-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
211 lines (176 loc) · 6.92 KB
/
train.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
import argparse
import os
import joblib
import numpy as np
import torch
import yaml
from expt.common import clf_map, synthetic_params, train_func_map
from sklearn.metrics import accuracy_score, roc_auc_score
from sklearn.model_selection import KFold, train_test_split
from utils import helpers
from utils.transformer import get_transformer
arrival_data = {"synthesis": False, "german": False, "sba": False, "gmc": False}
def eval_performance(model, X_test, y_test):
y_prob = model.predict_proba(X_test)
y_pred = np.argmax(y_prob, axis=-1)
accuracy = accuracy_score(y_test, y_pred)
auc = roc_auc_score(y_test, y_prob[:, 1])
return accuracy, auc
def train_model(X_train, y_train, X_test, y_test, train_func, clf, d, lr, num_epoch, idx, verbose, random_state):
print("training classifier ", idx, flush=True)
torch.manual_seed(random_state)
np.random.seed(random_state + 1)
model = clf(d)
train_func(model, X_train, y_train, lr, num_epoch, verbose)
acc, auc = eval_performance(model, X_test, y_test)
return model, acc, auc
def train(
clf_name,
data_name,
wdir,
lr,
num_epoch,
kfold=5,
num_future=100,
seed=123,
verbose=False,
num_proc=1,
append_arrival=True,
arrival_ratio=0.50,
train_shift_size=0.8,
):
transformer = get_transformer(data_name)
df, _ = helpers.get_dataset(data_name, params=synthetic_params)
y = df["label"].to_numpy()
X = df.drop("label", axis=1)
X = transformer.transform(X)
if not type(X) is np.ndarray:
X = X.toarray()
df_shift, _ = helpers.get_dataset(data_name + "shift", params=synthetic_params)
y_shift = df_shift["label"].to_numpy()
X_shift = df_shift.drop("label", axis=1)
X_shift = transformer.transform(X_shift)
if not type(X_shift) is np.ndarray:
X_shift = X_shift.toarray()
d = X.shape[1]
clf = clf_map[clf_name]
train_func = train_func_map[clf_name]
report = {}
# Train present data
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.8, random_state=42, stratify=y)
kf = KFold(n_splits=kfold)
jobs_args = []
for i, (train_index, cross_index) in enumerate(kf.split(X_train)):
X_training = X_train[train_index]
y_training = y_train[train_index]
jobs_args.append((X_training, y_training, X_test, y_test, train_func, clf, d, lr, num_epoch, i, verbose, seed))
rets = joblib.Parallel(n_jobs=num_proc)(joblib.delayed(train_model)(*args) for args in jobs_args)
# rets = []
# for i in range(len(jobs_args)):
# rets.append(train_model(*jobs_args[i]))
cur_auc = []
cur_acc = []
cur_models = []
for model, acc, auc in rets:
cur_acc.append(acc)
cur_auc.append(auc)
cur_models.append(model)
model = cur_models[0]
report["append_arrival"] = append_arrival
report["arrival_ratio"] = arrival_ratio
report["train_shift_size"] = train_shift_size
report["cur_acc_mean"] = float(np.mean(cur_acc))
report["cur_acc_std"] = float(np.std(cur_acc))
report["cur_auc_mean"] = float(np.mean(cur_auc))
report["cur_auc_std"] = float(np.std(cur_auc))
name = f"{clf_name}_{data_name}_{kfold}.pickle"
helpers.pdump(cur_models, name, wdir)
print(
"Trained classifier: {} on current dataset: {}, and saved to {}".format(
clf_name, data_name, os.path.join(wdir, name)
)
)
# Train shift data
for i, (train_index, cross_index) in enumerate(kf.split(X_train)):
jobs_args = []
X_training = X_train[train_index]
y_training = y_train[train_index]
for rng in range(num_future):
if append_arrival:
if arrival_ratio != 0:
arrival_X, _, arrival_y, _ = train_test_split(
X_shift, y_shift, train_size=arrival_ratio, random_state=rng, stratify=y_shift
)
future_X = np.vstack([X_training, arrival_X])
future_y = np.concatenate([y_training, arrival_y])
else:
future_X, future_y = X_training, y_training
else:
future_X, X_test, future_y, y_test = train_test_split(
X_shift, y_shift, train_size=train_shift_size, random_state=rng, stratify=y_shift
)
jobs_args.append(
(future_X, future_y, X_test, y_test, train_func, clf, d, lr, num_epoch, rng, verbose, seed)
)
rets = joblib.Parallel(n_jobs=num_proc)(joblib.delayed(train_model)(*args) for args in jobs_args)
shift_auc = []
shift_acc = []
models = []
for model, acc, auc in rets:
shift_acc.append(acc)
shift_auc.append(auc)
models.append(model)
report["shift_acc_mean"] = float(np.mean(shift_acc))
report["shift_acc_std"] = float(np.std(shift_acc))
report["shift_auc_mean"] = float(np.mean(shift_auc))
report["shift_auc_std"] = float(np.std(shift_auc))
name = f"{clf_name}_{data_name}_shift_{i}_{num_future}.pickle"
helpers.pdump(models, name, wdir)
print(
"Trained classifier: {} on shifted dataset: {}, and saved to {}".format(
clf_name, data_name, os.path.join(wdir, name)
)
)
return report
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Train a classifier")
parser.add_argument("--clf", "-c", dest="clfs", nargs="*")
parser.add_argument("--data", "-d", dest="datasets", nargs="*")
parser.add_argument("--lr", "-lr", default=1e-3, type=float)
parser.add_argument("--epoch", default=1000, type=int)
parser.add_argument("--kfold", default=5, type=int)
parser.add_argument("--num-future", "-nf", default=100, type=int)
parser.add_argument("--num-proc", default=1, type=int)
parser.add_argument("--run-id", default=0, type=int)
parser.add_argument("--verbose", "-v", action="store_true")
parser.add_argument("--seed", "-s", default=123, type=int)
args = parser.parse_args()
torch.set_printoptions(sci_mode=False)
seed = 46
torch.manual_seed(args.seed + 12)
np.random.seed(args.seed + 11)
np.set_printoptions(suppress=False)
wdir = f"results/run_{args.run_id}/checkpoints"
os.makedirs(wdir, exist_ok=True)
report = {}
for clf in args.clfs:
clf_report = {}
for data in args.datasets:
print("training on dataset: ", data)
data_report = train(
clf,
data,
wdir,
args.lr,
args.epoch,
args.kfold,
args.num_future,
args.seed,
args.verbose,
args.num_proc,
)
clf_report[data] = data_report
report[clf] = clf_report
filepath = f"{wdir}/report.txt"
with open(filepath, mode="w") as file:
yaml.dump(report, file)