脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - Python使用numpy实现BP神经网络

Python使用numpy实现BP神经网络

2021-01-20 00:48哇哇小仔 Python

这篇文章主要为大家详细介绍了Python使用numpy实现BP神经网络,具有一定的参考价值,感兴趣的小伙伴们可以参考一下

本文完全利用numpy实现一个简单的BP神经网络,由于是做regression而不是classification,因此在这里输出层选取的激励函数就是f(x)=x。BP神经网络的具体原理此处不再介绍。

?
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
import numpy as np
 
class NeuralNetwork(object):
  def __init__(self, input_nodes, hidden_nodes, output_nodes, learning_rate):
    # Set number of nodes in input, hidden and output layers.设定输入层、隐藏层和输出层的node数目
    self.input_nodes = input_nodes
    self.hidden_nodes = hidden_nodes
    self.output_nodes = output_nodes
 
    # Initialize weights,初始化权重和学习速率
    self.weights_input_to_hidden = np.random.normal(0.0, self.hidden_nodes**-0.5
                    ( self.hidden_nodes, self.input_nodes))
 
    self.weights_hidden_to_output = np.random.normal(0.0, self.output_nodes**-0.5
                    (self.output_nodes, self.hidden_nodes))
    self.lr = learning_rate
     
    # 隐藏层的激励函数为sigmoid函数,Activation function is the sigmoid function
    self.activation_function = (lambda x: 1/(1 + np.exp(-x)))
   
  def train(self, inputs_list, targets_list):
    # Convert inputs list to 2d array
    inputs = np.array(inputs_list, ndmin=2).T  # 输入向量的shape为 [feature_diemension, 1]
    targets = np.array(targets_list, ndmin=2).T 
 
    # 向前传播,Forward pass
    # TODO: Hidden layer
    hidden_inputs = np.dot(self.weights_input_to_hidden, inputs) # signals into hidden layer
    hidden_outputs = self.activation_function(hidden_inputs) # signals from hidden layer
 
     
    # 输出层,输出层的激励函数就是 y = x
    final_inputs = np.dot(self.weights_hidden_to_output, hidden_outputs) # signals into final output layer
    final_outputs = final_inputs # signals from final output layer
     
    ### 反向传播 Backward pass,使用梯度下降对权重进行更新 ###
     
    # 输出误差
    # Output layer error is the difference between desired target and actual output.
    output_errors = (targets_list-final_outputs)
 
    # 反向传播误差 Backpropagated error
    # errors propagated to the hidden layer
    hidden_errors = np.dot(output_errors, self.weights_hidden_to_output)*(hidden_outputs*(1-hidden_outputs)).T
 
    # 更新权重 Update the weights
    # 更新隐藏层与输出层之间的权重 update hidden-to-output weights with gradient descent step
    self.weights_hidden_to_output += output_errors * hidden_outputs.T * self.lr
    # 更新输入层与隐藏层之间的权重 update input-to-hidden weights with gradient descent step
    self.weights_input_to_hidden += (inputs * hidden_errors * self.lr).T
  
  # 进行预测  
  def run(self, inputs_list):
    # Run a forward pass through the network
    inputs = np.array(inputs_list, ndmin=2).T
     
    #### 实现向前传播 Implement the forward pass here ####
    # 隐藏层 Hidden layer
    hidden_inputs = np.dot(self.weights_input_to_hidden, inputs) # signals into hidden layer
    hidden_outputs = self.activation_function(hidden_inputs) # signals from hidden layer
     
    # 输出层 Output layer
    final_inputs = np.dot(self.weights_hidden_to_output, hidden_outputs) # signals into final output layer
    final_outputs = final_inputs # signals from final output layer 
     
    return final_outputs

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持服务器之家。

原文链接:http://blog.csdn.net/zhangyang10d/article/details/54804053

延伸 · 阅读

精彩推荐