-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
579 lines (461 loc) · 25.2 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
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
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
import numpy as np
import sys
import os
import glob
import gc
import matplotlib
import matplotlib.pyplot as plt
matplotlib.rcParams['agg.path.chunksize'] = 100000
matplotlib.use("Agg")
from generator import Generator
from discriminator import Discriminator
from voice_analy import Voice_Analy
from complex import to_polar, to_rect
from keras.optimizers import Adam, SGD
from keras.models import Sequential, model_from_json, Model
from keras.layers import Input
from distutils.dir_util import copy_tree
""" GPUがあれば、GPUメモリの使用量を指定 """
import tensorflow as tf
from tensorflow.python.client import device_lib
from keras.backend import tensorflow_backend
d = str(device_lib.list_local_devices())
if "GPU" in d:
# config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True)) # リアルタイムで必要な分だけ確保
config = tf.ConfigProto(gpu_options=tf.GPUOptions(per_process_gpu_memory_fraction=0.6)) # 空きメモリの何割を使うかを指定
# config = tf.ConfigProto(device_count={"GPU": 0}) # GPUを使わない(GPUメモリがどうしても足らないとき)
session = tf.Session(config=config)
tensorflow_backend.set_session(session)
class Train:
def __init__(self):
self.save_path = "save/"
os.makedirs(self.save_path, exist_ok=True)
self.data_path = "materials/"
self.res_path = "result/"
os.makedirs(self.res_path + "data_bA", exist_ok=True)
os.makedirs(self.res_path + "data_aB", exist_ok=True)
os.makedirs(self.res_path + "pic/data_bA", exist_ok=True)
os.makedirs(self.res_path + "pic/data_aB", exist_ok=True)
self.fft_len = 1024 // 2
self.window = np.hamming(self.fft_len)
self.step = self.fft_len // 4
self.input_shape = (self.fft_len,)
self.g_optimizer = Adam(0.0002, 0.5)
self.d_optimizer = SGD()
if os.path.exists(self.save_path + "log.txt"):
""" [Z] """
self.g_aB_z = model_from_json(open(self.save_path + "g_ab_z.json", "r").read())
self.g_aB_z.load_weights(self.save_path + "g_ab_w_z.h5")
self.d_B_z = model_from_json(open(self.save_path + "d_b_z.json", "r").read())
self.d_B_z.load_weights(self.save_path + "d_b_w_z.h5")
self.d_B_z.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
self.g_bA_z = model_from_json(open(self.save_path + "g_ba_z.json", "r").read())
self.g_bA_z.load_weights(self.save_path + "g_ba_w_z.h5")
self.d_A_z = model_from_json(open(self.save_path + "d_a_z.json", "r").read())
self.d_A_z.load_weights(self.save_path + "d_a_w_z.h5")
self.d_A_z.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
input_a = Input(shape=self.input_shape)
input_b = Input(shape=self.input_shape)
# create fake data
fake_a = self.g_bA_z(input_b) # genuine_B -> fake_A
fake_b = self.g_aB_z(input_a) # genuine_A -> fake_B
# reconstruct
recon_b = self.g_aB_z(fake_a) # (genuine_B -> ) fake_A -> genuine_B
recon_a = self.g_bA_z(fake_b) # (genuine_A -> ) fake_B -> genuine_A
self.g_aBA_z = Model(inputs=input_a, outputs=recon_a)
self.g_bAB_z = Model(inputs=input_b, outputs=recon_b)
# not convert
nc_a = self.g_bA_z(input_a) # genuine_A -> genuine_A (gen: B -> A)
nc_b = self.g_aB_z(input_b) # genuine_B -> genuine_B (gen: A -> B)
self.nc_bA_z = Model(inputs=input_a, outputs=nc_a)
self.nc_aB_z = Model(inputs=input_b, outputs=nc_b)
self.d_A_z.trainable = False
self.d_B_z.trainable = False
# deceive disc
deceive_A = self.d_A_z(fake_a)
deceive_B = self.d_B_z(fake_b)
