-
Notifications
You must be signed in to change notification settings - Fork 1
/
ml_baseline.py
320 lines (268 loc) · 15.5 KB
/
ml_baseline.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
from sklearn.model_selection import StratifiedKFold, KFold, train_test_split
from sklearn.metrics import roc_auc_score, make_scorer, average_precision_score, precision_recall_curve, accuracy_score, \
f1_score, auc, mean_squared_error
from sklearn.linear_model import RidgeCV, LassoCV, ElasticNet, SGDClassifier, SGDRegressor
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor, RandomForestClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler
from scipy.stats import pearsonr, spearmanr
from collections import defaultdict
from functools import partial
from data import DataProvider
import random
import json
import numpy as np
import numpy.ma as ma
import pandas as pd
def auprc(y_true, y_score):
lr_precision, lr_recall, _ = precision_recall_curve(y_true=y_true, probas_pred=y_score)
return auc(lr_recall, lr_precision)
def pearson(y_true, y_pred):
r_val, p_val = pearsonr(x=y_true, y=y_pred)
return r_val
def spearman(y_true, y_pred):
r_val, p_val = spearmanr(a=y_true, b=y_pred)
return r_val
classify_scoring = {
'auroc': 'roc_auc',
'auprc': make_scorer(auprc, needs_proba=True),
'acc': make_scorer(accuracy_score),
'f1': make_scorer(f1_score),
'aps': make_scorer(average_precision_score, needs_proba=True)
}
regress_scoring = {
'mse': make_scorer(mean_squared_error),
'pearsonr': make_scorer(pearson),
'spearmanr': make_scorer(spearman),
}
def classify_with_rf(train_features, y_train, cv_split_rf, metric='auroc'):
try:
# logger.debug("Training Random Forest model")
# mx_depth: trees' maximum depth
# n_estimators: number of trees to use
# n_jobs = -1 means to run the jobs in parallel
rf_tuning_parameters = [{'n_estimators': [10, 50, 200, 500, 1000], 'max_depth': [10, 50, 100, 200, 500]}]
# rf_tuning_parameters = [{'n_estimators': [5], 'max_depth': [10]}]
rf = GridSearchCV(RandomForestClassifier(), rf_tuning_parameters, n_jobs=-1, cv=cv_split_rf,
verbose=2, scoring=classify_scoring, refit=metric)
scaler = StandardScaler()
train_features = scaler.fit_transform(train_features)
rf.fit(train_features, y_train) # , groups=train_groups
# logger.debug("Trained Random Forest successfully")
return rf, scaler
except Exception as e:
# logger.debug("Fail to Train Random Forest, caused by %s" % e.message)
raise e
def regress_with_rf(train_features, y_train, cv_split_rf, metric='pearsonr'):
try:
# logger.debug("Training Random Forest model")
# mx_depth: trees' maximum depth
# n_estimators: number of trees to use
# n_jobs = -1 means to run the jobs in parallel
rf_tuning_parameters = [{'n_estimators': [10, 50, 200, 500, 1000], 'max_depth': [10, 50, 100, 200, 500]}]
# rf_tuning_parameters = [{'n_estimators': [5], 'max_depth': [10]}]
rf = GridSearchCV(RandomForestRegressor(), rf_tuning_parameters, n_jobs=-1, cv=cv_split_rf,
verbose=2, scoring=regress_scoring, refit=metric)
scaler = StandardScaler()
train_features = scaler.fit_transform(train_features)
rf.fit(train_features, y_train) # , groups=train_groups
# logger.debug("Trained Random Forest successfully")
return rf, scaler
except Exception as e:
# logger.debug("Fail to Train Random Forest, caused by %s" % e.message)
raise e
def classify_with_enet(train_features, y_train, cv_split_enet, metric='auroc'):
try:
# logger.debug("Training elastic net regression model")
alphas = [0.1, 0.01, 0.001, 0.0001, 0.00001]
l1_ratios = [0.0, 0.25, 0.5, 0.75, 1.0]
# alphas = [0.1]
# l1_ratios = [0.25]
base_enet = SGDClassifier(loss='log', penalty='elasticnet', random_state=12345)
enet_param_grid = dict(alpha=alphas, l1_ratio=l1_ratios)
enet = GridSearchCV(estimator=base_enet, param_grid=enet_param_grid, n_jobs=-1, cv=cv_split_enet, verbose=2,
scoring=classify_scoring, refit=metric)
scaler = StandardScaler()
train_features = scaler.fit_transform(train_features)
enet.fit(train_features, y_train)
# logger.debug("Trained Elastic net classification model successfully")
return enet, scaler
except Exception as e:
raise e
def regress_with_enet(train_features, y_train, cv_split_enet, metric='pearsonr'):
try:
# logger.debug("Training elastic net regression model")
alphas = [0.1, 0.01, 0.001, 0.0001, 0.00001]
