/
FullCNN0801.py
143 lines (133 loc) · 5.54 KB
/
FullCNN0801.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
from datetime import datetime
import os
import numpy as np
import tensorflow as tf
import gc
from sklearn.model_selection import KFold
import json
def listmap(o,p):
return list(map(o,p))
def loadPath():
with open("config.json") as f:
#这是用于自验证的代码
#with open("configSelfValid.json") as f:
config=json.loads(f.read())
return config["datapath"],config["sharepath"],config["rootpath"],config["selfvalidpath"],config["startdate"],config["days"]
datapath,sharepath,rootpath,selfvalidpath,startdate,days=loadPath()
lossplots=[]
outputs=None
def splitData(tensor,n_output,n_pred):
print(tensor.shape)
n_known=tensor.shape[0]-n_pred
n_input=tensor.shape[1]-n_output
knownX = tensor[0: n_known, 0: n_input]
knownY = tensor[0: n_known, n_input: n_input + n_output]
preX = tensor[n_known: n_known+n_pred, 0: n_input]
return (knownX,knownY,preX)
def fcn(trainX,trainY,hiddennum,times,keep,modelname,cutime,validX,validY,lr):
global outputs
inputnum=len(trainX[0])
outputnum=len(trainY[0])
losscol=[]
npx=trainX
npy=trainY
npx_test=validX
npy_test=validY
nodenums=[inputnum]+hiddennum+[outputnum]
sess=tf.InteractiveSession()
x=tf.placeholder(tf.float32, [None,inputnum])
y_=tf.placeholder(tf.float32,[None,outputnum])
keep_prob=tf.placeholder(tf.float32)
hiddens=[]
drops=[x]
for i in range(len(nodenums)-1):
if(i==len(nodenums)-2):
Wi=tf.Variable(tf.truncated_normal([nodenums[i],nodenums[i+1]],mean=10, stddev=0.1), name="W"+str(i)+cutime)
else:
Wi=tf.Variable(tf.truncated_normal([nodenums[i],nodenums[i+1]],mean=0, stddev=0.1), name="W"+str(i)+cutime)
bi= tf.Variable(tf.ones(nodenums[i+1]), name="b"+str(i)+cutime)
if i<len(nodenums)-2:
hiddeni = tf.nn.relu(tf.add(tf.matmul(drops[i],Wi),bi))
hiddens.append(hiddeni)
dropi=tf.nn.dropout(hiddeni,keep_prob)
drops.append(dropi)
else:
y=tf.add(tf.matmul(drops[i],Wi),bi)
lossfun=tf.reduce_mean(tf.abs(tf.subtract(y/y_,1)))
train_step=tf.train.AdamOptimizer(learning_rate=lr).minimize(lossfun)
init = tf.global_variables_initializer()
ts=0
while(True):
sess.run(init)
sess.run(train_step,feed_dict={x:npx,y_:npy,keep_prob:keep})
loss1=sess.run(lossfun,feed_dict={x:npx,y_:npy,keep_prob:1})
if(loss1<30):
break
ts+=1
if(ts>1000):
break
for i in range(times):
sess.run(train_step,feed_dict={x:npx,y_:npy,keep_prob:keep})
loss1=sess.run(lossfun,feed_dict={x:npx,y_:npy,keep_prob:1})
loss2=sess.run(lossfun,feed_dict={x:npx_test,y_:npy_test,keep_prob:1})
losscol.append([loss1,loss2,loss1+loss2])
losscolnp=np.array(losscol)
lossplots.append("'"+cutime+","+",".join(listmap(str,losscol[-1])))
print(losscol[-1])
# valid
preY_valid = sess.run(y, feed_dict={x: npx_test, keep_prob: 1})
np.savetxt(path + "preY_valid.csv", preY_valid.reshape(-1, 1), fmt="%.8f", delimiter=',')
# test
preY_test = sess.run(y, feed_dict={x: preX, keep_prob: 1})
np.savetxt(path + "preY_test.csv", preY_test.reshape(-1, 1), fmt="%.8f", delimiter=',')
if outputs is None:
outputs=preY_test.reshape(-1, 1)
else:
outputs=np.c_[outputs, preY_test.reshape(-1, 1)]
np.savetxt(path+"losscol.csv",losscolnp,fmt="%.8f",delimiter=',')
del sess
del hiddens
del drops
del losscolnp
del losscol
gc.collect()
allTask=listmap(str, list(range(1,6)))
if not os.path.exists(sharepath):
os.makedirs(sharepath)
for taskname in allTask:
outputs=None
inpath=rootpath+taskname+"\\"
sharetaskpath=sharepath+taskname+"\\"
if not os.path.exists(sharetaskpath):
os.makedirs(sharetaskpath)
cutime=datetime.strftime(datetime.now(),"%Y%m%d%H%M%S")
tensor=np.loadtxt(inpath+"tensor_fill.csv",delimiter=',')
knownX,knownY,preX=splitData(tensor,30,days)
alltimes=0
for i in range(5):
kf = KFold(n_splits=10)
for train_index, valid_index in kf.split(knownX):
print("TRAIN:", train_index, "VALID:", valid_index)
cutime=datetime.strftime(datetime.now(),"%Y%m%d%H%M%S")
path=inpath+cutime+"\\"
os.makedirs(path)
print(path)
trainX, validX = knownX[train_index], knownX[valid_index]
trainY, validY = knownY[train_index], knownY[valid_index]
print(trainX.shape,trainY.shape,validX.shape,validY.shape,preX.shape)
np.savetxt(path+"validYtrue.csv",validY.reshape(-1, 1), fmt="%.8f",delimiter=',')
lossi=fcn(trainX, trainY, [60,60,30], int(3e4), 0.88, taskname, cutime, validX, validY, 3e-4)
with open(path +"lossplots"+datetime.strftime(datetime.now(),"%Y%m%d%H%M%S")+".csv","w") as f:
f.write("\n".join(lossplots))
if(alltimes>=9):
break
else:
alltimes=alltimes+1
if(alltimes>=9):
break
cutime=datetime.strftime(datetime.now(),"%Y%m%d%H%M%S")
# np.savetxt(sharetaskpath+ "outputsmedian.csv", np.median(outputs,1), fmt="%.8f", delimiter=',')
# np.savetxt(sharetaskpath+ "outputsmean.csv", np.mean(outputs,1), fmt="%.8f", delimiter=',')
#本地Valid时使用下述代码生成本地预测结果
np.savetxt(datapath + "selfValid\\outputsmedian.csv", np.median(outputs, 1), fmt="%.8f", delimiter=',')
np.savetxt(datapath + "selfValid\\outputsmean.csv", np.mean(outputs, 1), fmt="%.8f", delimiter=',')