-
Notifications
You must be signed in to change notification settings - Fork 40
/
train_seg_patch.py
190 lines (164 loc) · 7.24 KB
/
train_seg_patch.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
import cPickle as pickle
import string
import sys
import time
from itertools import izip
import lasagne as nn
import numpy as np
import theano
from datetime import datetime, timedelta
import utils
import logger
import theano.tensor as T
import buffering
from configuration import config, set_configuration
import pathfinder
# theano.config.warn_float64 = 'raise'
if len(sys.argv) < 2:
sys.exit("Usage: train.py <configuration_name>")
config_name = sys.argv[1]
set_configuration('configs_seg_patch', config_name)
expid = utils.generate_expid(config_name)
print
print "Experiment ID: %s" % expid
print
# metadata
metadata_dir = utils.get_dir_path('models', pathfinder.METADATA_PATH)
metadata_path = metadata_dir + '/%s.pkl' % expid
# logs
logs_dir = utils.get_dir_path('logs', pathfinder.METADATA_PATH)
sys.stdout = logger.Logger(logs_dir + '/%s.log' % expid)
sys.stderr = sys.stdout
print 'Build model'
model = config().build_model()
all_layers = nn.layers.get_all_layers(model.l_out)
all_params = nn.layers.get_all_params(model.l_out)
num_params = nn.layers.count_params(model.l_out)
print ' number of parameters: %d' % num_params
print string.ljust(' layer output shapes:', 36),
print string.ljust('#params:', 10),
print 'output shape:'
for layer in all_layers:
name = string.ljust(layer.__class__.__name__, 32)
num_param = sum([np.prod(p.get_value().shape) for p in layer.get_params()])
num_param = string.ljust(num_param.__str__(), 10)
print ' %s %s %s' % (name, num_param, layer.output_shape)
train_loss = config().build_objective(model, deterministic=False)
valid_loss = config().build_objective(model, deterministic=True)
learning_rate_schedule = config().learning_rate_schedule
learning_rate = theano.shared(np.float32(learning_rate_schedule[0]))
updates = config().build_updates(train_loss, model, learning_rate)
x_shared = nn.utils.shared_empty(dim=len(model.l_in.shape))
y_shared = nn.utils.shared_empty(dim=len(model.l_target.shape))
idx = T.lscalar('idx')
givens_train = {}
givens_train[model.l_in.input_var] = x_shared[idx * config().batch_size:(idx + 1) * config().batch_size]
givens_train[model.l_target.input_var] = y_shared[idx * config().batch_size:(idx + 1) * config().batch_size]
givens_valid = {}
givens_valid[model.l_in.input_var] = x_shared
givens_valid[model.l_target.input_var] = y_shared
# theano functions
iter_train = theano.function([idx], train_loss, givens=givens_train, updates=updates)
iter_get_predictions = theano.function([idx], nn.layers.get_output(model.l_out), givens=givens_train,
on_unused_input='ignore')
iter_get_targets = theano.function([idx], nn.layers.get_output(model.l_target), givens=givens_train,
on_unused_input='ignore')
iter_get_inputs = theano.function([idx], nn.layers.get_output(model.l_in), givens=givens_train,
on_unused_input='ignore')
iter_validate = theano.function([], valid_loss, givens=givens_valid)
if config().restart_from_save:
print 'Load model parameters for resuming'
resume_metadata = utils.load_pkl(config().restart_from_save)
nn.layers.set_all_param_values(model.l_out, resume_metadata['param_values'])
start_chunk_idx = resume_metadata['chunks_since_start'] + 1
chunk_idxs = range(start_chunk_idx, config().max_nchunks)
lr = np.float32(utils.current_learning_rate(learning_rate_schedule, start_chunk_idx))
print ' setting learning rate to %.7f' % lr
learning_rate.set_value(lr)
losses_eval_train = resume_metadata['losses_eval_train']
losses_eval_valid = resume_metadata['losses_eval_valid']
else:
chunk_idxs = range(config().max_nchunks)
losses_eval_train = []
losses_eval_valid = []
start_chunk_idx = 0
train_data_iterator = config().train_data_iterator
valid_data_iterator = config().valid_data_iterator
print
print 'Data'
print 'n train: %d' % train_data_iterator.nsamples
print 'n validation: %d' % valid_data_iterator.nsamples
print 'n chunks per epoch', config().nchunks_per_epoch
print
print 'Train model'
chunk_idx = 0
start_time = time.time()
prev_time = start_time
tmp_losses_train = []
# use buffering.buffered_gen_threaded()
for chunk_idx, (x_chunk_train, y_chunk_train, id_train) in izip(chunk_idxs, buffering.buffered_gen_threaded(
train_data_iterator.generate())):
if chunk_idx in learning_rate_schedule:
lr = np.float32(learning_rate_schedule[chunk_idx])
print ' setting learning rate to %.7f' % lr
print
learning_rate.set_value(lr)
# load chunk to GPU
x_shared.set_value(x_chunk_train)
y_shared.set_value(y_chunk_train)
# make nbatches_chunk iterations
chunk_train_losses = []
for b in xrange(config().nbatches_chunk):
loss = iter_train(b)
chunk_train_losses.append(loss)
tmp_losses_train.append(loss)
print chunk_idx, np.mean(chunk_train_losses)
if ((chunk_idx + 1) % config().validate_every) == 0:
print
print 'Chunk %d/%d' % (chunk_idx + 1, config().max_nchunks)
# calculate mean train loss since the last validation phase
mean_train_loss = np.mean(tmp_losses_train)
print 'Mean train loss: %7f' % mean_train_loss
losses_eval_train.append(mean_train_loss)
tmp_losses_train = []
# load validation data to GPU
tmp_losses_valid = []
for i, (x_chunk_valid, y_chunk_valid, ids_batch) in enumerate(
buffering.buffered_gen_threaded(valid_data_iterator.generate(),
buffer_size=2)):
x_shared.set_value(x_chunk_valid)
y_shared.set_value(y_chunk_valid)
l_valid = iter_validate()
print i, l_valid
tmp_losses_valid.append(l_valid)
# calculate validation loss across validation set
valid_loss = np.mean(tmp_losses_valid)
# TODO: taking mean is not correct if chunks have different sizes!!!
print 'Validation loss: ', valid_loss
losses_eval_valid.append(valid_loss)
now = time.time()
time_since_start = now - start_time
time_since_prev = now - prev_time
prev_time = now
est_time_left = time_since_start * (config().max_nchunks - chunk_idx + 1.) / (chunk_idx + 1. - start_chunk_idx)
eta = datetime.now() + timedelta(seconds=est_time_left)
eta_str = eta.strftime("%c")
print " %s since start (%.2f s)" % (utils.hms(time_since_start), time_since_prev)
print " estimated %s to go (ETA: %s)" % (utils.hms(est_time_left), eta_str)
print
if ((chunk_idx + 1) % config().save_every) == 0:
print
print 'Chunk %d/%d' % (chunk_idx + 1, config().max_nchunks)
print 'Saving metadata, parameters'
with open(metadata_path, 'w') as f:
pickle.dump({
'configuration_file': config_name,
'git_revision_hash': utils.get_git_revision_hash(),
'experiment_id': expid,
'chunks_since_start': chunk_idx,
'losses_eval_train': losses_eval_train,
'losses_eval_valid': losses_eval_valid,
'param_values': nn.layers.get_all_param_values(model.l_out)
}, f, pickle.HIGHEST_PROTOCOL)
print ' saved to %s' % metadata_path
print