-
Notifications
You must be signed in to change notification settings - Fork 0
/
utilDANN.py
executable file
·296 lines (244 loc) · 13.8 KB
/
utilDANN.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
# -*- coding: utf-8 -*-
from __future__ import print_function
import numpy as np
import matplotlib.pyplot as plt
import gc
from sklearn.metrics import accuracy_score
from sklearn.utils import shuffle
from keras import backend as K
from keras.callbacks import EarlyStopping, TensorBoard
from keras.models import load_model
import matplotlib.pyplot as plt
import tensorflow as tf
import util
import utilIO
import utilMetrics
# ----------------------------------------------------------------------------
def get_dann_weights_filename(folder, from_dataset, to_dataset, config):
if config.grl_position == 0:
grl_position_str = ""
else:
grl_position_str = "_gpos" + str(config.grl_position)
if config.filter == False:
filter_str = ""
else:
filter_str = "_filter"
return '{}{}/weights_dannCONV_model_from_{}_to_{}_w{}_s{}_l{}_f{}_k{}_drop{}_page{}_super{}_e{}_b{}_lda{}_lda_inc{}_dmodel{}{}{}.npy'.format(
folder,
('/truncated' if config.truncate else ''),
from_dataset, to_dataset,
config.window, config.step,
config.nb_layers,
config.nb_filters, config.k_size,
str(config.dropout),
str(config.page), str(config.nb_super_epoch),
str(config.epochs), str(config.batch),
str(config.lda), str(config.lda_inc),
str(config.domain_model_version),
grl_position_str,
filter_str)
# ----------------------------------------------------------------------------
def get_dann_logs_directory(folder, from_dataset, to_dataset, config):
weights_filename = get_dann_weights_filename(folder, from_dataset, to_dataset, config)
return weights_filename.replace("/weights_dann", "/logs_dann")
# ----------------------------------------------------------------------------
def get_dann_csv_logs_directory(folder, from_dataset, to_dataset, config):
weights_filename = get_dann_weights_filename(folder, from_dataset, to_dataset, config)
return weights_filename.replace("/weights_dann", "/csv_logs_dann")
# ----------------------------------------------------------------------------
def batch_generator(
x_data, y_data=None,
batch_size=1,
shuffle_data=True):
len_data = len(x_data)
index_arr = np.arange(len_data)
if shuffle_data:
np.random.shuffle(index_arr)
start = 0
while len_data > start + batch_size:
batch_ids = index_arr[start:start + batch_size]
start += batch_size
if y_data is not None:
x_batch = x_data[batch_ids]
y_batch = y_data[batch_ids]
yield x_batch, y_batch
else:
x_batch = x_data[batch_ids]
yield x_batch
# ----------------------------------------------------------------------------
def train_dann_batch(config, dann_model, src_generator, target_genenerator, target_x_train, batch_size, with_filter, model_cnn, histogram_source, threshold_correl_pearson):
#### NEW MODEL ####
#domain0 = np.zeros(target_x_train.shape[1:], dtype=int)
#domain1 = np.ones(target_x_train.shape[1:], dtype=int)
domain_output_shape = dann_model.output_shape[0]
#print(domain_output_shape)
batchYd = np.concatenate(( np.zeros((batch_size // 2,) + domain_output_shape[1:], dtype=int),
np.ones((batch_size // 2,) + domain_output_shape[1:], dtype=int)) )
# np.tile(domain0, [batch_size // 2, 1, 1, 1]),
# np.tile(domain1, [batch_size // 2, 1, 1, 1])))
#print(batchYd)
#print(batchYd.shape)
for batchXs, batchYs in src_generator:
try:
batchXd = next(target_genenerator)
except: # Restart...
