-
Notifications
You must be signed in to change notification settings - Fork 8
/
train_OWG.py
392 lines (333 loc) · 20.3 KB
/
train_OWG.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
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
# ▄▄▌ ▐ ▄▌ ▄▄ •
#▪ ██· █▌▐█▐█ ▀ ▪
# ▄█▀▄ ██▪▐█▐▐▌▄█ ▀█▄
#▐█▌.▐▌▐█▌██▐█▌▐█▄▪▐█
# ▀█▄▀▪ ▀▀▀▀ ▀▪·▀▀▀▀
#
## train_OWG.py
## A script to train optical wave gauges for 4 models and 4 batch sizes
## GPU with 8+GB memory is recommended
## Written by Daniel Buscombe,
## Northern Arizona University
## daniel.buscombe.nau.edu
# import libraries
import numpy as np
import pandas as pd
import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt
import zipfile
import sys, getopt, os, gc
import time, datetime, json
from glob import glob
from sklearn.model_selection import train_test_split
os.environ['CUDA_VISIBLE_DEVICES'] = '0' ##use GPU. 0=first, 1=second, etc
from utils import *
from tensorflow.keras.applications.mobilenet import MobileNet
from tensorflow.keras.applications.inception_v3 import InceptionV3
from tensorflow.keras.applications.inception_resnet_v2 import InceptionResNetV2
## the original implementation used MobileNetV2
## however, for an unknown reason, keras implementation of MobileNetV2 in Tensorflow 2 is not giving the same result as in Tensorflow 1
## so below I have replaced MobileNetV2 with DenseNet201 which is comparable accuracy
#from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2
from tensorflow.keras.applications.densenet import DenseNet201
from tensorflow.keras.layers import GlobalAveragePooling2D, Dense, Dropout, Flatten, BatchNormalization
from tensorflow.keras.models import Sequential
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import tensorflow as tf
#==============================================================
## script starts here
if __name__ == '__main__':
# start time
print ("start time - {}".format(datetime.datetime.now().strftime("%Y-%m-%d %H:%M")))
start = time.time()
argv = sys.argv[1:]
try:
opts, args = getopt.getopt(argv,"h:c:")
except getopt.GetoptError:
print('python train_OWG.py -c configfile.json')
sys.exit(2)
for opt, arg in opts:
if opt == '-h':
print('Example usage: python3 train_OWG.py -c config_IR_H.json')
sys.exit()
elif opt in ("-c"):
configfile = arg
# load the user configs
with open(os.getcwd()+os.sep+'config'+os.sep+configfile) as f:
config = json.load(f)
# start time
print ("start time - {}".format(datetime.datetime.now().strftime("%Y-%m-%d %H:%M")))
start = time.time()
# # exploit multiple GPUs if available
# strategy = tf.distribute.MirroredStrategy()
# print ('Number of devices: {}'.format(strategy.num_replicas_in_sync))
print(config)
# config variables
imsize = int(config["img_size"])
num_epochs = int(config["num_epochs"])
test_size = float(config["test_size"])
height_shift_range = float(config["height_shift_range"])
width_shift_range = float(config["width_shift_range"])
rotation_range = float(config["rotation_range"])
samplewise_std_normalization = config["samplewise_std_normalization"]
horizontal_flip = config["horizontal_flip"]
vertical_flip = config["vertical_flip"]
samplewise_center = config["samplewise_center"]
shear_range = float(config["shear_range"])
zoom_range = float(config["zoom_range"])
dropout_rate = float(config["dropout_rate"])
epsilon = float(config["epsilon"])
min_lr = float(config["min_lr"])
factor = float(config["factor"])
input_image_format = config["input_image_format"]
input_csv_file = config["input_csv_file"]
category = config["category"]
fill_mode = config["fill_mode"]
prc_lower_withheld = config['prc_lower_withheld']
prc_upper_withheld = config['prc_upper_withheld']
base_dir = os.path.normpath(os.getcwd())
## download files and unzip
if input_csv_file=='IR-training-dataset.csv':
print('Downloading IR imagery ...')
