forked from smitharauco/GeoFacies_DL
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_geofacies.py
executable file
·161 lines (143 loc) · 8.56 KB
/
train_geofacies.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
import argparse
import numpy as np
from pathlib import Path
from Model.DCVAE import DCVAE, DCVAE_Style
from Model.GeoGans import CycleGAN_MPS, GAN2D_MPS, AlphaGAN_MPS, WGAN2D_MPS
from Model.Utils import MPS_Generator
from keras.optimizers import RMSprop, Adam
from sklearn.model_selection import train_test_split
def get_args():
parser = argparse.ArgumentParser(description="train Geofacies Class",
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("--train_dataset_path", type=str, required=True,
help="train dataset path (tfrecorfs)")
parser.add_argument("--test_dataset_path", type=str, default=None,
help="test dataset path (tfrecorfs)")
parser.add_argument("--filters", type=str, default='32-32-32',
help="Filters number")
parser.add_argument("--kernel_dim", type=str, default='3-3-3',
help="Dimension of the Kernel")
parser.add_argument("--strides_values", type=str, default='2-2-2',
help="Strides values")
parser.add_argument("--hidden_dim", type=int, default=1024,
help="Dimension of the hidden layer")
parser.add_argument("--latent_dim", type=int, default=500,
help="Dimension of the latent vector")
parser.add_argument("--batch_size", type=int, default=32,
help="batch size")
parser.add_argument("--nb_epochs", type=int, default=500,
help="number of epochs")
parser.add_argument("--lr", type=float, default=0.001,
help="learning rate")
parser.add_argument("--dropout", type=float, default=0.1,
help="dropout rate")
parser.add_argument("--steps", type=int, default=500,
help="steps per epoch")
parser.add_argument("--save_path_weights", type=str, default=None,
help="path to save the weights")
parser.add_argument("--weight", type=str, default=None,
help="weight file for restart")
parser.add_argument("--output_path", type=str, default="checkpoints",
help="checkpoint dir")
parser.add_argument("--activation", type=str, default="sigmoid",
help="activation in the last layer")
parser.add_argument("--optimizer", type=str, default="RMSprop",
help="optimizer ('RMSprop','Adam' or other)")
parser.add_argument("--model", type=str, default="cvae",
help="model architecture ('cvae','cvae-style','0AlphaGAN','CycleGAN','GAN2D_AE' and 'WGAN2D_AE')")
parser.add_argument("--kl_weight", type=float, default=2.0,
help="weight for the KL loss")
parser.add_argument("--style_weight", type=float, default=3.125e-05,
help="weight for the style loss")
parser.add_argument("--patience", type=int, default=20,
help="step patience to stop train")
parser.add_argument("--epsilon", type=float, default=0.3,
help="epsilon to compute PCA")
parser.add_argument("--Nr", type=int, default=5000,
help="Samples number to compute PCA")
parser.add_argument("--Nt", type=int, default=5000,
help="Number de samples generate by PCA model")
parser.add_argument("--alpha", type=int, default=10,
help="Hiperparameter of AlphaGans and CycleGans networks")
parser.add_argument("--clip", type=float, default=0.05,
help="clip value for WGans-AE networks")
args = parser.parse_args()
return args
def load_data_set(path_tfRecord, isArray=False, batch=4, isTanh=False):
gen_train = MPS_Generator(path_tfRecord, batch)
if isArray:
gen_train = MPS_Generator(path_tfRecord, gen_train.num)
x_train = gen_train.get_numpy_batch().astype('float32')
if isTanh:
x_train = x_train*2-1
else:
x_train = gen_train.mps_generator()
return x_train, gen_train.num, gen_train.image_dim
def load_data_by_class(args,path):
