-
Notifications
You must be signed in to change notification settings - Fork 561
/
train_ke.py
315 lines (282 loc) · 8.58 KB
/
train_ke.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
"""
Script for training model on Keras.
"""
import argparse
import time
import logging
import os
import numpy as np
import random
import keras
from keras.models import load_model
from keras.callbacks import ModelCheckpoint
import mxnet as mx
# from common.logger_utils import initialize_logging
from cvutil.logger import initialize_logging
from keras_.utils import prepare_ke_context, prepare_model, get_data_rec, get_data_generator, backend_agnostic_compile
def parse_args():
"""
Parse python script parameters.
Returns
-------
ArgumentParser
Resulted args.
"""
parser = argparse.ArgumentParser(
description="Train a model for image classification (Keras)",
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument(
"--rec-train",
type=str,
default="../imgclsmob_data/imagenet_rec/train.rec",
help="the training data")
parser.add_argument(
"--rec-train-idx",
type=str,
default="../imgclsmob_data/imagenet_rec/train.idx",
help='the index of training data')
parser.add_argument(
"--rec-val",
type=str,
default="../imgclsmob_data/imagenet_rec/val.rec",
help="the validation data")
parser.add_argument(
"--rec-val-idx",
type=str,
default="../imgclsmob_data/imagenet_rec/val.idx",
help="the index of validation data")
parser.add_argument(
"--model",
type=str,
required=True,
help="type of model to use. see model_provider for options")
parser.add_argument(
"--use-pretrained",
action="store_true",
help="enable using pretrained model from github repo")
parser.add_argument(
"--dtype",
type=str,
default="float32",
help="data type for training")
parser.add_argument(
"--resume",
type=str,
default="",
help="resume from previously saved parameters if not None")
parser.add_argument(
"--resume-state",
type=str,
default="",
help="resume from previously saved optimizer state if not None")
parser.add_argument(
"--input-size",
type=int,
default=224,
help="size of the input for model")
parser.add_argument(
"--resize-inv-factor",
type=float,
default=0.875,
help="inverted ratio for input image crop")
parser.add_argument(
"--num-gpus",
type=int,
default=0,
help="number of gpus to use")
parser.add_argument(
"-j",
"--num-data-workers",
dest="num_workers",
default=4,
type=int,
help="number of preprocessing workers")
parser.add_argument(
"--batch-size",
type=int,
default=512,
help="training batch size per device (CPU/GPU)")
parser.add_argument(
"--num-epochs",
type=int,
default=120,
help="number of training epochs")
parser.add_argument(
"--start-epoch",
type=int,
default=1,
help="starting epoch for resuming, default is 1 for new training")
parser.add_argument(
"--attempt",
type=int,
default=1,
help="current number of training")
parser.add_argument(
"--optimizer-name",
type=str,
default="nag",
help="optimizer name")
parser.add_argument(
"--lr",
type=float,
default=0.1,
help="learning rate")
parser.add_argument(
"--momentum",
type=float,
default=0.9,
help="momentum value for optimizer")
parser.add_argument(
"--wd",
type=float,
default=0.0001,
help="weight decay rate")
parser.add_argument(
"--log-interval",
type=int,
default=50,
help="number of batches to wait before logging")
parser.add_argument(
"--save-interval",
type=int,
default=4,
help="saving parameters epoch interval, best model will always be saved")
parser.add_argument(
"--save-dir",
type=str,
default="",
help="directory of saved models and log-files")
parser.add_argument(
"--logging-file-name",
type=str,
default="train.log",
help="filename of training log")
parser.add_argument(
"--seed",
type=int,
default=-1,
help="Random seed to be fixed")
parser.add_argument(
"--log-packages",
type=str,
default="keras",
help="list of python packages for logging")
parser.add_argument(
"--log-pip-packages",
type=str,
default="keras, keras-mxnet, keras-applications, keras-preprocessing",
help="list of pip packages for logging")
args = parser.parse_args()
return args
def init_rand(seed):
if seed <= 0:
seed = np.random.randint(10000)
random.seed(seed)
np.random.seed(seed)
mx.random.seed(seed)
return seed
def prepare_trainer(net,
optimizer_name,
momentum,
lr,
num_gpus,
state_file_path=None):
optimizer_name = optimizer_name.lower()
if (optimizer_name == "sgd") or (optimizer_name == "nag"):
optimizer = keras.optimizers.SGD(
lr=lr,
momentum=momentum,
nesterov=(optimizer_name == "nag"))
else:
raise ValueError("Usupported optimizer: {}".format(optimizer_name))
backend_agnostic_compile(
model=net,
loss="categorical_crossentropy",
optimizer=optimizer,
metrics=[keras.metrics.categorical_accuracy, keras.metrics.top_k_categorical_accuracy],
num_gpus=num_gpus)
if (state_file_path is not None) and state_file_path and os.path.exists(state_file_path):
net = load_model(filepath=state_file_path)
return net
def train_net(net,
train_gen,
val_gen,
train_num_examples,
val_num_examples,
num_epochs,
checkpoint_filepath,
start_epoch1):
checkpointer = ModelCheckpoint(
filepath=checkpoint_filepath,
verbose=1,
save_best_only=True)
tic = time.time()
net.fit_generator(
generator=train_gen,
samples_per_epoch=train_num_examples,
epochs=num_epochs,
verbose=True,
callbacks=[checkpointer],
validation_data=val_gen,
validation_steps=val_num_examples,
class_weight=None,
max_queue_size=10,
workers=1,
use_multiprocessing=False,
shuffle=True,
initial_epoch=(start_epoch1 - 1))
logging.info("Time cost: {:.4f} sec".format(
time.time() - tic))
def main():
"""
Main body of script.
"""
args = parse_args()
args.seed = init_rand(seed=args.seed)
_, _ = initialize_logging(
logging_dir_path=args.save_dir,
logging_file_name=args.logging_file_name,
main_script_path=__file__,
script_args=args)
batch_size = prepare_ke_context(
num_gpus=args.num_gpus,
batch_size=args.batch_size)
net = prepare_model(
model_name=args.model,
use_pretrained=args.use_pretrained,
pretrained_model_file_path=args.resume.strip())
num_classes = net.classes if hasattr(net, "classes") else 1000
input_image_size = net.in_size if hasattr(net, "in_size") else (args.input_size, args.input_size)
train_data, val_data = get_data_rec(
rec_train=args.rec_train,
rec_train_idx=args.rec_train_idx,
rec_val=args.rec_val,
rec_val_idx=args.rec_val_idx,
batch_size=batch_size,
num_workers=args.num_workers,
input_image_size=input_image_size,
resize_inv_factor=args.resize_inv_factor)
train_gen = get_data_generator(
data_iterator=train_data,
num_classes=num_classes)
val_gen = get_data_generator(
data_iterator=val_data,
num_classes=num_classes)
net = prepare_trainer(
net=net,
optimizer_name=args.optimizer_name,
momentum=args.momentum,
lr=args.lr,
num_gpus=args.num_gpus,
state_file_path=args.resume_state)
train_net(
net=net,
train_gen=train_gen,
val_gen=val_gen,
train_num_examples=1281167,
val_num_examples=50048,
num_epochs=args.num_epochs,
checkpoint_filepath=os.path.join(args.save_dir, "imagenet_{}.h5".format(args.model)),
start_epoch1=args.start_epoch)
if __name__ == "__main__":
main()