-
Notifications
You must be signed in to change notification settings - Fork 0
/
fit.py
308 lines (262 loc) · 12.9 KB
/
fit.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
import argparse
import csv
import gc
import os
import warnings
from collections import OrderedDict
from datetime import datetime
import matplotlib.pyplot as plt
import pandas as pd
import yaml
from keras.callbacks import CSVLogger, ModelCheckpoint, ReduceLROnPlateau, EarlyStopping, TensorBoard
from keras.optimizers import Adam
from config import MAX_TRAIN_STEPS, DEFAULT_LEARNING_RATE, DEFAULT_LOAD_WEIGHTS, MAX_TRAIN_EPOCHS, INPUT_SHAPE, \
DEFAULT_FREEZE_LAYERS, DEFAULT_MODEL_NAME, DEFAULT_MARGIN, BATCH_SIZE, VALID_IMG_NUM, CONFIG_HISTORY_FILE, \
DEFAULT_TRAINING_CONFIG, CHECKPOINTS_DIR, DEFAULT_HARD_SAMPLING_BATCH, DEFAULT_TIMESTAMP, DEFAULT_USE_HARD_BATCH, \
USE_SIAMESE_MODEL
from models import AVAILABLE_MODELS
from utils.data_generators import TripletDataGenerator, PairDataGenerator
from utils.losses_and_metrics import triplet_loss, triplet_acc
def warn(*args, **kwargs):
pass
warnings.warn = warn
gc.enable() # memory is tight
class TrainingConfigurator:
def __init__(self, kwargs):
self.model_name = kwargs['model_name']
self.learning_rate = kwargs['learning_rate']
self.load_weights = kwargs['load_weights'] # to be removed
self.freeze_layers = kwargs['freeze_layers']
self.margin = kwargs['margin']
self.hard_sampling_batch_size = kwargs['hard_sampling_batch_size']
self.batch_size = kwargs['batch_size']
self.number_of_validation_imgs = kwargs['number_of_validation_imgs']
self.input_shape = kwargs['input_shape']
self.model = None
self.keras_model = None
self.encoder = None
self.timestamp = kwargs['timestamp']
self.hard_batch_approach = DEFAULT_USE_HARD_BATCH
self.siamese_model = USE_SIAMESE_MODEL
if self.timestamp is None:
self._prepare_directory()
self._load_models()
self._prepare_models()
def _prepare_directory(self):
self.timestamp = datetime.utcnow().strftime('%Y%m%d%H%M%S')
os.mkdir(os.path.join(CHECKPOINTS_DIR, self.timestamp))
def _load_models(self):
"""
Prepare instance of SiameseNetwork and Keras models: full siamese and encoders
"""
ModelClass = AVAILABLE_MODELS.get(self.model_name)
self.model = ModelClass(self)
self.keras_model = self.model.get_model()
self.encoder = self.keras_model.get_layer('encoder')
def _prepare_models(self):
"""
Prepare Keras models basing on training configuration (Freeze layers, load weights)
"""
if self.freeze_layers is not None:
self._set_freeze_layers()
self._load_weight_if_possible()
print(self.keras_model.summary())
self.show_configuration()
def _load_weight_if_possible(self):
"""
Load weights for Keras Siamese model if they exist
"""
try:
self.keras_model.load_weights(self.model.WEIGHT_PATH)
print('Weights loaded!')
except OSError:
print('No file with weights available! Starting from scratch...')
def _set_freeze_layers(self):
"""
Make some number of first few layers non-trainable. Number is defined by self.freeze_layers parameter.
