-
Notifications
You must be signed in to change notification settings - Fork 3
/
train.py
299 lines (221 loc) · 16.4 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
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
import tables
import os
import math
from time import gmtime, strftime
from functools import partial
from shutil import rmtree
from keras.callbacks import ModelCheckpoint, CSVLogger, Callback, LearningRateScheduler
from keras.utils import plot_model
from keras.backend import clear_session
#import config_files.isles_config as isles_config
#import config_files.edema_config as edema_config
#import config_files.tumor1_config as tumor1_config
#import config_files.tumor2_config as tumor2_config
#import config_files.nonenhancing1_config as nonenhancing1_config
#import config_files.nonenhancing2_config as nonenhancing2_config
#import config_files.downsampled_edema_config as downsampled_edema_config
#import config_files.upsample_config as upsample_config
#import config_files.upsample_preloaded_config as upsample_preloaded_config
#import config_files.old_edema_config as old_edema_config
#import config_files.fms_config as fms_config
from model import n_net_3d, u_net_3d, split_u_net_3d, w_net_3d, load_old_model, vox_net, parellel_unet_3d
from load_data import DataCollection
from data_generator import get_data_generator, get_patch_data_generator
from data_utils import pickle_dump, pickle_load
from predict import model_predict_patches_hdf5
from augment import *
from file_util import split_folder
def learning_pipeline(overwrite=False, delete=False, config=None, parameters=None):
# Modifications to add in regardless of config file
####################################################
# append_prefix_to_config(config, ["hdf5_train", "hdf5_validation", "hdf5_test"], 'downsample_')
# append_prefix_to_config(config, ["model_file"], strftime("%Y-%m-%d_%H:%M:%S", gmtime()))
# config['predictions_name'] = 'perpetual_patch_32'
# config['predictions_replace_existing'] = True
# config['overwrite_trainval_data'] = True
# config['patch_shape'] = (24, 24, 4)
# config['training_patches_multiplier'] = 50
# config['validation_patches_multiplier'] = 10
# config["downsize_filters_factor"] = 1
# config["initial_learning_rate"] = 0.00001
# config["regression"] = True
# config["n_epochs"] = 200
# config["batch_size"] = 1
# config["image_shape"] = None
fill_config_keys(config=config)
update_config(config=config, parameters=parameters)
create_directories(delete=delete, config=config)
modality_dict = config['training_modality_dict']
validation_files = []
# Load training and validation data.
if config['overwrite_trainval_data'] or not os.path.exists(os.path.abspath(config["hdf5_train"])):
print 'WRITING DATA', '\n'
flip_augmentation_group = AugmentationGroup({'input_modalities': Flip_Rotate_2D(flip=True, rotate=False), 'ground_truth': Flip_Rotate_2D(flip=True, rotate=False)}, multiplier=2)
# Find Data
training_data_collection = DataCollection(config['train_dir'], modality_dict, brainmask_dir=config['brain_mask_dir'], roimask_dir=config['roi_mask_dir'], patch_shape=config['patch_shape'])
training_data_collection.fill_data_groups()
# # Training - with patch augmentation
if not config['perpetual_patches']:
training_patch_augmentation = ExtractPatches(config['patch_shape'], config['patch_extraction_conditions'])
training_patch_augmentation_group = AugmentationGroup({'input_modalities': training_patch_augmentation, 'ground_truth': training_patch_augmentation}, multiplier=config['training_patches_multiplier'])
training_data_collection.append_augmentation(training_patch_augmentation_group)
training_data_collection.append_augmentation(flip_augmentation_group)
training_data_collection.write_data_to_file(output_filepath = config['hdf5_train'], save_masks=config["overwrite_masks"])
if config['validation_dir'] is not None:
# Validation - with patch augmentation
validation_data_collection = DataCollection(config['validation_dir'], modality_dict, brainmask_dir=config['brain_mask_dir'], roimask_dir=config['roi_mask_dir'], patch_shape=config['patch_shape'])
validation_data_collection.fill_data_groups()
if not config['perpetual_patches']:
validation_patch_augmentation = ExtractPatches(config['patch_shape'], config['patch_extraction_conditions'])
validation_patch_augmentation_group = AugmentationGroup({'input_modalities': validation_patch_augmentation, 'ground_truth': validation_patch_augmentation}, multiplier=config['validation_patches_multiplier'])
validation_data_collection.append_augmentation(validation_patch_augmentation_group)
validation_data_collection.append_augmentation(flip_augmentation_group)
validation_data_collection.write_data_to_file(output_filepath = config['hdf5_validation'], save_masks=config["overwrite_masks"], store_masks=True)
