-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_net.py
executable file
·158 lines (138 loc) · 5.45 KB
/
train_net.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
import numpy as np
from loader import TieLoader
from networks import make_FCN
from train_tools import make_train,make_test
import theano
import theano.tensor as T
from sklearn.metrics import average_precision_score
import datetime
from utils import get_pp_pn
from train_tools import fit
from test_tools import test_network
import argparse
from easydict import EasyDict
from utils import tee
import os
def train_network(cfg):
try:
os.mkdir(cfg.NAME)
except Exception as e:
print 'cannot create dir ',cfg.NAME
print e
if( cfg.TRAIN.EPOCH < 0):
epoch = np.array([int(i[:-len('.npz')]) for i in os.listdir(cfg.NAME+"/models")]).max()
cfg.TRAIN.EPOCH = int(epoch)
print 'start epoch ',int(epoch)
logger_name = cfg.NAME+'/train.log'
train_loader = TieLoader('data/train_ties%d'%cfg.DATASET.T_SIZE,
cfg.DATASET.TRAIN_MINR,
cfg.DATASET.TRAIN_MAXR,
t_size=cfg.DATASET.T_SIZE,
mask_size=cfg.OUT_SIZE,
sample_size=cfg.TILE_SIZE,
cache_samples=cfg.DATASET.CASHE_SAMPLES)
test_loader = TieLoader('data/test_ties%d'%cfg.DATASET.T_SIZE,
cfg.DATASET.TEST_MINR,
cfg.DATASET.TEST_MAXR,
t_size=cfg.DATASET.T_SIZE,
mask_size=cfg.OUT_SIZE,
sample_size=cfg.TILE_SIZE,
cache_samples=cfg.DATASET.CASHE_SAMPLES)
data = T.tensor4(name='data')
label = T.tensor3(name='label')
net = make_FCN(cfg.NETWORK,data,
ndim=cfg.NDIM,
model_name='%s/models/%03d'%(cfg.NAME,cfg.TRAIN.EPOCH) if cfg.TRAIN.EPOCH > 0 else '',
input_shape = (None,3,cfg.TILE_SIZE,cfg.TILE_SIZE),
pad = 'same')
non_learn_params={'min_cov' : theano.shared(1e-8),
'lr' : theano.shared(np.array(1e-2, dtype=theano.config.floatX)),
'width': theano.shared(4.),
'total_grad_constraint': 10,
'histogram_bins' : 100,
'use_approx_grad' : True,
'ndim' : cfg.NDIM,
'gm_num' : cfg.GM_NUM}
train_fn = make_train(net,data,label,non_learn_params)
print 'train_fn compiled'
test_fn = make_test(net,data,label,non_learn_params)
print 'test_fn compiled'
metrix = { 'aps' : average_precision_score,
'int_pp_pn' : lambda l,pred : get_pp_pn(l,pred)}
def update_params(epoch,params):
if(epoch == 0):
params['lr'].set_value(5e-2)
if(epoch == 4):
params['lr'].set_value(5e-3)
if(epoch == 10):
params['lr'].set_value(1e-3)
logger = open(logger_name,'a')
tee('\n################### train network '+cfg.NAME+ ' ' + str(datetime.datetime.now())+'################\n',logger)
tee('config\n'+str(cfg),logger)
tee('non_learn_params\n'+str(non_learn_params),logger)
fit(cfg.NAME,net,train_fn,test_fn,train_loader,test_loader,non_learn_params,
update_params=update_params,
metrix = metrix,
logger=logger,
train_esize = cfg.TRAIN.TRAIN_EPOCH_SIZE ,
test_esize = cfg.TRAIN.TEST_EPOCH_SIZE ,
epochs=cfg.TRAIN.EPOCH_NUM,
start_epoch=cfg.TRAIN.EPOCH)
tee('\n################### test network '+cfg.NAME+ ' ' + str(datetime.datetime.now())+'################\n',logger)
test_network(cfg.NAME,cfg.NETWORK,cfg.NDIM,cfg.TRAIN.EPOCH_NUM-1,cfg.GM_NUM,im_size=(320,240),train_size=100,test_size=400)
tee('\n################### done #######################\n',logger)
cfg = EasyDict()
cfg.SEQ_LENGTH = 250
cfg.TILE_SIZE = 9
cfg.OUT_SIZE = 1
cfg.GM_NUM = 4
cfg.NDIM = 4
cfg.NAME_PREFIX = ''
cfg.NETWORK = ''
cfg.TRAIN = EasyDict()
cfg.TRAIN.EPOCH = 0
cfg.TRAIN.TRAIN_EPOCH_SIZE = 1500
cfg.TRAIN.TEST_EPOCH_SIZE = 750
cfg.TRAIN.EPOCH_NUM = 20
cfg.DATASET = EasyDict()
cfg.DATASET.TRAIN_MINR = 0.2
cfg.DATASET.TRAIN_MAXR = 0.3
cfg.DATASET.TEST_MINR = 0.2
cfg.DATASET.TEST_MAXR = 0.3
cfg.DATASET.T_SIZE = 32
cfg.DATASET.CASHE_SAMPLES = True
def iter_all_keys_in_dict(prefix,d):
res = []
for k in d.keys():
try:
d[k].keys()
res = res + iter_all_keys_in_dict((prefix+'.' if( len(prefix) > 0) else '')+k,d[k])
except:
names = (prefix+'.' if( len(prefix) > 0) else '')+k
default = d[k]
res.append((names,default))
return res
def set_values_to_config(args,cfg):
for arg in vars(args):
val = getattr(args, arg)
if(val is None):
continue
arg = arg.upper()
c = cfg
path = arg.split('.')
for i in path[:-1]:
c = c[i]
c[path[-1]] = val
print 'set %s to '%arg,val
return cfg
parser = argparse.ArgumentParser(description='Train gmm segmentation network')
reqiered_args = ['ndim','network_prefix','network']
for name,default in iter_all_keys_in_dict('',cfg):
if(name.lower() in reqiered_args):
parser.add_argument('--'+name.lower(), type = type(default),default=None)
else:
parser.add_argument('--'+name.lower(), type = type(default),default=None,nargs='?')
args = parser.parse_args()
cfg = set_values_to_config(args,cfg)
cfg.NAME='experiments/'+cfg.NAME_PREFIX+cfg.NETWORK+'%d'%(cfg.NDIM)
train_network(cfg)