l1_ratios = [0.0, 0.25, 0.5, 0.75, 1.0]
# alphas = [0.1]
# l1_ratios = [0.25]
# base_enet = SGDRegressor(penalty='elasticnet', random_state=12345)
base_enet = ElasticNet(random_state=12345, max_iter=5000)
enet_param_grid = dict(alpha=alphas, l1_ratio=l1_ratios)
enet = GridSearchCV(estimator=base_enet, param_grid=enet_param_grid, n_jobs=-1, cv=cv_split_enet, verbose=2,
scoring=regress_scoring, refit=metric)
scaler = StandardScaler()
train_features = scaler.fit_transform(train_features)
enet.fit(train_features, y_train)
# logger.debug("Trained Elastic net classification model successfully")
return enet, scaler
except Exception as e:
raise e
def n_time_cv_classify(train_data, n=10, model_fn=classify_with_enet, test_data=None, random_state=2020,
metric='auroc'):
# metric_list = ['auroc', 'acc', 'aps', 'f1']
metric_list = ['auroc', 'acc', 'aps', 'f1', 'auprc']
random.seed(random_state)
seeds = random.sample(range(100000), k=n)
train_history = defaultdict(list)
test_history = defaultdict(list)
models = []
for seed in seeds:
kfold = StratifiedKFold(n_splits=5, shuffle=True, random_state=seed)
cv_split = kfold.split(*train_data)
trained_model, scaler = model_fn(*train_data, list(cv_split), metric=metric)
for metric in metric_list:
train_history[metric].append(trained_model.cv_results_[f'mean_test_{metric}'][trained_model.best_index_])
if test_data is not None:
# preds = trained_model.predict(test_data[0])
# pred_scores = trained_model.predict_proba(test_data[0])[:, 1]
preds = trained_model.predict(scaler.transform(test_data[0]))
pred_scores = trained_model.predict_proba(scaler.transform(test_data[0]))[:, 1]
# print(preds)
# print(pred_scores)
test_history['auroc'].append(roc_auc_score(y_true=test_data[1], y_score=pred_scores))
test_history['acc'].append(accuracy_score(y_true=test_data[1], y_pred=preds))
test_history['aps'].append(average_precision_score(y_true=test_data[1], y_score=pred_scores))
test_history['f1'].append(f1_score(y_true=test_data[1], y_pred=preds))
test_history['auprc'].append(auprc(y_true=test_data[1], y_score=pred_scores))
models.append(trained_model)
return (train_history, models) if test_data is None else (train_history, test_history, models)
def multi_regress(train_data, output_file_name, model_fn=regress_with_enet, test_data=None, random_state=2020):
train_feature_df = train_data[0]
train_target_df = train_data[1]
train_pred_df = pd.DataFrame(np.full_like(train_target_df, fill_value=-1),
index=train_target_df.index,
columns=train_target_df.columns)
if test_data is not None:
test_feature_df = test_data[0]
test_target_df = test_data[1]
test_pred_df = pd.DataFrame(np.full_like(test_target_df, fill_value=-1),
index=test_target_df.index,
columns=test_target_df.columns)
assert all(train_pred_df.columns == test_pred_df.columns)
for drug in train_pred_df.columns:
print("Training: {}".format(drug))
y = train_target_df.loc[~train_target_df[drug].isna(), drug]
sample_ids = train_feature_df.index.intersection(y.index)
X = train_feature_df.loc[sample_ids]
# print(X.columns)
y = y.reindex(sample_ids)
outer_kfold = KFold(n_splits=5, shuffle=True, random_state=random_state)
for train_index, test_index in outer_kfold.split(X, y):
X_train, X_test = X.iloc[train_index, :], X.iloc[test_index, :]
y_train, y_test = y[train_index], y[test_index]
kfold = KFold(n_splits=5, shuffle=True, random_state=random_state)
cv_split = kfold.split(X_train)
try:
trained_model, scaler = model_fn(X_train, y_train, list(cv_split))
train_pred_df.loc[y.index[test_index], drug] = trained_model.predict(scaler.transform(X_test))
except Exception as e:
print(e)
assert all(train_pred_df.index==train_target_df.index)
train_pred_df.to_csv(f'./predictions/{output_file_name}.csv', index_label='Sample')
if test_data is not None:
outer_kfold = KFold(n_splits=5, shuffle=True, random_state=random_state)
cv_split = outer_kfold.split(X)
try:
trained_model, scaler = model_fn(X, y, list(cv_split))
test_pred_df.loc[test_feature_df.index, drug] = trained_model.predict(scaler.transform(test_feature_df))
except Exception as e:
print(e)
assert all(test_pred_df.index == test_target_df.index)
test_pred_df.to_csv(f'./predictions/test_{output_file_name}.csv', index_label='Sample')
return (train_target_df.values, train_pred_df.values) if test_data is None else (