print('... file is ~2GB - takes a while')
url = 'https://drive.google.com/file/d/1rToH8sebTCptSv8vMSo-UToSy1Nm7KgI/view?usp=sharing'
image_dir = 'IR_images'+os.sep+'data'
if not os.path.isdir(os.path.join(base_dir,image_dir)):
file_id = '1rToH8sebTCptSv8vMSo-UToSy1Nm7KgI'
destination = 'IR_images.zip'
download_file_from_google_drive(file_id, destination)
print('download complete ... unzipping')
zip_ref = zipfile.ZipFile(destination, 'r')
zip_ref.extractall(os.getcwd())
zip_ref.close()
os.remove(destination)
elif input_csv_file=='snap-training-dataset.csv':
print('Downloading nearshore imagery ...')
print('... file is ~421MB - takes a while')
url = 'https://drive.google.com/file/d/11xd3HWcfEE_yMYcSsr1StFFPSdXNAjKH/view?usp=sharing'
image_dir = 'snap_images'+os.sep+'data'
if not os.path.isdir(os.path.join(base_dir,image_dir)):
file_id = '11xd3HWcfEE_yMYcSsr1StFFPSdXNAjKH'
destination = 'snap_images.zip'
download_file_from_google_drive(file_id, destination)
print('download complete ... unzipping')
zip_ref = zipfile.ZipFile(destination, 'r')
zip_ref.extractall(os.getcwd())
zip_ref.close()
os.remove(destination)
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
print('Downloading nearshore oblique imagery ...')
print('... file is ~257MB - takes a while')
url = 'https://drive.google.com/file/d/1N2iaH7eD9msBPtqBUHJJrasInf9tICou/view?usp=sharing'
image_dir = 'snap'+os.sep+'data'
if not os.path.isdir(os.path.join(base_dir,image_dir)):
file_id = '1N2iaH7eD9msBPtqBUHJJrasInf9tICou'
destination = 'snap.zip'
download_file_from_google_drive(file_id, destination)
print('download complete ... unzipping')
zip_ref = zipfile.ZipFile(destination, 'r')
zip_ref.extractall(os.getcwd())
zip_ref.close()
os.remove(destination)
IMG_SIZE = (imsize, imsize)
# call the utils.py function get_and_tidy_df
new_df, df = get_and_tidy_df(base_dir, input_csv_file, image_dir, category)
## making subsets of data based on prc_lower_withheld and prc_upper_withheld
if (prc_lower_withheld>0) & (prc_upper_withheld>0):
up = np.percentile(new_df[category], 100-prc_upper_withheld)
low = np.percentile(new_df[category], prc_lower_withheld)
extreme_df = new_df.loc[(new_df[category] < low) | (new_df[category] > up)]
new_df = new_df.loc[(new_df[category] >= low) & (new_df[category] <= up)]
elif (prc_lower_withheld>0) & (prc_upper_withheld==0):
low = np.percentile(new_df[category], prc_lower_withheld)
extreme_df = new_df.loc[(new_df[category] < low)]
new_df = new_df.loc[(new_df[category] >= low)]
elif (prc_lower_withheld==0) & (prc_upper_withheld>0):
up = np.percentile(new_df[category], 100-prc_upper_withheld)
extreme_df = new_df.loc[(new_df[category] > up)]
new_df = new_df.loc[(new_df[category] <= up)]
print('New Data Size:', new_df.shape[0], 'Old Size:', df.shape[0])
train_df, valid_df = train_test_split(new_df,
test_size = test_size,
random_state = 2018,
stratify = new_df['category'])
print('train', train_df.shape[0], 'validation', valid_df.shape[0])
im_gen = ImageDataGenerator(samplewise_center=samplewise_center,
samplewise_std_normalization=samplewise_std_normalization,
horizontal_flip = horizontal_flip,
vertical_flip = vertical_flip,
height_shift_range = height_shift_range,
width_shift_range = width_shift_range,
rotation_range = rotation_range,
shear_range = shear_range,
fill_mode = fill_mode,
zoom_range= zoom_range)
# call the utils.py function gen_from_def
train_X, train_Y = gen_from_def(IMG_SIZE, train_df, image_dir, category, im_gen)
test_X, test_Y = gen_from_def(IMG_SIZE, valid_df, image_dir, category, im_gen)
if (prc_lower_withheld>0) | (prc_upper_withheld>0):
ex_X, ex_Y = gen_from_def(IMG_SIZE, extreme_df, image_dir, category, im_gen)
## loop through 4 different batch sizes
for batch_size in [128]: #[16,32,64,128]:
print ("[INFO] Batch size = "+str(batch_size))
## this is the original list of models
##archs = {'1':MobileNet, '2':MobileNetV2, '3':InceptionV3, '4':InceptionResNetV2}
## however, for an unknown reason, keras implementation of MobileNetV2 in Tensoeflow 2 is not giving the same result as in Tensorflow 1, so below I have replaced MobileNetV2 with DenseNet201 which is comparable accuracy
archs = {'1':MobileNet, '2':DenseNet201, '3':InceptionV3, '4':InceptionResNetV2}
