-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathDNN.py
157 lines (135 loc) · 6.23 KB
/
DNN.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
# coding: utf-8
import tensorflow as tf
import matplotlib.pyplot as plt
from sklearn.preprocessing import scale
from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np
import os
import time
# 创建保存模型和结果的文件夹
result_path = './result'
model_path = './result/model'
if not os.path.exists(model_path):
os.makedirs(model_path)#存放模型
if not os.path.exists(result_path):
os.makedirs(result_path)#存放日志和结果
# 创建日志文件
if not os.path.exists('./result/date.txt'):
with open('./result/date.txt', 'w') as acc_file:
pass
with open('./result/date.txt', 'a') as acc_file:
acc_file.write('\n%s \n' % (time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(time.time()))))
#数据导入
data = pd.read_csv("data/data.csv",)
x = data.iloc[:,:13]
y = data.MEDV
X_train,X_test,y_train,y_test = train_test_split(x,y,test_size=0.1,random_state=0)
X_train = np.array(X_train)
X_test = np.array(X_test)
y_train = np.array(y_train)
y_test = np.array(y_test)
y_train = y_train.reshape((-1,1))
y_test = y_test.reshape((-1,1))
#此处为将数据变为正态分布，可变也可不变
# X_train = scale(X_train)
# X_test = scale(X_test)
# y_train = np.array(y_train)
# y_test = np.array(y_test)
# y_train = scale(y_train.reshape((-1,1)))
# y_test = scale(y_test.reshape((-1,1)))
#NETWORK TOPOLOGIES设置模型框架的参数
n_hidden_1 = 10
n_hidden_2 = 5
n_hidden_3 = 3
n_input = 13
n_classes = 1
xs = tf.placeholder(shape=[None,X_train.shape[1]],dtype=tf.float32)
ys = tf.placeholder(shape=[None,1],dtype=tf.float32)
keep_prob_s = tf.placeholder(dtype=tf.float32)
pos = tf.placeholder(dtype=tf.float32)
#设置初始的权重和偏置项
stddev = 0.1#标准差
weights = {
'W1': tf.Variable(tf.random_normal([n_input,n_hidden_1], stddev=stddev)),
'W2': tf.Variable(tf.random_normal([n_hidden_1,n_hidden_2], stddev=stddev)),
'W3': tf.Variable(tf.random_normal([n_hidden_2,n_hidden_3], stddev=stddev)),
'out': tf.Variable(tf.random_normal([n_hidden_3,n_classes],stddev=stddev))
}
biases = {
'b1' : tf.Variable(tf.random_normal([n_hidden_1])),
'b2' : tf.Variable(tf.random_normal([n_hidden_2])),
'b3' : tf.Variable(tf.random_normal([n_hidden_3])),
'out' : tf.Variable(tf.random_normal([n_classes]))
}
#搭建神经网络(我这里分别试了一下2层、3层、4层的神经网络，并且发现随着层数的增加，非线性越来越好，对应的拟合效果也越好，但是层数肯定有极限！)
def multilayer_perceptron(_X, _weight, _biases,_keepratio):
layer_1 = tf.nn.relu(tf.add(tf.matmul(_X, _weight['W1']), _biases['b1']))
dropout_1 = tf.nn.dropout(layer_1,keep_prob=_keepratio)
layer_2 = tf.nn.relu(tf.add(tf.matmul(dropout_1,_weight['W2']),_biases['b2']))
dropout_2 = tf.nn.dropout(layer_2,keep_prob=_keepratio)
layer_3 = tf.matmul(dropout_2,_weight['W3']) + _biases['b3']
dropout_3 = tf.nn.dropout(layer_3,keep_prob=_keepratio)
layer_4 = tf.matmul(dropout_3,_weight['out']) + _biases['out']
dropout_4 = tf.nn.dropout(layer_4,keep_prob=_keepratio)
return dropout_4
#设置超参数
keep_prob=1 # 防止过拟合，取值一般在0.5到0.8。我这里是1，没有做过拟合处理
ITER =10000 # 一共迭代1万次，考虑样本数据比较少
p = 0.1
#预测
pred = multilayer_perceptron(xs,weights,biases,keep_prob_s)
#定义损失函数
loss = tf.reduce_mean(tf.reduce_sum(tf.square(y_train - pred),reduction_indices=[1]))
#优化器（这里有两个优化器都可以选择，可以比较结果那个好）
optm = tf.train.AdamOptimizer(learning_rate = pos).minimize(loss)
# optm = tf.train.GradientDescentOptimizer(learning_rate=0.05).minimize(loss)
#没迭代100次保存一个最优的模型
save_step = 100
saver = tf.train.Saver(max_to_keep=1) #最多保存1个模型
one = tf.constant(0.1,dtype=tf.float32)#学习率的迭代
def fit(X,y,n,keep_prob,p):
init = tf.global_variables_initializer()
stage_epochs = [5000,3000,2000]#分批次进行训练
with tf.Session() as sess:
tf.global_variables()
sess.run(init)
end = time.time()
loss1 = 50#初始的损失值设为50，一般的回归问题，初始loss是无穷大的
for i in range(n):
if (i + 1) in np.cumsum(stage_epochs)[:-1]:
p = sess.run(tf.multiply(p,one))
print(' ---Step into next stage---learning_rate: %.5f'%p)
with open('./result/date.txt', 'a') as acc_file:
acc_file.write('-----Step into next stage---learning_rate: %.5f----\n'%p)
feed_dict_train = {ys: y, xs: X, keep_prob_s : keep_prob,pos : p}
_loss,_ = sess.run([loss,optm], feed_dict = feed_dict_train)
else:
feed_dict_train = {ys: y, xs: X, keep_prob_s : keep_prob,pos : p}
_loss,_ = sess.run([loss,optm], feed_dict = feed_dict_train)
#每100次进行打印一次结果。
if i%200 == 0:
print('epoch:%d\t loss:%.8f\t Time:%.5f\t' % (i,_loss,time.time()-end))
end = time.time()
with open('./result/date.txt', 'a') as acc_file:
acc_file.write('epoch:%d\t loss:%.8f\t\n' % (i,_loss))
#保存模型
if i%save_step == 0:
if _loss < loss1:
saver.save(sess, "result/model/model.ckpt")
loss1 = _loss
fit(X=X_train,y=y_train,n=ITER,keep_prob = keep_prob,p = p)
#打开保存的最优model并预测测试集数据
with tf.Session() as sess:
saver = tf.train.import_meta_graph('./result/model/model.ckpt.meta')
saver.restore(sess, tf.train.latest_checkpoint("result/model/"))
y_pred = sess.run(pred, feed_dict={ys: y_test, xs: X_test, keep_prob_s : keep_prob,pos : p})
print("Test set predicted complete !/n")
#可视化比较测试集与预测的差别
print("Visually demonstrate test and predicted values")
plt.plot(range(len(y_test)),y_test,'b')
plt.plot(range(len(y_pred)),y_pred,'r--')
plt.legend(['y_test','y_pred'])
plt.title("Test set and predicted values")
plt.savefig('./result/result%s.jpg'%time.strftime('%H-%M-%S',time.localtime(time.time())))
plt.show()