs_r = train_seqs.reshape((train_seqs.shape[0], -1))
accuracies = []
precisions = []
recalls = []
aurocs = []
auprcs = []
for col_num in range(train_targets.shape[1]):
estimator = RandomForestClassifier(n_estimators=100)
estimator.fit(train_seqs_r, train_targets[:,col_num])
accuracies.append(estimator.score(test_seqs_r, test_targets[:,col_num]))
preds = estimator.predict(test_seqs_r)
precisions.append(precision_score(y_pred=preds, y_true=test_targets[:,col_num]))
recalls.append(recall_score(y_pred=preds, y_true=test_targets[:,col_num]))
preds = estimator.predict_proba(test_seqs_r)
fpr, tpr, thresholds = roc_curve(test_targets[:,col_num], preds[:,1])
aurocs.append(auc(fpr, tpr))
precision, recall, thresholds = precision_recall_curve(test_targets[:,col_num], preds[:,1])
auprcs.append(auc(recall, precision))
print(accuracies)
print(precisions)
print(recalls)
print(aurocs)
print(auprcs)
print ("----------------------------")
aurocs_all_runs.append(aurocs)
auprcs_all_runs.append(auprcs)
allr = np.vstack(aurocs_all_runs)
allp = np.vstack(auprcs_all_runs)
tstr_aurocs_mean = allr.mean(axis=0)
tstr_aurocs_sem = sem(allr, axis=0)
tstr_auprcs_mean = allp.mean(axis=0)
tstr_auprcs_sem = sem(allp, axis=0)
# get AUROC/AUPRC for real, random data
print ("Experiment with real data.")
print ("loading data...")
samples, labels = data_utils.eICU_task()
train_seqs = samples['train'].reshape(-1,16,4)
vali_seqs = samples['vali'].reshape(-1,16,4)
test_seqs = samples['test'].reshape(-1,16,4)
train_targets = labels['train']
vali_targets = labels['vali']
test_targets = labels['test']
train_seqs, vali_seqs, test_seqs = data_utils.scale_data(train_seqs, vali_seqs, test_seqs)
print ("data loaded.")
train_seqs_r = train_seqs.reshape((train_seqs.shape[0], -1))
vali_seqs_r = vali_seqs.reshape((vali_seqs.shape[0], -1))
test_seqs_r = test_seqs.reshape((test_seqs.shape[0], -1))
aurocs_all = []
auprcs_all = []
for i in range(5):
accuracies = []
precisions = []
recalls = []
aurocs = []
auprcs = []
for col_num in range(train_targets.shape[1]):
estimator = RandomForestClassifier(n_estimators=100)
estimator.fit(train_seqs_r, train_targets[:,col_num])
accuracies.append(estimator.score(test_seqs_r, test_targets[:,col_num]))
preds = estimator.predict(test_seqs_r)
precisions.append(precision_score(y_pred=preds, y_true=test_targets[:,col_num]))
recalls.append(recall_score(y_pred=preds, y_true=test_targets[:,col_num]))
preds = estimator.predict_proba(test_seqs_r)
fpr, tpr, thresholds = roc_curve(test_targets[:,col_num], preds[:,1])
aurocs.append(auc(fpr, tpr))
precision, recall, thresholds = precision_recall_curve(test_targets[:,col_num], preds[:,1])
auprcs.append(auc(recall, precision))
print(accuracies)
print(precisions)
print(recalls)
print(aurocs)
print(auprcs)
aurocs_all.append(aurocs)
auprcs_all.append(auprcs)
real_aurocs_mean = np.array(aurocs_all).mean(axis=0)
real_aurocs_sem = sem(aurocs_all, axis=0)
real_auprcs_mean = np.array(auprcs_all).mean(axis=0)
real_auprcs_sem = sem(auprcs_all, axis=0)
print ("Experiment with random predictions.")
#random score
test_targets_random = copy.deepcopy(test_targets)
random.shuffle(test_targets_random)
accuracies = []
precisions = []
recalls = []
aurocs = []
auprcs = []
for col_num in range(train_targets.shape[1]):
accuracies.append(accuracy_score(y_pred=test_targets_random[:,col_num], y_true=test_targets[:,col_num]))
precisions.append(precision_score(y_pred=test_targets_random[:,col_num], y_true=test_targets[:,col_num]))
recalls.append(recall_score(y_pred=test_targets_random[:,col_num], y_true=test_targets[:,col_num]))
preds = np.random.rand(len(test_targets[:,col_num]))
fpr, tpr, thresholds = roc_curve(test_targets[:,col_num], preds)
aurocs.append(auc(fpr, tpr))
precision, recall, thresholds = precision_recall_curve(test_targets[:,col_num], preds)
auprcs.append(auc(recall, precision))
print(accuracies)
print(precisions)
print(recalls)
print(aurocs)
print(auprcs)
random_aurocs = aurocs
random_auprcs = auprcs
print("Results")
print("------------")
print("------------")
print("TSTR")
print(tstr_aurocs_mean)
print(tstr_aurocs_sem)
print(tstr_auprcs_mean)
print(tstr_auprcs_sem)
print("------------")
print("Real")
print(real_aurocs_mean)
print(real_aurocs_sem)
print(real_auprcs_mean)
print(real_auprcs_sem)
print("------------")
print("Random")
print(random_aurocs)
print(random_auprcs)