self.c_aB_z = Model(inputs=input_a, outputs=deceive_B) # genuine_B -> fake_A -> Genuine(expected value)
self.c_bA_z = Model(inputs=input_b, outputs=deceive_A) # genuine_A -> fake_B -> Genuine(expected value)
self.c_aB_z.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.c_bA_z.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.g_aBA_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.g_bAB_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_bA_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_aB_z.compile(loss="mae",
optimizer=self.g_optimizer)
# =======================================================================
""" [theta] """
self.g_aB_t = model_from_json(open(self.save_path + "g_ab_t.json", "r").read())
self.g_aB_t.load_weights(self.save_path + "g_ab_w_t.h5")
self.d_B_t = model_from_json(open(self.save_path + "d_b_t.json", "r").read())
self.d_B_t.load_weights(self.save_path + "d_b_w_t.h5")
self.d_B_t.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
self.g_bA_t = model_from_json(open(self.save_path + "g_ba_t.json", "r").read())
self.g_bA_t.load_weights(self.save_path + "g_ba_w_t.h5")
self.d_A_t = model_from_json(open(self.save_path + "d_a_t.json", "r").read())
self.d_A_t.load_weights(self.save_path + "d_a_w_t.h5")
self.d_A_t.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
input_a = Input(shape=self.input_shape)
input_b = Input(shape=self.input_shape)
# create fake data
fake_a = self.g_bA_t(input_b) # genuine_B -> fake_A
fake_b = self.g_aB_t(input_a) # genuine_A -> fake_B
# reconstruct
recon_b = self.g_aB_t(fake_a) # (genuine_B -> ) fake_A -> genuine_B
recon_a = self.g_bA_t(fake_b) # (genuine_A -> ) fake_B -> genuine_A
self.g_aBA_t = Model(inputs=input_a, outputs=recon_a)
self.g_bAB_t = Model(inputs=input_b, outputs=recon_b)
# not convert
nc_a = self.g_bA_t(input_a) # genuine_A -> genuine_A (gen: B -> A)
nc_b = self.g_aB_t(input_b) # genuine_B -> genuine_B (gen: A -> B)
self.nc_bA_t = Model(inputs=input_a, outputs=nc_a)
self.nc_aB_t = Model(inputs=input_b, outputs=nc_b)
self.d_A_t.trainable = False
self.d_B_t.trainable = False
# deceive disc
deceive_A = self.d_A_t(fake_a)
deceive_B = self.d_B_t(fake_b)
self.c_aB_t = Model(inputs=input_a, outputs=deceive_B) # genuine_B -> fake_A -> Genuine(expected value)
self.c_bA_t = Model(inputs=input_b, outputs=deceive_A) # genuine_A -> fake_B -> Genuine(expected value)
self.c_aB_t.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.c_bA_t.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.g_aBA_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.g_bAB_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_bA_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_aB_t.compile(loss="mae",
optimizer=self.g_optimizer)
# open(self.save_path + "log.txt", "w").write("") # log.txtをクリア
else:
self.gen = Generator(input_shape=self.input_shape) # 1 + self.fft_len // 2 + 1
self.disc = Discriminator(input_shape=self.input_shape) # 1 + self.fft_len // 2 + 1
""" [Z] """
# genuine_A -> fake_B
self.g_aB_z = self.gen.build_generator_z(filters=16)
# B -> genuine or fake
self.d_B_z = self.disc.build_discriminator(filters=16)
self.d_B_z.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
# genuine_B -> fake_A
self.g_bA_z = self.gen.build_generator_z(filters=16)
# A -> genuine or fake
self.d_A_z = self.disc.build_discriminator(filters=16)
self.d_A_z.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
input_a_z = Input(shape=self.input_shape)
input_b_z = Input(shape=self.input_shape)