train_target_df.values, train_pred_df.values, test_target_df.values, test_pred_df.values)
def n_time_cv_regress(train_data, output_file_name, n=5, model_fn=regress_with_enet, test_data=None, random_state=2020):
random.seed(random_state)
seeds = random.sample(range(100000), k=int(n))
train_history = defaultdict(list)
test_history = defaultdict(list)
for seed in seeds:
if test_data is None:
train_y_truths, train_y_preds = multi_regress(train_data, f'{output_file_name}_{seed}',
model_fn=model_fn, test_data=test_data,
random_state=seed)
else:
train_y_truths, train_y_preds, test_y_truths, test_y_preds = multi_regress(train_data,
f'{output_file_name}_{seed}',
model_fn=model_fn,
test_data=test_data,
random_state=random_state)
test_history['dpearsonr'].append(
np.mean([pearsonr(test_y_truths[:, i][~ma.masked_invalid(test_y_truths[:, i]).mask],
test_y_preds[:, i][~ma.masked_invalid(test_y_truths[:, i]).mask])[
0]
for i in range(test_y_truths.shape[1])]).item())
test_history['cpearsonr'].append(
np.mean([pearsonr(test_y_truths[i, :][~ma.masked_invalid(test_y_truths[i, :]).mask],
test_y_preds[i, :][~ma.masked_invalid(test_y_truths[i, :]).mask])[
0]
for i in range(test_y_truths.shape[0])]).item())
test_history['dspearman'].append(
np.mean([spearmanr(test_y_truths[:, i][~ma.masked_invalid(test_y_truths[:, i]).mask],
test_y_preds[:, i][~ma.masked_invalid(test_y_truths[:, i]).mask])[
0]
for i in range(test_y_truths.shape[1])]).item())
test_history['cspearman'].append(
np.mean([spearmanr(test_y_truths[i, :][~ma.masked_invalid(test_y_truths[i, :]).mask],
test_y_preds[i, :][~ma.masked_invalid(test_y_truths[i, :]).mask])[
0]
for i in range(test_y_truths.shape[0])]).item())
test_history['drmse'].append(
np.mean(np.nanmean(np.square((test_y_truths - test_y_preds)), axis=0)).item())
test_history['crmse'].append(
np.mean(np.nanmean(np.square((test_y_truths - test_y_preds)), axis=1)).item())
train_history['dpearsonr'].append(
np.mean([pearsonr(train_y_truths[:, i][~ma.masked_invalid(train_y_truths[:, i]).mask],
train_y_preds[:, i][~ma.masked_invalid(train_y_truths[:, i]).mask])[
0]
for i in range(train_y_truths.shape[1])]).item())
train_history['cpearsonr'].append(
np.mean([pearsonr(train_y_truths[i, :][~ma.masked_invalid(train_y_truths[i, :]).mask],
train_y_preds[i, :][~ma.masked_invalid(train_y_truths[i, :]).mask])[
0]
for i in range(train_y_truths.shape[0])]).item())
train_history['dspearman'].append(
np.mean([spearmanr(train_y_truths[:, i][~ma.masked_invalid(train_y_truths[:, i]).mask],
train_y_preds[:, i][~ma.masked_invalid(train_y_truths[:, i]).mask])[
0]
for i in range(train_y_truths.shape[1])]).item())
train_history['cspearman'].append(
np.mean([spearmanr(train_y_truths[i, :][~ma.masked_invalid(train_y_truths[i, :]).mask],
train_y_preds[i, :][~ma.masked_invalid(train_y_truths[i, :]).mask])[
0]
for i in range(train_y_truths.shape[0])]).item())
train_history['drmse'].append(np.mean(np.nanmean(np.square((train_y_truths - train_y_preds)), axis=0)).item())
train_history['crmse'].append(np.mean(np.nanmean(np.square((train_y_truths - train_y_preds)), axis=1)).item())
return (train_history, test_history)
if __name__ == '__main__':
data_provider = DataProvider()
labeled_samples, mut_only_labeled_samples = data_provider.get_labeled_samples()
labeled_target_df = data_provider.target_df.loc[labeled_samples]
labeled_mut_only_target_df = data_provider.target_df.loc[mut_only_labeled_samples]
label_gex_df = data_provider.gex_dat.loc[labeled_samples]
label_mut_df = data_provider.mut_dat.loc[labeled_samples]
label_mut_only_df = data_provider.mut_dat.loc[mut_only_labeled_samples]
train_gex_data = (label_gex_df, labeled_target_df)
train_mut_data = (label_mut_df, labeled_target_df)
test_data = (label_mut_only_df, labeled_mut_only_target_df)
gex_train_history, _ = n_time_cv_regress(train_gex_data, 'gex_pred', n=5, model_fn=regress_with_enet,
test_data=None, random_state=2020)
with open('./predictions/gex_pred.json', 'w') as f:
json.dump(gex_train_history, f)
mut_train_history, mut_test_history = n_time_cv_regress(train_mut_data, 'mut_pred', n=5, model_fn=regress_with_enet,
test_data=test_data, random_state=2020)
with open('./predictions/mut_pred.json', 'w') as f:
json.dump(mut_train_history, f)
with open('./predictions/test_mut_pred.json', 'w') as f:
json.dump(mut_test_history, f)