-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathBP.py
144 lines (121 loc) · 4.39 KB
/
BP.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
#!/usr/bin/env python
# -*- coding:utf-8 -*-
# author: 徐聪
# datetime: 2022-10-23 11:35
# software: PyCharm
import numpy as np
import scipy.special as ssp
class BPNet:
def __init__(self, input_num, hide_num, output_num):
self.input_num = input_num
self.hide_num = hide_num
self.output_num = output_num
# 激活函数
self.active_fun = lambda x: ssp.expit(x)
# 生成隐藏层以及输出层的权重矩阵，正态分布，期望为0，方差为hide_num的-0.5次方
self.wih = np.random.normal(0.0, pow(self.hide_num, -0.5), (self.hide_num, self.input_num))
self.who = np.random.normal(0.0, pow(self.hide_num, -0.5), (self.output_num, self.hide_num))
def train(self, X, y, rate, epochs):
"""
训练神经网络
:param X: 样本特征
:param y: 标签
:param rate: 学习率
:param epochs: 迭代次数
:return:
"""
# 生成隐藏层以及输出层的权重矩阵，正态分布，期望为0，方差为hide_num的-0.5次方
self.wih = np.random.normal(0.0, pow(self.hide_num, -0.5), (self.hide_num, self.input_num))
self.who = np.random.normal(0.0, pow(self.hide_num, -0.5), (self.output_num, self.hide_num))
# 训练样本数目
num = X.shape[0]
erro_list = []
acc_list = []
epoch_list = []
# 神经网络模型的训练
print("BP-NeuralNetwork training begins")
for i in range(epochs):
erro = 0
# 遍历训练集中的数据
for j in range(num):
# 数据预处理
data = X[j, :]
label = np.ones(self.output_num) * 0.01
label[y[j, 0]] = 0.99
# 训练模型
t = self.net_train(data, label, rate)
erro += sum(t.T.dot(t)) / 3
if i % 50 == 0:
print(f"epoch={i}, erro={(erro / num)[0]}, acc={self.getAcc(X, y)}")
erro_list.append(erro / num)
acc_list.append(self.getAcc(X, y))
epoch_list.append(i)
# print(f"训练迭代次数：{i + 1}")
print("BP-NeuralNetwork training end")
return erro_list, acc_list, epoch_list
def getAcc(self, X, y):
"""
获取模型的准确率
:param X: 样本的特征矩阵
:param y: 标签
:return:
"""
y_predict = self.query(X)
cnt = 0
for i in range(y.shape[0]):
if y_predict[i, 0] == y[i, 0]:
cnt += 1
return cnt / y.shape[0]
def net_train(self, X, y, rate):
"""
模型训练
:param rate: 学习率
:param X: 特征
:param y: 标签
:return:
"""
# 数据初始化
data = np.array(X, ndmin=2).T
label = np.array(y, ndmin=2).T
# 前向传播
hide_z = self.wih.dot(data)
hide_a = self.active_fun(hide_z)
final_z = self.who.dot(hide_a)
final_a = self.active_fun(final_z)
# 反向传播
final_error = label - final_a
hide_error = self.who.T.dot(final_error)
# 利用公式计算梯度
final_gradient = np.dot((final_error * final_a * (1.0 - final_a)), np.transpose(hide_a))
hide_gradient = np.dot((hide_error * hide_a * (1.0 - hide_a)), np.transpose(data))
# 梯度下降更新权重矩阵 因为前面计算err时，是真实值-预测值，所以这里加梯度
self.who += rate * final_gradient
self.wih += rate * hide_gradient
return final_error
def query(self, X):
"""
获取预测分类结果
:param X: 样本特征
:return:
"""
num = X.shape[0]
y_predict = [0] * num
for i in range(num):
y_predict[i] = np.argmax(self.net_query(X[i]))
self.net_query(X[i])
y_predict = np.array(y_predict).reshape((num, 1))
return y_predict
def net_query(self, X):
"""
神经网络识别函数
:param X: 需要识别的数据
:return:
"""
# 数据初始化
data = np.array(X, ndmin=2).T
# 前向传播计算
hide_z = self.wih.dot(data)
hide_a = self.active_fun(hide_z)
final_z = self.who.dot(hide_a)
final_a = self.active_fun(final_z)
return final_a