# create fake data
fake_a_z = self.g_bA_z(input_b_z) # genuine_B -> fake_A
fake_b_z = self.g_aB_z(input_a_z) # genuine_A -> fake_B
# reconstruct
recon_b_z = self.g_aB_z(fake_a_z) # (genuine_B -> ) fake_A -> genuine_B
recon_a_z = self.g_bA_z(fake_b_z) # (genuine_A -> ) fake_B -> genuine_A
self.g_aBA_z = Model(inputs=input_a_z, outputs=recon_a_z)
self.g_bAB_z = Model(inputs=input_b_z, outputs=recon_b_z)
# not convert
nc_a_z = self.g_bA_z(input_a_z) # genuine_A -> genuine_A (gen: B -> A)
nc_b_z = self.g_aB_z(input_b_z) # genuine_B -> genuine_B (gen: A -> B)
self.nc_bA_z = Model(inputs=input_a_z, outputs=nc_a_z)
self.nc_aB_z = Model(inputs=input_b_z, outputs=nc_b_z)
self.d_A_z.trainable = False
self.d_B_z.trainable = False
# deceive disc
deceive_A_z = self.d_A_z(fake_a_z)
deceive_B_z = self.d_B_z(fake_b_z)
self.c_aB_z = Model(inputs=input_a_z, outputs=deceive_B_z) # genuine_B -> fake_A -> Genuine(expected value)
self.c_bA_z = Model(inputs=input_b_z, outputs=deceive_A_z) # genuine_A -> fake_B -> Genuine(expected value)
self.c_aB_z.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.c_bA_z.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.g_aBA_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.g_bAB_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_bA_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_aB_z.compile(loss="mae",
optimizer=self.g_optimizer)
self.model_save(self.g_aB_z, "g_ab_z.json")
self.model_save(self.d_B_z, "d_b_z.json")
self.model_save(self.g_bA_z, "g_ba_z.json")
self.model_save(self.d_A_z, "d_a_z.json")
# =========================================================
""" [theta] """
# genuine_A -> fake_B
self.g_aB_t = self.gen.build_generator_t(filters=16)
# B -> genuine or fake
self.d_B_t = self.disc.build_discriminator(filters=16)
self.d_B_t.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
# genuine_B -> fake_A
self.g_bA_t = self.gen.build_generator_t(filters=16)
# A -> genuine or fake
self.d_A_t = self.disc.build_discriminator(filters=16)
self.d_A_t.compile(loss="mse",
optimizer=self.d_optimizer,
metrics=["accuracy"])
input_a_t = Input(shape=self.input_shape)
input_b_t = Input(shape=self.input_shape)
# create fake data
fake_a_t = self.g_bA_t(input_b_t) # genuine_B -> fake_A
fake_b_t = self.g_aB_t(input_a_t) # genuine_A -> fake_B
# reconstruct
recon_b_t = self.g_aB_t(fake_a_t) # (genuine_B -> ) fake_A -> genuine_B
recon_a_t = self.g_bA_t(fake_b_t) # (genuine_A -> ) fake_B -> genuine_A
self.g_aBA_t = Model(inputs=input_a_t, outputs=recon_a_t)
self.g_bAB_t = Model(inputs=input_b_t, outputs=recon_b_t)
# not convert
nc_a_t = self.g_bA_t(input_a_t) # genuine_A -> genuine_A (gen: B -> A)
nc_b_t = self.g_aB_t(input_b_t) # genuine_B -> genuine_B (gen: A -> B)
self.nc_bA_t = Model(inputs=input_a_t, outputs=nc_a_t)
self.nc_aB_t = Model(inputs=input_b_t, outputs=nc_b_t)
self.d_A_t.trainable = False
self.d_B_t.trainable = False
# deceive disc
deceive_A_t = self.d_A_t(fake_a_t)
deceive_B_t = self.d_B_t(fake_b_t)
self.c_aB_t = Model(inputs=input_a_t, outputs=deceive_B_t) # genuine_B -> fake_A -> Genuine(expected value)
self.c_bA_t = Model(inputs=input_b_t, outputs=deceive_A_t) # genuine_A -> fake_B -> Genuine(expected value)
self.c_aB_t.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.c_bA_t.compile(loss="mse",
optimizer=self.g_optimizer,
metrics=["accuracy"])