if path is None:
return None,None,None
if args.model == "cvae":
x_train,nt,image_dim = load_data_set(path,
batch=args.batch_size)
elif args.model == "cvae-style" :
x_train,nt,image_dim = load_data_set(path,isArray=True)
x_train = np.expand_dims(np.argmax(x_train,axis=-1),axis=-1)
x_train = x_train*2-1
elif args.model == 'AlphaGAN' or args.model == "CycleGAN" or args.model == "GAN2D_AE" or args.model == 'WGAN2D_AE':
x_train,nt,image_dim= load_data_set(path,
isArray=True,isTanh=True)
else:
print("Don't load dataSet")
return x_train,nt,image_dim
def main():
args = get_args()
kernel = [int(i) for i in args.kernel_dim.split('-')]
strides = [int(i) for i in args.strides_values.split('-')]
filters = [int(i) for i in args.filters.split('-')]
x_train,nt,image_dim = load_data_by_class(args,args.train_dataset_path)
x_val,vs,_ = load_data_by_class(args,args.test_dataset_path)
if args.optimizer == 'RMSprop':
opt = RMSprop(lr=args.lr)
if args.optimizer == 'Adam':
opt = Adam(lr=args.lr)
if args.model == 'AlphaGAN':
model = AlphaGAN_MPS(input_shape = image_dim,
d_filters=filters[0], g_filters=filters[0], e_filters=filters[0], c_filters=3500,alpha=args.alpha,
d_ksize=kernel[0], g_ksize=kernel[0], e_ksize=kernel[0],
z_size=args.latent_dim, batch_size=args.batch_size,
saving_path = args.save_path_weights, name='geo_AlphaGAN_', summary=True)
model.train(x_train, data_val_=x_val, epochs=args.nb_epochs, patience=args.patience, plots=False,reset_model=True)
if args.model == 'WGAN2D_AE':
model = WGAN2D_MPS(input_shape = image_dim,
d_filters=filters[0], g_filters=filters[0], e_filters=filters[0],clip_value=args.clip,
d_ksize=kernel[0], g_ksize=kernel[0], e_ksize=kernel[0],
z_size=args.latent_dim, batch_size=args.batch_size,
saving_path = args.save_path_weights, name='Wgeo_', summary=True)
model.train(x_train, data_val=x_val,epochs=args.nb_epochs, patience=args.patience,
plots=False,reset_model=True)
if args.model == 'GAN2D_AE' :
model = GAN2D_MPS(input_shape = image_dim,
d_filters=filters[0], g_filters=filters[0], e_filters=filters[0],
d_ksize=kernel[0], g_ksize=kernel[0], e_ksize=kernel[0], z_size=args.latent_dim, batch_size=args.batch_size,
saving_path = args.save_path_weights, name='geo_', summary=True)
model.train(x_train,data_val=x_val,epochs=args.nb_epochs, patience=args.patience, plots=False,reset_model=True)
if args.model == 'CycleGAN':
model = CycleGAN_MPS(batch_size=args.batch_size, saving_path = args.save_path_weights, input_shape = image_dim,
filters=filters[0], epsilon= args.epsilon, Nr=args.Nr, Nt=args.Nt,alpha=args.alpha,
model_file=None, name='geo_CycleGAN_', summary=True)
model.train(x_train, epochs=args.nb_epochs, patience=args.patience, plots=False,reset_model=True)
if args.model == 'cvae-style':
model = DCVAE_Style(input_shape=x_train.shape[1:],filters=filters,strides=strides,KernelDim=kernel,
style_weight=args.style_weight,kl_weight=args.kl_weight,act='tanh',
hidden_dim=args.hidden_dim,latent_dim=args.latent_dim,isTerminal=True,opt=opt,dropout=args.dropout, filepath = args.save_path_weights)
model.fit(x_train,x_v=x_val,num_epochs=args.nb_epochs, verbose=1,batch_size = args.batch_size)
if args.model == 'cvae':
model = DCVAE(input_shape=image_dim,filters=filters,strides=strides,KernelDim=kernel,
hidden_dim=args.hidden_dim,latent_dim=args.latent_dim,isTerminal=True,opt=opt,dropout=args.dropout, filepath = args.save_path_weights)
model.fit_generator(x_train,
num_epochs=args.nb_epochs, verbose=1,
steps_per_epoch = nt//args.batch_size,
val_set = x_val,
validation_steps = vs//args.batch_size)
if __name__ == '__main__':
main()