# Create a new model if necessary. Preferably, load an existing one.
if not config["overwrite_model"] and os.path.exists(config["model_file"]):
model = load_old_model(config["model_file"])
else:
model = u_net_3d(input_shape=(len(modality_dict['input_modalities']),) + config['patch_shape'], output_shape=(len(modality_dict['ground_truth']),) + config['patch_shape'], downsize_filters_factor=config['downsize_filters_factor'], initial_learning_rate=config['initial_learning_rate'], regression=config['regression'], num_outputs=(len(modality_dict['ground_truth'])))
plot_model(model, to_file='model_image.png', show_shapes=True)
# Create data generators and train the model.
if config["overwrite_training"]:
# Get training and validation generators, either split randomly from the training data or from separate hdf5 files.
if config["validation_dir"] is None:
open_validation_hdf5 = []
validation_generator, num_validation_steps = None, None
elif os.path.exists(os.path.abspath(config["hdf5_validation"])):
print "Validation data found"
open_validation_hdf5 = tables.open_file(config["hdf5_validation"], "r")
if config['perpetual_patches']:
validation_generator, num_validation_steps = get_patch_data_generator(open_validation_hdf5, batch_size=config["validation_batch_size"], data_labels = ['input_modalities', 'ground_truth'], patch_multiplier=config['validation_patch_multiplier'], patch_shape=config['patch_shape'], roi_ratio=config['roi_ratio'])
else:
validation_generator, num_validation_steps = get_data_generator(open_validation_hdf5, batch_size=config["validation_batch_size"], data_labels = ['input_modalities', 'ground_truth'])
print validation_generator
print num_validation_steps
else:
print "Validation data not found! Exiting pipeline."
open_train_hdf5 = tables.open_file(config["hdf5_train"], "r")
if config['perpetual_patches']:
train_generator, num_train_steps = get_patch_data_generator(open_train_hdf5, batch_size=config["training_batch_size"], data_labels = ['input_modalities', 'ground_truth'], patch_multiplier=config['training_patch_multiplier'], patch_shape=config['patch_shape'], roi_ratio=config['roi_ratio'], num_epoch_steps=config['num_epoch_steps'], num_subepochs=config['num_subepochs'], num_subepoch_patients=config['num_subepoch_patients'])
else:
train_generator, num_train_steps = get_data_generator(open_train_hdf5, batch_size=config["training_batch_size"], data_labels = ['input_modalities', 'ground_truth'])
# Train model.. TODO account for no validation
train_model(model=model, model_file=config["model_file"], training_generator=train_generator, validation_generator=validation_generator, steps_per_epoch=num_train_steps, validation_steps=num_validation_steps, initial_learning_rate=config["initial_learning_rate"], learning_rate_drop=config["learning_rate_drop"], learning_rate_epochs=config["decay_learning_rate_every_x_epochs"], n_epochs=config["n_epochs"])
# Close training and validation files, no longer needed.
open_train_hdf5.close()
if validation_files:
open_validation_hdf5.close()
# Load testing data
if config['overwrite_test_data'] or not os.path.exists(os.path.abspath(config["hdf5_test"])):
modality_dict = config['test_modality_dict']
testing_data_collection = DataCollection(config['test_dir'], modality_dict)
testing_data_collection.fill_data_groups()
testing_data_collection.write_data_to_file(output_filepath = config['hdf5_test'], save_masks=False, store_masks=False)
# Run prediction step.
if config['overwrite_prediction']:
open_test_hdf5 = tables.open_file(config["hdf5_test"], "r")
model_predict_patches_hdf5(data_file=open_test_hdf5, input_data_label=config['predictions_input'], patch_shape=config['patch_shape'], output_directory=config['predictions_folder'], output_name=config['predictions_name'], ground_truth_data_label=config['predictions_groundtruth'], model=model, replace_existing=config['predictions_replace_existing'])
clear_session()
def create_directories(delete=False, config=None):
# Create required directories
for directory in [config['model_file'], config['hdf5_train'], config['hdf5_test'], config['hdf5_validation'], config['predictions_folder'], config['brain_mask_dir'], config['roi_mask_dir']]:
if directory is not None:
directory = os.path.abspath(directory)
if not os.path.isdir(directory):
directory = os.path.dirname(directory)
if delete:
rmtree(directory)
if not os.path.exists(directory):
os.makedirs(directory)
def update_config(config, parameters):
if parameters is None:
return
for key in parameters.keys():
config[key] = parameters[key]
def fill_config_keys(config):
if 'num_subepochs' not in config.keys():
config['num_subepochs'] = None
if 'num_epoch_steps' not in config.keys():
config['num_epoch_steps'] = 10
def append_prefix_to_config(config, keys, prefix):
for key in keys:
config[key] = '/'.join(str.split(config[key], '/')[0:-1]) + '/' + prefix + str.split(config[key], '/')[-1]
def train_model(model, model_file, training_generator, validation_generator, steps_per_epoch, validation_steps, initial_learning_rate, learning_rate_drop, learning_rate_epochs, n_epochs):
if validation_generator is None:
model.fit_generator(generator=training_generator, steps_per_epoch=steps_per_epoch, epochs=n_epochs, pickle_safe=True, callbacks=get_callbacks(model_file, initial_learning_rate=initial_learning_rate, learning_rate_drop=learning_rate_drop,learning_rate_epochs=learning_rate_epochs))
else:
model.fit_generator(generator=training_generator, steps_per_epoch=steps_per_epoch, epochs=n_epochs, validation_data=validation_generator, validation_steps=validation_steps, pickle_safe=True, callbacks=get_callbacks(model_file, initial_learning_rate=initial_learning_rate, learning_rate_drop=learning_rate_drop,learning_rate_epochs=learning_rate_epochs))
model.save(model_file)
""" The following three functions/classes are mysterious to me.