self.g_aBA_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.g_bAB_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_bA_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.nc_aB_t.compile(loss="mae",
optimizer=self.g_optimizer)
self.model_save(self.g_aB_t, "g_ab_t.json")
self.model_save(self.d_B_t, "d_b_t.json")
self.model_save(self.g_bA_t, "g_ba_t.json")
self.model_save(self.d_A_t, "d_a_t.json")
def model_save(self, model, name):
j = model.to_json()
open(self.save_path + name, "w").write(j)
def save_weight(self):
""" [Z] """
self.g_aB_z.save_weights(self.save_path + "g_ab_w_z.h5")
self.d_B_z.save_weights(self.save_path + "d_b_w_z.h5")
self.g_bA_z.save_weights(self.save_path + "g_ba_w_z.h5")
self.d_A_z.save_weights(self.save_path + "d_a_w_z.h5")
""" [theta] """
self.g_aB_t.save_weights(self.save_path + "g_ab_w_t.h5")
self.d_B_t.save_weights(self.save_path + "d_b_w_t.h5")
self.g_bA_t.save_weights(self.save_path + "g_ba_w_t.h5")
self.d_A_t.save_weights(self.save_path + "d_a_w_t.h5")
def train(self, epochs=1000, batch_size=20, save_interval=5):
va = Voice_Analy()
wave_files_a = glob.glob(self.data_path + "data_A/*.wav")
wave_files_b = glob.glob(self.data_path + "data_B/*.wav")
real_label = np.ones((batch_size, 1))
fake_label = np.zeros((batch_size, 1))
for epoch in range(epochs):
for (file_a, file_b) in zip(wave_files_a, wave_files_b):
fs_a, data_a = va.get_data_from_wave(file_a)
data_a = np.array(data_a, dtype=np.float32).copy()
data_a /= 32768.0
stft_a = va.stft(data_a, self.window, self.step)
stft_a_z, stft_a_theta = to_polar(stft_a)
stft_a_z /= 256.0
stft_a_theta /= np.pi
fs_b, data_b = va.get_data_from_wave(file_b)
data_b = np.array(data_b, dtype=np.float32).copy()
data_b /= 32768.0
stft_b = va.stft(data_b, self.window, self.step)
stft_b_z, stft_b_theta = to_polar(stft_b)
stft_b_z /= 256.0
stft_b_theta /= np.pi
num_batches = 50
print("A:", os.path.basename(file_a))
print("B:", os.path.basename(file_b))
for batch in range(num_batches):
try:
idx_a = np.random.randint(0, stft_a.shape[0], batch_size)
input_a_z = stft_a_z[idx_a]
input_a_t = stft_a_theta[idx_a]
idx_b = np.random.randint(0, stft_b.shape[0], batch_size)
input_b_z = stft_b_z[idx_b]
input_b_t = stft_b_theta[idx_b]
# ======================================================
""" [Z] """
""" B -> A """
fake_a_z = self.g_bA_z.predict(input_b_z)
d_loss_real_ba = self.d_A_z.train_on_batch(input_a_z, real_label)
d_loss_fake_ba = self.d_A_z.train_on_batch(fake_a_z, fake_label)
d_loss_ba_z = 0.5 * np.add(d_loss_real_ba, d_loss_fake_ba)
c_loss_ba_z = self.c_bA_z.train_on_batch(input_b_z, real_label)
r_loss_ba_z = self.g_bAB_z.train_on_batch(input_b_z, input_b_z)
nc_loss_ba_z = self.nc_bA_z.train_on_batch(input_a_z, input_a_z)
""" A -> B """
fake_b_z = self.g_aB_z.predict(input_a_z)
d_loss_real_ab = self.d_B_z.train_on_batch(input_b_z, real_label)
d_loss_fake_ab = self.d_B_z.train_on_batch(fake_b_z, fake_label)
d_loss_ab_z = 0.5 * np.add(d_loss_real_ab, d_loss_fake_ab)
c_loss_ab_z = self.c_aB_z.train_on_batch(input_a_z, real_label)
r_loss_ab_z = self.g_aBA_z.train_on_batch(input_a_z, input_a_z)
nc_loss_ab_z = self.nc_aB_z.train_on_batch(input_b_z, input_b_z)
# ======================================================
""" [theta] """
""" B -> A """
fake_a_t = self.g_bA_t.predict(input_b_t)
d_loss_real_ba = self.d_A_t.train_on_batch(input_a_t, real_label)
d_loss_fake_ba = self.d_A_t.train_on_batch(fake_a_t, fake_label)
d_loss_ba_t = 0.5 * np.add(d_loss_real_ba, d_loss_fake_ba)
c_loss_ba_t = self.c_bA_t.train_on_batch(input_b_t, real_label)
