-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_tabular.py
202 lines (160 loc) · 7.09 KB
/
train_tabular.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
import argparse
import pandas as pd
import numpy as np
import os
import joblib
from sklearn.model_selection import RandomizedSearchCV, GridSearchCV
from skopt import BayesSearchCV
from skopt.space import Real, Categorical, Integer
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.utils.fixes import loguniform
from sklearn.ensemble import RandomForestClassifier
from xgboost import XGBClassifier
from sklearn.model_selection import PredefinedSplit
from sklearn.metrics import classification_report
from sklearn.impute import SimpleImputer, MissingIndicator
from sklearn.preprocessing import StandardScaler
from sklearn.feature_selection import RFECV, VarianceThreshold
import eco2ai
from models.timeseries.tabular import PreProcessor
from models.utils import set_seed
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--features", type=str, default='tsfresh')
parser.add_argument("--data_dir", type=str, default="dataset/cinc2020/raw/")
parser.add_argument("--csv_dir", type=str, default="dataset/cinc2020/processed/")
parser.add_argument("--log_dir", type=str, default="./logs/tabular/")
parser.add_argument("--debug", action='store_true', default=False)
parser.add_argument("--cache", action='store_true', default=False)
parser.add_argument("--n_iter", type=int, default=50)
parser.add_argument("--seed", type=int, default=42)
args = parser.parse_args()
return args
def train(args):
set_seed(args.seed)
os.makedirs(args.log_dir, exist_ok=True)
eco_tracker = eco2ai.Tracker(
file_name=os.path.join(args.log_dir, "emission.csv")
)
if args.debug:
args.cache = False
train_data = pd.read_csv(os.path.join(args.csv_dir, 'y_train.csv'))
val_data = pd.read_csv(os.path.join(args.csv_dir, 'y_val.csv'))
test_data = pd.read_csv(os.path.join(args.csv_dir, 'y_test.csv'))
if args.debug:
train_data = train_data.head(100)
val_data = val_data.head(50)
test_data = test_data.head(50)
target_names = train_data.drop(columns=['idx']).columns.to_list()
X_train = train_data['idx'].to_list()
y_train = train_data.drop(columns=['idx']).to_numpy()
X_val = val_data['idx'].to_list()
y_val = val_data.drop(columns=['idx']).to_numpy()
X_test = test_data['idx'].to_list()
y_test = test_data.drop(columns=['idx']).to_numpy()
if args.features == 'ecg':
preprocessor = PreProcessor(data_dir=args.data_dir)
if not args.cache:
X_train = preprocessor.fit_transform(X_train, verbose=True)
X_test = preprocessor.transform(X_test, verbose=True)
X_val = preprocessor.transform(X_val, verbose=True)
if not args.debug:
os.makedirs(
os.path.join(args.log_dir, 'cache'),
exist_ok=True,
)
joblib.dump(X_train, os.path.join(args.log_dir, 'cache/X_train.joblib'))
joblib.dump(X_val, os.path.join(args.log_dir, 'cache/X_val.joblib'))
joblib.dump(X_test, os.path.join(args.log_dir, 'cache/X_test.joblib'))
else:
X_train = joblib.load(os.path.join(args.log_dir, 'cache/X_train.joblib'))
X_test = joblib.load(os.path.join(args.log_dir, 'cache/X_test.joblib'))
X_val = joblib.load(os.path.join(args.log_dir, 'cache/X_val.joblib'))
elif args.features == 'tsfresh':
X_train = pd.read_csv(os.path.join(args.csv_dir, 'train_features.csv'))
X_val = pd.read_csv(os.path.join(args.csv_dir, 'val_features.csv'))
X_test = pd.read_csv(os.path.join(args.csv_dir, 'test_features.csv'))
colnames = [i for i in X_train.columns if i.startswith('x__')]
X_train = X_train[colnames]
X_val = X_val[colnames]
X_test = X_test[colnames]
X = np.vstack([X_train, X_val])
y = np.vstack([y_train, y_val])
test_fold = X_train.shape[0] * [-1] + X_val.shape[0] * [0]
splitter = PredefinedSplit(test_fold)
print('Data shape:', X.shape, y.shape)
preprocessor = Pipeline([
('scl', StandardScaler()),
('missing_handler', SimpleImputer(strategy='constant', fill_value=-999)),
# ('missing_handler', FeatureUnion(
# transformer_list=[
# ('features', SimpleImputer(strategy='constant', fill_value=-999)),
# ('indicators', MissingIndicator())
# ]
# )),
('vt', VarianceThreshold(threshold=0.16)),
('fs', RFECV(RandomForestClassifier(n_jobs=-1, random_state=42), step=100, cv=3, scoring='f1_micro')),
])
pipeline = Pipeline([
('pre', preprocessor),
('clf', None)
])
params = [
# {
# 'pre__fs__min_features_to_select': np.arange(100, 501, 100),
# 'clf': [RandomForestClassifier(random_state=args.seed, n_jobs=-1)],
# 'clf__n_estimators': [100, 500, 1000, 2000],
# 'clf__max_depth': np.arange(10, 100, 10),
# 'clf__class_weight': ['balanced', None],
# 'clf__max_features': ['sqrt', 'log2'],
# # 'clf__min_samples_split': [2, 10, 50],
# # 'clf__min_samples_leaf': [1, 5, 10],
# },
{
'pre__fs__min_features_to_select': Integer(10, X.shape[1]),
'clf': [XGBClassifier(tree_method='gpu_hist', gpu_id=0)],
'clf__max_depth': Integer(2, 100),
'clf__gamma': Real(1e-3, 1e+3, prior='log-uniform'),
'clf__eta': Real(1e-3, 1e+3, prior='log-uniform'),
'clf__scale_pos_weight': Real(1e-1, 1e+3, prior='log-uniform'),
'clf__reg_lambda': Real(1e-3, 1e+3, prior='log-uniform'),
'clf__reg_alpha': Real(1e-3, 1e+3, prior='log-uniform'),
# 'clf__objective': ['binary:logistic'],
# 'clf__min_child_weight': [1, 2, 5, 10],
# 'clf__subsample': [0.1, 0.2, 0.3],
# 'clf__colsample_bytree': np.linspace(0, 0.9, 10),
}
]
# import ipdb; ipdb.set_trace()
search = BayesSearchCV(
pipeline,
params,
cv=splitter,
scoring='f1_micro',
n_jobs=-1,
verbose=10,
random_state=args.seed,
n_iter=args.n_iter,
)
eco_tracker.start()
search.fit(X, y)
print('Best params:', search.best_params_)
y_pred = search.predict(X_test)
y_prob = search.predict_proba(X_test)
report = classification_report(y_test, y_pred, target_names=target_names)
print(report)
np.savez_compressed(os.path.join(args.log_dir, 'test_prob'), y_prob)
# Save report csv
report = classification_report(y_test, y_pred,
target_names=target_names, zero_division=0, output_dict=True)
report = pd.DataFrame(report).round(2).T.to_csv(
os.path.join(args.log_dir, 'report.csv')
)
# Save the results
cv_results = pd.DataFrame(search.cv_results_)
cv_results.to_csv(os.path.join(args.log_dir, 'cv_results.csv'), index=False)
joblib.dump(search, os.path.join(args.log_dir, 'search.pkl'))
eco_tracker.stop()
if __name__ == "__main__":
args = parse_args()
train(args)