"""
for layer in self.encoder.layers[:self.freeze_layers]:
layer.trainable = False
def show_configuration(self):
"""
Show configuration
"""
keys = self.get_configuration_parameters_names()
data = self.get_configuration_parameters_values()
print('\nTRAINING CONFIGURATION:\n')
for pos, param in enumerate(keys):
print('{}: {}'.format(param, data[pos]))
print('\n')
def show_loss(self, acc=True):
"""
Print loss history
:param acc: optionally print accuracy history
"""
loss_history = pd.read_csv(self.model.FIT_HISTORY_PATH, sep=';')
epochs = loss_history.epoch
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(22, 10))
_ = ax1.plot(epochs, loss_history.loss, 'b-',
epochs, loss_history.val_loss, 'r-')
ax1.legend(['Training', 'Validation'])
ax2.set_xlabel('epochs')
ax2.set_ylabel('Loss')
ax1.set_title('Loss')
if acc:
_ = ax2.plot(epochs, loss_history.acc, 'b-',
epochs, loss_history.val_acc, 'r-')
ax2.legend(['Training', 'Validation'])
ax2.set_xlabel('epochs')
ax2.set_ylabel('Acc')
ax2.set_title('Acc')
def _get_callbacks(self):
"""
Prepare Keras callback for training
:return: List of Keras callbacks
"""
csv_logger = CSVLogger(self.model.FIT_HISTORY_PATH, append=False, separator=';')
checkpoint = ModelCheckpoint(self.model.WEIGHT_PATH,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
save_weights_only=True)
reduceLROnPlat = ReduceLROnPlateau(monitor='val_loss',
factor=0.2,
patience=1,
verbose=1,
mode='min',
min_delta=0.0001,
cooldown=0,
min_lr=1e-10)
early = EarlyStopping(monitor="val_loss", mode="min", verbose=2, patience=5, min_delta=0.0001)
tb = TensorBoard(log_dir="./Graph", write_grads=True,
histogram_freq=1, write_images=True)
return [checkpoint, early, reduceLROnPlat, csv_logger]
def _fit_siamese(self):
self.keras_model.compile(optimizer=Adam(self.learning_rate, decay=0.000001),
loss='binary_crossentropy',
metrics=['binary_crossentropy', 'acc'])
data_generator = PairDataGenerator(self)
(valid_A_x, valid_B_x), valid_y = data_generator.get_validation_dataset()
aug_gen = data_generator.get_training_gen()
callbacks_list = self._get_callbacks()
loss_hist = [self.keras_model.fit_generator(aug_gen,
steps_per_epoch=MAX_TRAIN_STEPS,
epochs=MAX_TRAIN_EPOCHS,
validation_data=([valid_A_x, valid_B_x], valid_y),
callbacks=callbacks_list,
workers=1,
max_queue_size=1)]
return loss_hist
def _fit_triplet_loss(self):
"""
Start training with triplet loss
:return: Keras history object
"""
self.keras_model.compile(optimizer=Adam(self.learning_rate, decay=0.000001),
loss=triplet_loss(margin=self.margin),
metrics=[triplet_acc(margin=0), triplet_acc(margin=self.margin)])
data_generator = TripletDataGenerator(self)
aug_gen = data_generator.get_training_gen()
validation_images, y = data_generator.get_hard_preprocessed_validation_dataset()
callbacks_list = self._get_callbacks()
loss_hist = [self.keras_model.fit_generator(aug_gen,
steps_per_epoch=MAX_TRAIN_STEPS,
epochs=MAX_TRAIN_EPOCHS,
validation_data=[validation_images, y],
callbacks=callbacks_list,
workers=0,
max_queue_size=1)]
return loss_hist
def get_configuration_parameters_values(self):
"""
Get all configuration parameters
:return: Python Tuple with all configuration parameters in a proper order
"""
return (self.timestamp, self.model_name, self.model.WEIGHT_PATH, self.model.FIT_HISTORY_PATH,
self.learning_rate, self.load_weights, self.freeze_layers, self.margin,
self.hard_sampling_batch_size, self.batch_size, self.number_of_validation_imgs,
self.input_shape)
def get_configuration_parameters_names(self):
"""
Get all configuration parameters names
:return: Python Tuple with all configuration parameters names in a proper order