"""
def step_decay(epoch, initial_lrate, drop, epochs_drop):
return initial_lrate * math.pow(drop, math.floor((1+epoch)/float(epochs_drop)))
class SaveLossHistory(Callback):
def on_train_begin(self, logs={}):
self.losses = []
def on_batch_end(self, batch, logs={}):
self.losses.append(logs.get('loss'))
pickle_dump(self.losses, "loss_history.pkl")
def get_callbacks(model_file, initial_learning_rate, learning_rate_drop, learning_rate_epochs, logging_dir="."):
""" Currently do not understand callbacks.
"""
model_checkpoint = ModelCheckpoint(model_file, monitor="loss", save_best_only=True)
logger = CSVLogger(os.path.join(logging_dir, "training.log"))
history = SaveLossHistory()
scheduler = LearningRateScheduler(partial(step_decay, initial_lrate=initial_learning_rate, drop=learning_rate_drop, epochs_drop=learning_rate_epochs))
return [model_checkpoint, logger, history, scheduler]
def pipeline(config_name='', mode='default'):
from configs import config_map
config_type = config_map[config_name]
if config_type:
if mode == 'default':
config_mode = config_type.default_config()
elif mode == 'train':
config_mode = config_type.train_config()
elif mode == 'predict':
config_mode = config_type.predict_config()
elif mode == 'test':
config_mode = config_type.test_config()
elif mode == 'train_data':
config_mode = config_type.train_data_config()
else:
raise NotImplementedError("Config option '{}' not implemented!".format(mode))
learning_pipeline(config=config_mode, overwrite=False)
else:
pass
# ISLES CONFIGURATION OPTIONS
# learning_pipeline(config=isles_config.default_config(), overwrite=False)
# learning_pipeline(config=isles_config.train_config(), overwrite=False)
# learning_pipeline(config=isles_config.predict_config(), overwrite=False)
# BRATS CONFIGURATION OPTIONS
# learning_pipeline(config=old_edema_config.predict_config(), overwrite=False)
# learning_pipeline(config=edema_config.default_config(), overwrite=False)
# learning_pipeline(config=edema_config.train_config(), overwrite=False)
# learning_pipeline(config=edema_config.predict_config(), overwrite=False)
# learning_pipeline(config=downsampled_edema_config.default_config(), overwrite=False)
# learning_pipeline(config=fms_config.test_config(), overwrite=False)
# learning_pipeline(config=upsample_config.test_config(), overwrite=False)
# learning_pipeline(config=upsample_preloaded_config.test_config(), overwrite=False)
# learning_pipeline(config=tumor1_config.train_config(), overwrite=False)
# learning_pipeline(config=nonenhancing1_config.default_config(), overwrite=False)
# learning_pipeline(config=nonenhancing1_config.test_config(), overwrite=False)
# learning_pipeline(config=tumor2_config.train_data_config(), overwrite=False)
# learning_pipeline(config=tumor2_config.test_config(), overwrite=False)
# learning_pipeline(config=nonenhancing2_config.train_data_config(), overwrite=False)
# learning_pipeline(config=nonenhancing2_config.test_config(), overwrite=False)
# learning_pipeline(config=upsample_config.train_data_config(), overwrite=False)
# learning_pipeline(config=edema_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Train'})
# learning_pipeline(config=tumor1_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Train'})
# learning_pipeline(config=nonenhancing1_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Train'})
# learning_pipeline(config=tumor2_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Train'})
# learning_pipeline(config=nonenhancing2_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Train'})
# learning_pipeline(config=edema_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Validation'})
# learning_pipeline(config=tumor1_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Validation'})
# learning_pipeline(config=nonenhancing1_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Validation'})
# learning_pipeline(config=tumor2_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Validation'})
# learning_pipeline(config=nonenhancing2_config.test_config(), overwrite=False, parameters={"test_dir": '/mnt/jk489/sharedfolder/BRATS2017/Validation'})
# split_folder('/mnt/jk489/sharedfolder/BRATS2017/Val', .2, ['/mnt/jk489/sharedfolder/BRATS2017/Val_Train', '/mnt/jk489/sharedfolder/BRATS2017/Val_Val'])
# learning_pipeline(config=reconciliation_config.train_config(), overwrite=False)
# learning_pipeline(config=reconciliation_config.test_config(), overwrite=False)
if __name__ == '__main__':
pass