counter =1
## loop through 4 different base models
for arch in archs:
print("==========================================================")
print("==========================================================")
print("==========================================================")
print(arch)
# call the utils.py function get_weights_path
weights_path = get_weights_path(input_csv_file, category, counter, imsize, batch_size, num_epochs)
model_checkpoint = ModelCheckpoint(weights_path, monitor='val_loss', verbose=1,
save_best_only=True, mode='min', save_weights_only = True)
reduceloss_plat = ReduceLROnPlateau(monitor='val_loss', factor=factor, patience=5, verbose=1, mode='auto', epsilon=epsilon, cooldown=5, min_lr=min_lr)
earlystop = EarlyStopping(monitor="val_loss", mode="min", patience=25)
callbacks_list = [model_checkpoint, earlystop, reduceloss_plat]
print ("[INFO] Training optical wave gauge")
base_model = archs[arch](input_shape = (IMG_SIZE[0], IMG_SIZE[1],1), include_top = False, weights = None)
OWG = Sequential()
OWG.add(BatchNormalization(input_shape = (IMG_SIZE[0], IMG_SIZE[1],1)))
OWG.add(base_model)
OWG.add(BatchNormalization())
OWG.add(GlobalAveragePooling2D())
OWG.add(Dropout(dropout_rate))
OWG.add(Dense(1, activation = 'linear' ))
OWG.compile(optimizer = 'rmsprop', loss = 'mse', #adam
metrics = [mae_metric])
OWG.summary()
history = OWG.fit(train_X, train_Y, batch_size=batch_size,
validation_data = (test_X, test_Y),
epochs=num_epochs, callbacks = callbacks_list)
OWG.load_weights(weights_path)
# serialize model to JSON
model_json = OWG.to_json()
with open(weights_path.replace('.hdf5','.json'), "w") as json_file:
json_file.write(model_json)
print ("[INFO] Testing optical wave gauge")
if category=='zscore':
# or if the model predicts zscores - recover value using pop. mean and standard dev.
pred_Y = div*np.squeeze(OWG.predict(test_X, batch_size = batch_size, verbose = True))+mean
if (prc_lower_withheld>0) | (prc_upper_withheld>0):
pred_extreme_Y = div*np.squeeze(OWG.predict(ex_X, batch_size = batch_size, verbose = True))+mean
else:
pred_Y = np.squeeze(OWG.predict(test_X, batch_size = batch_size, verbose = True))
if (prc_lower_withheld>0) | (prc_upper_withheld>0):
pred_extreme_Y = np.squeeze(OWG.predict(ex_X, batch_size = batch_size, verbose = True))
print ("[INFO] Creating plots ")
fig, ax1 = plt.subplots(1,1, figsize = (6,6))
ax1.plot(test_Y, pred_Y, 'b.', label = 'predictions')
ax1.plot(test_Y, test_Y, 'k-', label = 'actual')
if (prc_lower_withheld>0) | (prc_upper_withheld>0):
ax1.plot(ex_Y, pred_extreme_Y, 'r*', label = 'extreme predictions')
ax1.legend()
if category == 'H':
ax1.set_xlabel('Actual H (m)')
ax1.set_ylabel('Predicted H (m)')
if input_csv_file=='IR-training-dataset.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_IR.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='snap-training-dataset.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_nearshore.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_oblique.png', dpi=300, bbox_inches='tight')
else:
ax1.set_xlabel('Actual T (s)')
ax1.set_ylabel('Predicted T (s)')
if input_csv_file=='IR-training-dataset.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_IR.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='snap-training-dataset.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_nearshore.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
plt.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_oblique.png', dpi=300, bbox_inches='tight')
plt.close('all')
fig = plt.figure(figsize = (6,6))
plt.subplot(121)
# summarize history for accuracy
plt.plot(history.history['mae_metric'])
plt.plot(history.history['val_mae_metric'])
plt.ylabel('Mean absolute error')
plt.xlabel('Epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.subplot(122)
# summarize history for loss
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['train', 'test'], loc='upper left')
if input_csv_file=='IR-training-dataset.csv':