"""
return (
'timestamp', 'model_name', 'weight_path', 'fit_history_path', 'learning_rate', 'load_weights',
'freeze_layers', 'margin', 'hard_sampling_batch_size', 'batch_size',
'number_of_validation_imgs', 'input_shape')
def get_configuration_data(self):
"""
Get OrderedDict with all parameters in a proper order
:return: Python OrderedDict with all parameters in a proper order
"""
keys = self.get_configuration_parameters_names()
data = self.get_configuration_parameters_values()
ordered_data = OrderedDict()
for pos, key in enumerate(keys):
ordered_data[key] = data[pos]
return ordered_data
def _save_configuration_to_csv(self):
"""
Save configuration to CONFIG_HISTORY_FILE CSV file
"""
if not os.path.exists(CONFIG_HISTORY_FILE):
with open(CONFIG_HISTORY_FILE, "w") as f:
writer = csv.writer(f)
titles = self.get_configuration_parameters_names()
writer.writerow(titles)
with open(CONFIG_HISTORY_FILE, "a") as f:
writer = csv.writer(f)
data_row = self.get_configuration_parameters_values()
writer.writerow(data_row)
def _save_configuration_to_yml(self):
"""
Save configuration to "config_" + self.model.timestamp + ".yml" file
"""
data = self.get_configuration_data()
timestamp = self.model.timestamp
with open(os.path.join(CHECKPOINTS_DIR, timestamp, 'config_{}.yml'.format(timestamp)), 'w') as outfile:
yaml.dump(dict(data), outfile, default_flow_style=False)
def train(self):
"""
Start training
:return: Keras history object
"""
if self.siamese_model:
loss_hist = self._fit_triplet_loss()
else:
foss_hist = self._fit_siemese()
print('Saving configuration to yml and csv files')
self._save_configuration_to_yml()
self._save_configuration_to_csv()
return loss_hist
if __name__ == '__main__':
ap = argparse.ArgumentParser()
ap.add_argument('-mn', '--model_name', default=DEFAULT_MODEL_NAME, help="Specify name of the model to use")
ap.add_argument('-lr', '--learning_rate', type=float, default=DEFAULT_LEARNING_RATE)
ap.add_argument('-lw', '--load_weights', action='store_true', default=DEFAULT_LOAD_WEIGHTS,
help='Specify if try load weights before training. Needs to have --timestamp parameter specified.')
ap.add_argument('-tc', '--training_config', type=str, default=DEFAULT_TRAINING_CONFIG,
help='Config file to load, by default configuration is taken from config.py')
ap.add_argument('-fr', '--freeze_layers', type=int, default=DEFAULT_FREEZE_LAYERS,
help="Specify number of layers that should be non-trainable")
ap.add_argument('-bs', '--batch_size', type=int, default=BATCH_SIZE)
ap.add_argument('-vi', '--number_of_validation_imgs', type=int, default=VALID_IMG_NUM,
help="Specify size of validation dataset. By default it uses 5000 samples.")
ap.add_argument('-is', '--input_shape', nargs='+', type=int, default=INPUT_SHAPE,
help="Specify images input shape to Neural Network. i.e.: -is 224 224 3")
ap.add_argument('-mr', '--margin', type=float, default=DEFAULT_MARGIN,
help='Value of margin if triplet loss used')
ap.add_argument('-hb', '--hard_sampling_batch_size', type=int, default=DEFAULT_HARD_SAMPLING_BATCH,
help='If triplet loss and hard sampling is used then specify among how many samples'
'we should search for a hard one during training step')
ap.add_argument('-ts', '--timestamp', type=str, default=DEFAULT_TIMESTAMP,
help='Specify if you want to use one of old timestamp directories, i.e. for evaluation purposes.')
kwargs = vars(ap.parse_args())
if kwargs['training_config'] is not None:
with open(kwargs['training_config'], 'r') as cfg_file:
kwargs = yaml.load(cfg_file)
train = TrainingConfigurator(kwargs)
loss_history = train.train()
gc.collect()