r_loss_ba_t = self.g_bAB_t.train_on_batch(input_b_t, input_b_t)
nc_loss_ba_t = self.nc_bA_t.train_on_batch(input_a_t, input_a_t)
""" A -> B """
fake_b_t = self.g_aB_t.predict(input_a_t)
d_loss_real_ab = self.d_B_t.train_on_batch(input_b_t, real_label)
d_loss_fake_ab = self.d_B_t.train_on_batch(fake_b_t, fake_label)
d_loss_ab_t = 0.5 * np.add(d_loss_real_ab, d_loss_fake_ab)
c_loss_ab_t = self.c_aB_t.train_on_batch(input_a_t, real_label)
r_loss_ab_t = self.g_aBA_t.train_on_batch(input_a_t, input_a_t)
nc_loss_ab_t = self.nc_aB_t.train_on_batch(input_b_t, input_b_t)
# ==============================================================
if batch % save_interval == 0:
""" B -> A """
z = self.g_bA_z.predict(stft_b_z) * 256
t = self.g_bA_t.predict(stft_b_theta) * np.pi
data_bA = to_rect(z, t, stft_b.shape)
istft = va.istft(data_bA, self.window, self.step)
res = (istft * 32768.0).astype(np.int16)
p = self.res_path + "data_bA/" + os.path.basename(file_b)
p = p[:-4] + "_%d.wav" % epoch
va.save_wave(p, fs_a, res)
plt.figure()
plt.subplot(211)
plt.plot(data_b)
plt.subplot(212)
plt.plot(istft)
plt.ylim(-1, 1)
plt.savefig((self.res_path + "pic/data_bA/{}").format(
os.path.basename(file_a)[:-4] + "_%d.png" % epoch))
# plt.clf()
plt.close()
""" A -> B """
z = self.g_aB_z.predict(stft_a_z) * 256
t = self.g_aB_t.predict(stft_a_theta) * np.pi
data_aB = to_rect(z, t, stft_a.shape)
istft = va.istft(data_aB, self.window, self.step)
res = (istft * 32768.0).astype(np.int16)
p = self.res_path + "data_aB/" + os.path.basename(file_a)
p = p[:-4] + "_%d.wav" % epoch
va.save_wave(p, fs_b, res)
plt.figure()
plt.subplot(211)
plt.plot(data_a)
plt.subplot(212)
plt.plot(istft)
plt.ylim(-1, 1)
plt.savefig((self.res_path + "pic/data_aB/{}").format(
os.path.basename(file_b)[:-4] + "_%d.png" % epoch))
# plt.clf()
plt.close()
self.save_weight()
# メモリ解放
del z, t,
del data_bA, data_aB
del istft, res
gc.collect()
log = ("epoch:{:4}, batch:{:2} " +
"Z[dl_ba:{:.4f}, da:{:.2f}, cl_ba:{:.4f}, ca:{:.2f}, rl_ba:{:.4f}, nl_ba:{:.4f}] " +
"T[dl_ba:{:.4f}, da:{:.2f}, cl_ba:{:.4f}, ca:{:.2f}, rl_ba:{:.4f}, nl_ba:{:.4f}] " +
"Z[dl_ab:{:.4f}, da:{:.2f}, cl_ab:{:.4f}, ca:{:.2f}, rl_ab:{:.4f}, nl_ab:{:.4f}] " +
"T[dl_ab:{:.4f}, da:{:.2f}, cl_ab:{:.4f}, ca:{:.2f}, rl_ab:{:.4f}, nl_ab:{:.4f}] "). \
format(
epoch,
batch,
d_loss_ba_z[0],
d_loss_ba_z[1],
c_loss_ba_z[0],
c_loss_ba_z[1],
r_loss_ba_z,
nc_loss_ba_z,
d_loss_ba_t[0],
d_loss_ba_t[1],
c_loss_ba_t[0],
c_loss_ba_t[1],
r_loss_ba_t,
nc_loss_ba_t,
d_loss_ab_z[0],
d_loss_ab_z[1],
c_loss_ab_z[0],
c_loss_ab_z[1],
r_loss_ab_z,
nc_loss_ab_z,
d_loss_ab_t[0],
d_loss_ab_t[1],
c_loss_ab_t[0],
c_loss_ab_t[1],
r_loss_ab_t,
nc_loss_ab_t,
)
open(self.save_path + "log.txt", "a").write(log + "\n")
print(log)
except:
self.save_weight()
print(sys.exc_info())
sys.exit()
if __name__ == "__main__":
print("test finished")
sys.exit()
# メモリ解放
del input_a_z, input_a_t
del input_b_z, input_b_t
del idx_a, idx_b
del fake_a_z, fake_a_t
del fake_b_z, fake_b_t
del d_loss_ab_z, d_loss_ab_t
del d_loss_ba_z, d_loss_ba_t
del c_loss_ab_z, c_loss_ab_t
del c_loss_ba_z, c_loss_ba_t
del r_loss_ab_z, r_loss_ab_t
del r_loss_ba_z, r_loss_ba_t
del nc_loss_ab_z, nc_loss_ab_t
del nc_loss_ba_z, nc_loss_ba_t
del log
gc.collect()
# メモリ解放
del data_a, data_b
del fs_a, fs_b
del stft_a, stft_a_z, stft_a_theta
del stft_b, stft_b_z, stft_b_theta
gc.collect()
copy_tree("./save", "./backup")
print("backup finished : ./save to ./backup")
if __name__ == "__main__":
t = Train()
t.train()