outfile = os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+category+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_'+category+'_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'_loss_acc_curves_IR.png'
elif input_csv_file=='snap-training-dataset.csv':
outfile = os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+category+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_'+category+'_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'_loss_acc_curves_nearshore.png'
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
outfile = os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+category+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_'+category+'_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'_loss_acc_curves_oblique.png'
plt.savefig(outfile, dpi=300, bbox_inches='tight')
plt.close('all')
print('Made '+outfile)
rand_idx = np.random.choice(range(test_X.shape[0]), 9)
fig, m_axs = plt.subplots(3, 3, figsize = (16, 32))
for (idx, c_ax) in zip(rand_idx, m_axs.flatten()):
c_ax.imshow(test_X[idx, :,:,0], cmap = 'gray')
if category == 'H':
c_ax.set_title('H: %0.3f\nPredicted H: %0.3f' % (test_Y[idx], pred_Y[idx]))
else:
c_ax.set_title('T: %0.3f\nPredicted T: %0.3f' % (test_Y[idx], pred_Y[idx]))
c_ax.axis('off')
if category == 'H':
if input_csv_file=='IR-training-dataset.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_IR.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='snap-training-dataset.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_nearshore.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'H'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveheight_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_oblique.png', dpi=300, bbox_inches='tight')
else:
if input_csv_file=='IR-training-dataset.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_IR.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='snap-training-dataset.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_nearshore.png', dpi=300, bbox_inches='tight')
elif input_csv_file=='Nearshore-Training-Oblique-cam2-snap.csv':
fig.savefig(os.getcwd()+os.sep+'im'+str(imsize)+os.sep+'res'+os.sep+str(num_epochs)+'epoch'+os.sep+'T'+os.sep+'model'+str(counter)+os.sep+'batch'+str(batch_size)+os.sep+'im'+str(IMG_SIZE[0])+'_waveperiod_predictions_model'+str(counter)+'_'+str(num_epochs)+'epoch'+str(batch_size)+'batch_oblique.png', dpi=300, bbox_inches='tight')
counter += 1
print ("[INFO] Finished training")
gc.collect()
# end time
end = time.time()
print ("end time - {}".format(datetime.datetime.now().strftime("%Y-%m-%d %H:%M")))
# # Call the distribution scope context manager
# with strategy.scope():
# from tensorflow.keras.metrics import mean_absolute_error
# # mean absolute error
# def mae_metric(in_gt, in_pred):
# return mean_absolute_error(in_gt, in_pred)
# # call the utils.py function get_weights_path
# weights_path = get_weights_path(input_csv_file, category, counter, imsize, batch_size, num_epochs)
#
# model_checkpoint = ModelCheckpoint(weights_path, monitor='val_loss', verbose=1,
# save_best_only=True, mode='min', save_weights_only = True)
# reduceloss_plat = ReduceLROnPlateau(monitor='val_loss', factor=factor, patience=5, verbose=1, mode='auto', epsilon=epsilon, cooldown=5, min_lr=min_lr)
# earlystop = EarlyStopping(monitor="val_loss", mode="min", patience=25)
# callbacks_list = [model_checkpoint, earlystop, reduceloss_plat]
#
# print ("[INFO] Training optical wave gauge")
# base_model = archs[arch](input_shape = (IMG_SIZE[0], IMG_SIZE[1],1), include_top = False, weights = None)
# OWG = Sequential()
# OWG.add(BatchNormalization(input_shape = (IMG_SIZE[0], IMG_SIZE[1],1)))
# OWG.add(base_model)
# OWG.add(BatchNormalization())
# OWG.add(GlobalAveragePooling2D())
# OWG.add(Dropout(dropout_rate))
# OWG.add(Dense(1, activation = 'linear' ))
# OWG.compile(optimizer = 'rmsprop', loss = 'mse', #adam
# metrics = [mae_metric])
# history = OWG.fit(train_X, train_Y, batch_size=batch_size,
# validation_data = (test_X, test_Y),
# epochs=num_epochs, callbacks = callbacks_list)
# with strategy.scope():
# # load the new model weights
# OWG.load_weights(weights_path)