target_genenerator = batch_generator(
target_x_train, None,
batch_size=batch_size // 2)
batchXd = next(target_genenerator)
# Combine the labeled and unlabeled data along with the discriminative results
combined_batchX = np.concatenate((batchXs, batchXd))
batch2Ys = np.concatenate((batchYs, batchYs))
#### NEW MODEL ####
#batchYd = np.concatenate((np.tile([0, 1], [batch_size // 2, 1]),
# np.tile([1, 0], [batch_size // 2, 1])))
#print(combined_batchX.shape, batch2Ys.shape, batchYd.shape)
result = dann_model.train_on_batch(combined_batchX,
{'classifier_output': batch2Ys,
'domain_output':batchYd})
#print(dann_model.metrics_names)
# print(result)
return result
# ----------------------------------------------------------------------------
def __train_dann_page(config,
dann_builder, source_x_train, source_y_train, source_x_test, source_y_test,
target_x_train, target_y_train, target_x_test, target_y_test,
nb_epochs, batch_size,
lda_inc,
weights_filename_dann, model_cnn,
histogram_source,
threshold_correl_pearson,
csv_logs_directory,
page,
with_tensorboard, tensorboard,
with_filter):
best_label_f1 = -1
target_genenerator = batch_generator(
target_x_train, None,
batch_size=batch_size // 2)
def named_logs(source_f1_train, source_f1_test, target_f1_train, target_f1_test, hp_lambda):
result = {}
result["source_f1_train"] = source_f1_train
result["source_f1_test"] = source_f1_test
result["target_f1_train"] = target_f1_train
result["target_f1_test"] = target_f1_test
result["lambda"] = hp_lambda
return result
util.mkdirp(csv_logs_directory)
csv_logs_filename = csv_logs_directory + "/" + str(page) + "_logs.csv"
csv_logs_file = open(csv_logs_filename,'a+')
csv_logs_file.write("#Epoch\tSRC_TRAIN\tTRG_TRAIN\tSRC_TEST\tTRG_TEST\tTRG_MSE\tTRG_LOSS\tLBL_MSE\tLBL_LOSS\tLAMBDA\n")
csv_logs_file.close()
for e in range(nb_epochs):
src_generator = batch_generator(source_x_train, source_y_train, batch_size=batch_size // 2)
# Update learning rates
if type(dann_builder.opt) is str:
lr = dann_builder.opt
else:
lr = float(K.get_value(dann_builder.opt.lr))* (1. / (1. + float(K.get_value(dann_builder.opt.decay)) * float(K.get_value(dann_builder.opt.iterations)) ))
print(' - Lr:', lr, ' / Lambda:', dann_builder.grl_layer.get_hp_lambda())
dann_builder.grl_layer.increment_hp_lambda_by(lda_inc)
# Train batch
logs = train_dann_batch(config,
dann_builder.dann_model, src_generator, target_genenerator, target_x_train, batch_size,
with_filter, model_cnn,
histogram_source, threshold_correl_pearson)
loss, domain_loss, label_loss, domain_acc, label_mse = logs
#print("Source train")
#source_prediction_train = dann_builder.label_model.predict(source_x_train, batch_size=32, verbose=0)
#source_f1_train, source_th_train, _, _ = utilMetrics.calculate_best_fm(source_prediction_train, source_y_train)
print("Source test")
source_prediction_test = dann_builder.label_model.predict(source_x_test, batch_size=32, verbose=0)
source_f1_test, source_th_test, _, _ = utilMetrics.calculate_best_fm(source_prediction_test, source_y_test)
#print("Target train")
#target_prediction_train = dann_builder.label_model.predict(target_x_train, batch_size=32, verbose=0)
#target_f1_train, target_th_train, _, _ = utilMetrics.calculate_best_fm(target_prediction_train, target_y_train)
print("Target test")
target_prediction_test = dann_builder.label_model.predict(target_x_test, batch_size=32, verbose=0)
