-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_funcs.py
191 lines (149 loc) · 6.7 KB
/
train_funcs.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
import glob
import time
import os
import numpy as np
import hickle as hkl
from proc_load import crop_and_mirror
def proc_configs(config):
if not os.path.exists(config['weights_dir']):
os.makedirs(config['weights_dir'])
print "Create folder: " + config['weights_dir']
return config
def unpack_configs(config, ext_data='.hkl', ext_label='.npy'):
flag_para_load = config['para_load']
# Load Training/Validation Filenames and Labels
train_folder = config['train_folder']
val_folder = config['val_folder']
label_folder = config['label_folder']
trainA_filenames = sorted(glob.glob(train_folder + '/*A' + ext_data))
valA_filenames = sorted(glob.glob(val_folder + '/*A' + ext_data))
trainB_filenames = sorted(glob.glob(train_folder + '/*B' + ext_data))
valB_filenames = sorted(glob.glob(val_folder + '/*B' + ext_data))
#train_labels = np.load(label_folder + 'train_labels' + ext_label)
#val_labels = np.load(label_folder + 'val_labels' + ext_label)
img_mean = np.load(config['mean_file'])
img_mean = img_mean[:, :, :, np.newaxis].astype('float32')
return (flag_para_load,
trainA_filenames, valA_filenames, trainB_filenames, valB_filenames,
img_mean)
def adjust_learning_rate(config, epoch, step_idx, val_record, learning_rate):
# Adapt Learning Rate
if config['lr_policy'] == 'step':
if epoch == config['lr_step'][step_idx]:
learning_rate.set_value(
np.float32(learning_rate.get_value() / 10))
step_idx += 1
if step_idx >= len(config['lr_step']):
step_idx = 0 # prevent index out of range error
print 'Learning rate changed to:', learning_rate.get_value()
if config['lr_policy'] == 'auto':
if (epoch > 5) and (val_record[-3] - val_record[-1] <
config['lr_adapt_threshold']):
learning_rate.set_value(
np.float32(learning_rate.get_value() / 10))
print 'Learning rate changed to::', learning_rate.get_value()
return step_idx
def get_val_error_loss(rand_arrs, shared_xquery, shared_xp, shared_xns,
filenamesA, filenamesB,
flag_para_load, img_mean,
batch_size, validate_model,
send_queue=None, recv_queue=None,
flag_top_5=False):
validation_losses = []
validation_errors = []
if flag_top_5:
validation_errors_top_5 = []
n_val_batches = len(filenamesA)
if flag_para_load:
# send the initial message to load data, before each epoch
send_queue.put(str(filenamesA[0]))
send_queue.put(str(filenamesB[0]))
send_queue.put(np.float32([0.5, 0.5, 0]))
send_queue.put('calc_finished')
for val_index in xrange(n_val_batches):
if flag_para_load:
# load by self or the other process
# wait for the copying to finish
msg = recv_queue.get()
assert msg == 'copy_finished'
if val_index + 1 < n_val_batches:
name_to_read1 = str(filenamesA[val_index + 1])
send_queue.put(name_to_read1)
send_queue.put(np.float32([0.5, 0.5, 0]))
name_to_read2 = str(filenamesB[val_index + 1])
send_queue.put(name_to_read2)
send_queue.put(np.float32([0.5, 0.5, 0]))
else:
img1 = hkl.load(str(filenamesA[val_index])) - img_mean
img2 = hkl.load(str(filenamesB[val_index])) - img_mean
param_rand = [0.5,0.5,0]
img1 = crop_and_mirror(img1, param_rand, flag_batch=True)
img2 = crop_and_mirror(img2, param_rand, flag_batch=True)
shared_xquery.set_value(img1)
shared_xp.set_value(img2)
# set false colour variant matches
for i in xrange(len(shared_xns)):
shared_xn = shared_xns[i]
N = img2.shape[3]
n_indexes = np.arange(N) + np.random.randint(1,N-5)
n_indexes %= N
imgN = img2[:,:,:,n_indexes]
shared_xn.set_value(imgN)
if flag_top_5:
loss, error, error_top_5 = validate_model()
else:
loss, error = validate_model()
if flag_para_load and (val_index + 1 < n_val_batches):
send_queue.put('calc_finished')
validation_losses.append(loss)
validation_errors.append(error)
if flag_top_5:
validation_errors_top_5.append(error_top_5)
this_validation_loss = np.mean(validation_losses)
this_validation_error = np.mean(validation_errors)
#if flag_top_5:
# this_validation_error_top_5 = np.mean(validation_errors_top_5)
# return this_validation_error, this_validation_error_top_5, this_validation_loss
#else:
return this_validation_error, this_validation_loss
def get_rand3d():
tmp_rand = np.float32(np.random.rand(3))
tmp_rand[2] = round(tmp_rand[2])
return tmp_rand
def train_model_wrap(train_model, shared_xquery, shared_xp, shared_xns, rand_arrs, img_mean,
count, minibatch_index, minibatch_range, batch_size,
filenamesA, filenamesB,
flag_para_load, flag_batch,
send_queue=None, recv_queue=None):
if flag_para_load:
# load by self or the other process
# wait for the copying to finish
msg = recv_queue.get()
assert msg == 'copy_finished'
if count < len(minibatch_range):
ind_to_read = minibatch_range[count]
name_to_read1 = str(filenamesA[ind_to_read])
send_queue.put(name_to_read1)
send_queue.put(get_rand3d())
name_to_read2 = str(filenamesB[ind_to_read])
send_queue.put(name_to_read2)
send_queue.put(get_rand3d())
else:
img1 = hkl.load(str(filenamesA[minibatch_index])) - img_mean
img2 = hkl.load(str(filenamesB[minibatch_index])) - img_mean
param_rand1 = get_rand3d()
param_rand2 = get_rand3d()
img1 = crop_and_mirror(img1, param_rand1, flag_batch=flag_batch)
img2 = crop_and_mirror(img2, param_rand2, flag_batch=flag_batch)
shared_xquery.set_value(img1)
shared_xp.set_value(img2)
# set false colour variant matches
for i in xrange(len(shared_xns)):
shared_xn = shared_xns[i]
N = img2.shape[3]
n_indexes = np.arange(N) + np.random.randint(1,N-5)
n_indexes %= N
imgN = img2[:,:,:,n_indexes]
shared_xn.set_value(imgN)
cost_ij = train_model()
return cost_ij