-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy pathtrainFullTensor.py
More file actions
executable file
·179 lines (147 loc) · 7.31 KB
/
trainFullTensor.py
File metadata and controls
executable file
·179 lines (147 loc) · 7.31 KB
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
#!/usr/bin/env python
from __future__ import print_function
import argparse
import time
import os
import sys
import tensorflow as tf
import tensorflow.keras
import numpy as np
#from .model import Model
from . import data
import importlib
import tensorflow.keras as KK
################################
def train(args):
if True:
os.environ["CUDA_VISIBLE_DEVICES"]=args.CUDA_VISIBLE_DEVICES
# load the model based on name and access as Model: from .model import args.model
myimport = importlib.import_module("tfccs.%s" % args.model)
Model = myimport.Model
model = Model(args)
model.model.save(args.modelsave)
with open("%s.model.json" % args.modelsave,"w") as f:
f.write(model.model.to_json())
model.model.save_weights("%s.model.h5" % args.modelsave)
t0=time.time()
data_loader = data.data( args.batch_size, sys.argv[2],
inputdatName=args.inputdatName,
outputdatName=args.outputdatName)
t1=time.time()
print("time data_loader",str(t1-t0))
# get test if there
data_loader_test = None
if len(sys.argv)>3:
t0=time.time()
data_loader_test = data.data( args.batch_size, sys.argv[3],
inputdatName=args.inputdatName,
outputdatName=args.outputdatName)
t1=time.time()
print("time data_loader test",str(t1-t0))
tfconfig=tf.ConfigProto()
with tf.Session( config=tfconfig ) as sess:
sess.run(tf.global_variables_initializer())
if hasattr(args, "modelrestore"):
model.model = KK.models.load_model(args.modelrestore)
print("restored model", args.modelrestore)
print("# args.num_epochs", args.num_epochs, "args.batch_size", args.batch_size, "num_batches", data_loader.num_batches)
testLossAvgMIN = 999.9E+99
first=True
for e in range(args.num_epochs):
storeloss = []
data_loader.reset_batch_pointer()
for b in range(data_loader.num_batches):
start = time.time()
x, y = data_loader.next_batch()
if first:
first=False
print("x.shape",x.shape)
print("y.shape",y.shape)
print("x[4]",x[4])
print("y[4]",y[4])
# for ii in range(len(x)):
# print("ii",ii,"x[ii].shape",x[ii].shape)
# for ii in range(len(y)):
# print("ii",ii,"y[ii].shape",y[ii].shape)
#myfit=model.model.fit( x, [yid,ylen], epochs=1, batch_size=1,verbose=2)
myfit = model.model.train_on_batch( x, y )
end = time.time()
#print("epoch %d batch %d time %f" % (e, b, end-start))
for (kk,vv) in zip(model.model.metrics_names,[myfit]):
print("epoch",e,"batch",b,"trainMetric",kk,vv)
if kk=="loss":
if not isinstance(vv,list): vv = [vv] # only single loss
# handle multiple inputs
if isinstance(x,list):
myx = x[0]
else:
myx=x
storeloss.append( (vv[0],myx.shape[0]) )
# compute average loss across all batches
trainnum = 0
trainsum = 0.0
for xx in storeloss:
trainsum += xx[0]*xx[1]
trainnum += xx[1]
train_loss = trainsum/float(trainnum)
print("epoch %d trainLossAvg %f" % (e , train_loss))
#### Training ran through all the batches
# if save_every or at tend then save and run validation test
if (e % args.save_every == 0) or (e == args.num_epochs-1):
model.model.save(args.modelsave)
with open("%s.model.json" % args.modelsave,"w") as f:
f.write(model.model.to_json())
model.model.save_weights("%s.model.h5" % args.modelsave)
# run the test set if there
if data_loader_test is not None:
storeloss = []
data_loader_test.reset_batch_pointer()
for b in range(data_loader_test.num_batches):
x, y = data_loader_test.next_batch()
#mytest=model.model.evaluate( x, [yid,ylen],verbose=0)
mytest = model.model.test_on_batch( x, y )
for (kk,vv) in zip(model.model.metrics_names,[mytest]):
#print("epoch",e,"batch",b,"trainMetric",kk,"=",vv,"batchsize",x.shape[0])
if kk=="loss":
if not isinstance(vv,list): vv = [vv] # only single loss
# handle multiple inputs
if isinstance(x,list):
myx = x[0]
else:
myx=x
storeloss.append( (vv[0],myx.shape[0]) ) # vv[0] for multiple losses
# compute average loss across all batches
testnum = 0
testsum = 0.0
for xx in storeloss:
testsum += xx[0]*xx[1]
testnum += xx[1]
testLossAvg = testsum/float(testnum)
print("epoch %d testLossAvg %f" % (e , testLossAvg))
if testLossAvg < testLossAvgMIN:
testLossAvgMIN = testLossAvg
cmd = "mv %s %s.best" % (args.modelsave, args.modelsave)
print(cmd)
os.system(cmd)
cmd = "mv %s.model.json %s.model.json.best" % (args.modelsave, args.modelsave)
print(cmd)
os.system(cmd)
cmd = "mv %s.model.h5 %s.model.h5.best" % (args.modelsave, args.modelsave)
print(cmd)
os.system(cmd)
if testLossAvg > 2.0*testLossAvgMIN:
print("EARLY STOPPING:",testLossAvg,testLossAvgMIN)
return()
#sys.exit(1)
if __name__ == '__main__':
print("time begin",str(time.strftime('%Y-%m-%d %H:%M %Z', time.localtime(time.time()))))
exec(open(sys.argv[1]).read())
args.init_from = None
for aa in sys.argv:
if "EXEC:" in aa:
toexec = aa.replace("EXEC:","")
print("toexec",toexec)
exec(toexec)
print("-------")
train(args)
print("time end",str(time.strftime('%Y-%m-%d %H:%M %Z', time.localtime(time.time()))))