-
Notifications
You must be signed in to change notification settings - Fork 328
/
Copy pathsolver_GOCD.py
executable file
·175 lines (149 loc) · 7.8 KB
/
solver_GOCD.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
# -*- coding: utf-8 -*-
import os
import logging
import time
class Solver(object):
def __init__(self,
mxnet_module,
trainset_dataiter,
net_symbol,
net_initializer,
optimizer_name,
optimizer_params,
data_names,
label_names,
context,
num_train_loops,
train_metric,
display_interval=10,
val_evaluation_interval=100,
valset_dataiter=None,
val_metric=None,
num_val_loops=0,
pretrained_model_param_path=None,
save_prefix=None,
start_index=0,
model_save_interval=None,
train_metric_update_frequency=1):
self.mxnet_module = mxnet_module
self.trainset_dataiter = trainset_dataiter
self.valset_dataiter = valset_dataiter
self.net_symbol = net_symbol
self.net_initializer = net_initializer
self.data_names = data_names
self.label_names = label_names
self.input_names = data_names + label_names
self.context = context
self.optimizer_name = optimizer_name
self.optimizer_params = optimizer_params
self.num_train_loops = num_train_loops
self.num_val_loops = num_val_loops
self.train_metric = train_metric
self.val_metric = val_metric
self.display_interval = display_interval
self.val_evaluation_interval = val_evaluation_interval
self.save_prefix = save_prefix
self.start_index = start_index
self.pretrained_model_param_path = pretrained_model_param_path
self.model_save_interval = model_save_interval
self.train_metric_update_frequency = \
train_metric_update_frequency if train_metric_update_frequency <= display_interval else display_interval
self.module = self.mxnet_module.module.Module(symbol=self.net_symbol,
data_names=self.data_names,
label_names=self.label_names,
context=self.context)
def __init_module(self):
arg_names = self.net_symbol.list_arguments()
arg_shapes, _, __ = self.net_symbol.infer_shape()
data_name_shape = [x for x in zip(arg_names, arg_shapes) if x[0] in self.data_names]
label_name_shape_temp = [x for x in zip(arg_names, arg_shapes) if x[0] in self.label_names]
# rearrange according to label_names
label_name_shape = []
for label_name in self.label_names:
for temp_item in label_name_shape_temp:
if temp_item[0] == label_name:
label_name_shape.append(temp_item)
break
self.module.bind(data_shapes=data_name_shape,
label_shapes=label_name_shape,
for_training=True,
grad_req='write')
if self.pretrained_model_param_path:
self.load_checkpoint()
self.module.params_initialized = True
else:
self.module.init_params(initializer=self.net_initializer,
allow_missing=True)
self.module.init_optimizer(kvstore='device',
optimizer=self.optimizer_name,
optimizer_params=self.optimizer_params)
def fit(self):
self.__init_module()
logging.info('Start training in %s.--------------------------------------------', str(self.context))
sum_time = 0
for i in range(self.start_index + 1, self.num_train_loops + 1):
start = time.time()
batch = self.trainset_dataiter.next()
self.module.forward(data_batch=batch, is_train=True)
self.module.backward()
# update parameters----------------------------------------------------------------------------------------
self.module.update()
outputs = [output for output in self.module.get_outputs() if not output.wait_to_read()]
# display training process----------------------------------------------------------------------------------
if i % self.train_metric_update_frequency == 0:
self.train_metric.update(batch.label, outputs)
sum_time += (time.time() - start)
if i % self.display_interval == 0:
names, values = self.train_metric.get()
logging.info('Iter[%d] -- Time elapsed: %.1f s. Speed: %.1f images/s.',
i, sum_time, self.display_interval * self.trainset_dataiter.get_batch_size() / sum_time)
for name, value in zip(names, values):
logging.info('%s: --> %.4f', name, value)
self.train_metric.reset()
sum_time = 0
if i % self.val_evaluation_interval == 0 and self.num_val_loops:
logging.info('Start validating-------------------------------------------')
for val_loop in range(self.num_val_loops):
val_batch = self.valset_dataiter.next()
self.module.forward(data_batch=val_batch, is_train=False)
outputs = [output for output in self.module.get_outputs() if not output.wait_to_read()]
self.val_metric.update(val_batch.label, outputs)
names, values = self.val_metric.get()
logging.info('Iter[%d] validation metric ------------- ', i)
for name, value in zip(names, values):
logging.info('%s: --> %.4f', name, value)
logging.info('End validating ---------------------------------------------')
self.val_metric.reset()
# save checkpoint--------------------------------------------------------------------------------
if i % self.model_save_interval == 0:
self.save_checkpoint(i)
def save_checkpoint(self, loop):
logging.info('\n<---------- Save checkpoint---------->')
save_model_name = '%s_iter_%d.params' % (self.save_prefix, loop)
if not os.path.exists(os.path.dirname(save_model_name)):
os.makedirs(os.path.dirname(save_model_name))
temp_arg_name_arrays, temp_aux_name_arrays = self.module.get_params()
# save model params
save_dict = {('arg:%s' % k): v.as_in_context(self.mxnet_module.cpu()) for k, v in temp_arg_name_arrays.items() if k not in self.input_names}
save_dict.update({('aux:%s' % k): v.as_in_context(self.mxnet_module.cpu()) for k, v in temp_aux_name_arrays.items()})
self.mxnet_module.nd.save(save_model_name, save_dict)
logging.info('Iter[%d] <--Save params to file: %s-->', loop, save_model_name)
def load_checkpoint(self):
logging.info('------>Load pre-trained model from file: %s.', self.pretrained_model_param_path)
# load model params
save_dict = self.mxnet_module.nd.load(self.pretrained_model_param_path)
arg_names = self.net_symbol.list_arguments()
# get the arg shapes
arg_name_arrays = {}
aux_name_arrays = {}
for k, v in save_dict.items():
tp, name = k.split(':', 1)
if name not in arg_names:
continue
if tp == 'arg':
arg_name_arrays.update({name: v.as_in_context(self.mxnet_module.cpu())})
if tp == 'aux':
aux_name_arrays.update({name: v.as_in_context(self.mxnet_module.cpu())})
self.module.init_params(self.net_initializer, arg_name_arrays, aux_name_arrays,
allow_missing=True,
force_init=True)