target_f1_test, target_th_test, _, _ = utilMetrics.calculate_best_fm(target_prediction_test, target_y_test, source_th_test)
saved = ""
if source_f1_test >= best_label_f1:
best_label_f1 = source_f1_test
dann_builder.save(weights_filename_dann)
saved = "SAVED"
#target_loss, target_mse = dann_builder.label_model.evaluate(target_x_train, target_y_train, batch_size=32, verbose=0)
target_loss, target_mse = dann_builder.label_model.evaluate(target_x_test, target_y_test, batch_size=32, verbose=0)
#print("Epoch [{}/{}]: source label loss={:.4f}, mse={:.4f} | domain loss={:.4f}, acc={:.4f} | target label loss={:.4f}, mse={:.4f} | {}".format(
# e+1, nb_epochs, label_loss, label_mse, domain_loss, domain_acc, target_loss, target_mse, saved))
print("Epoch [{}/{}]: source label loss={:.4f}, mse={:.4f}, f1={:.4f} | domain loss={:.4f}, acc={:.4f} | target label loss={:.4f}, mse={:.4f}, f1={:.4f} | {}".format(
e+1, nb_epochs, label_loss, label_mse, source_f1_test, domain_loss, domain_acc, target_loss, target_mse, target_f1_test, saved))
if with_tensorboard:
tensorboard.on_epoch_end(e, named_logs(
source_f1_train=0, #source_f1_train,
source_f1_test=source_f1_test,
target_f1_train=0, #target_f1_train,
target_f1_test=target_f1_test,
hp_lambda=dann_builder.grl_layer.get_hp_lambda()))
csv_logs_file = open(csv_logs_filename,'a+')
csv_logs_file.write("%d\t%f\t%f\t%f\t%f\t%f\t%f\t%f\t%f\t%f\n"%\
(e,\
0,\
#source_f1_train,\
0,\
#target_f1_train,\
source_f1_test,\
target_f1_test,\
target_mse,\
target_loss,\
label_mse,\
label_loss,\
dann_builder.grl_layer.get_hp_lambda()\
))
csv_logs_file.close()
gc.collect()
# ----------------------------------------------------------------------------
def train_dann(dann_builder, source, target,
weights_filename_dann, model_cnn, histogram_source, threshold_correl_pearson,
parent_logs_directory, csv_logs_directory,
config):
print('Training DANN model...')
dann_builder.grl_layer.set_hp_lambda(config.lda)
logs_directory = get_dann_logs_directory( parent_logs_directory, source['name'], target['name'], config)
util.deleteFolder(logs_directory)
csv_logs_directory = get_dann_csv_logs_directory( csv_logs_directory, source['name'], target['name'], config)
util.deleteFolder(csv_logs_directory)
if config.tboard:
tensorboard = TensorBoard(
log_dir=logs_directory,
histogram_freq=0,
batch_size=config.batch,
write_graph=True,
write_grads=True
)
tensorboard.set_model(dann_builder.dann_model)
else:
tensorboard = None
for se in range(config.nb_super_epoch):
print(80 * "-")
print("SUPER EPOCH: %03d/%03d" % (se+1, config.nb_super_epoch))
source['generator'].reset()
source['generator'].shuffle()
for source_x_train, source_y_train in source['generator']:
try:
target_x_train, target_y_train = next(target['generator'])
except: # Restart...
target['generator'].reset()
target['generator'].shuffle()
target_x_train, target_y_train = next(target['generator'])
print("> Train on %03d source page samples and %03d target page samples..." % (len(source_x_train), len(target_x_train)))
#print(source_x_train.shape)
#print(source_y_train.shape)
#print(target_x_train.shape)
page = source['generator'].get_pos()
__train_dann_page(config, dann_builder, source_x_train, source_y_train, source['x_test'], source['y_test'],
target_x_train, target_y_train, target['x_test'], target['y_test'],
config.epochs, config.batch,
config.lda_inc,
weights_filename_dann,
model_cnn,
histogram_source,
threshold_correl_pearson,
csv_logs_directory,
page,
config.tboard, tensorboard,
config.filter)