公告

测试，学习中

昵称： zz_BooM
园龄： 2年6个月
粉丝： 0
关注： 2

+加关注

随笔分类

随笔档案

阅读排行榜

回忆网络：2022/9/25

回忆

In [35]:

import numpy as np
#scipy.special for the sigmoid function expit()
import scipy.special

绘制数组

In [36]:

import numpy as np
import matplotlib.pyplot as plt
a = np.zeros([3,2])
a[0,1]=2
plt.imshow(a,interpolation="nearest")
# plt.imshow(a)

Out[36]:

<matplotlib.image.AxesImage at 0x195185612e0>

使用python制作神经网络¶

1.网络框架：¶

初始化函数：设定输入层节点、隐藏层节点、输出层节点数量
训练：学习给定训练集样本后，优化权重
查询：给定输入，从输出节点给定答案

In [37]:

#代码框架（神经网络框架）
class neuralNetwork:
    
    #initialise(初始化) the neural network
    def __init__():
        pass

    #train the neural network
    def train():
        pass

    #query(询问) the neural network
    def query():
        pass

2.初始化网络¶

设定输入层节点、隐藏层节点、输出层节点数量
1. 设定节点数量参数，该类参数定义了神经网络的形状与尺寸
2. 设置学习率

In [38]:

#initialise(初始化) the neural network

def __init__(self,inputnodes,hiddennodes,outpotnodes,learningrate):

    #set number of nodes(节点) in each input,hidden,output layer

    self.inodes = inputnodes
    self.hnodes = hiddennodes
    self.onodes = outputnodes

    #learning rate

    self.lr = learningrate
    pass
   

整合

In [39]:

 #代码框架（神经网络框架）
class neuralNetwork:
    
    #initialise(初始化) the neural network
    def __init__(self,inputnodes,hiddennodes,outputnodes,learningrate):

        #set number of nodes(节点) in each input,hidden,output layer
        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes
        #learning rate
        self.lr = learningrate
        
        pass

    #train the neural network
    def train():
        pass

    #query(询问) the neural network
    def query():
        pass

In [40]:

#此处为参数例子：创建一个每层3个节点、学习率为0.5的小型神经网络对象

#number of input，hidden and output nodes
input_nodes = 3
hidden_nodes = 3
output_nodes = 3

#learning rate is 0.5
learning_rate = 0.5

#create instance(实例) of neural network:

n= neuralNetwork(input_nodes,hidden_nodes,output_nodes,learning_rate)
#该实例还没什么用，因为函数还没设定

3.权重——网络的核心¶

创建网络的节点和链接，链接权重。网络使用权重来计算前向传播与反向传播，进而改进权重
1. 输入层与隐藏层之间的链接权重矩阵： $W_{i n p u t_h i d d e n}, W_{i n p u t_h i d d e n} . s h a p e （大小） = h i d d e n_n o d e s \times i n p u t_n o d e s$
2. 隐藏层与输出层之间的链接权重矩阵： $W_{h i d d e n_o u t p u t}, W_{h i d d e n_o u t p u t} . s h a p e （大小） = h i d d e n_n o d e s \times o u t p u t_n o d e s$

In [41]:

import numpy as np
# np.random.rand(rows,colimns)
# np.random.rand(3,3)#生成0~1之间的数，组成3*3的矩阵
np.random.rand(3,3)-0.5#生成-0.5~0.5之间的数，组成3*3的矩阵

#可用作初始化权重矩阵（均匀分布）:
#link weight matrices,wih and who

#weights inside the arrays ate w_i_j, where link is from node i to node j in the next layer

#w11 w21
#w12 w22 etc

#此处注释掉
# self.wih = (np.random.rand(self.hnodes,self.inodes)-0.5)
# self.who = (np.random.rand(self.onodes,self.inodes)-0.5)

Out[41]:

array([[-0.12651293,  0.11044255,  0.35225612],
       [-0.37125033,  0.09278935,  0.05095061],
       [-0.14624907,  0.17131378,  0.39858423]])

In [42]:

#另一种初始化权重的方法（标准正态分布）:
#pow(self.下一层节点,-0.5)-->开方

# self.wih = np.random.normal(0.0,pow(self.hnodes,-0.5),(self.hnodes,self.inodes))
# self.who = np.random.normal(0.0,pow(self.onodes,-0.5),(self.onodes,self.hnodes))

整合

In [43]:

 #代码框架（神经网络框架）
class neuralNetwork:
    
    #initialise(初始化) the neural network
    def __init__(self,inputnodes,hiddennodes,outputnodes,learningrate):

        #set number of nodes(节点) in each input,hidden,output layer
        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes

        # link weight matrices , wih and who
        # weights inside the arrays are w_i_j, where link is from node i to node j in the next layer

        #w11 w21
        #w12 w22 etc
        self.wih = (np.random.rand(self.hnodes,self.inodes)-0.5)
        self.who = (np.random.rand(self.onodes,self.inodes)-0.5)

        #learning rate
        self.lr = learningrate
        
        pass

    #train the neural network
    def train():
        pass

    #query(询问) the neural network
    def query():
        pass

4.查询网络¶

编写query()函数:接受神经网络的输入，返回网络的输出

信号输入->输入层节点->隐藏层节点->输出层节点->信号输出（注意不是训练哦！）

链接权重，如： $$

 X_{hidden} = W_{input\_hidden} \cdot I_{input}

#使用numpy实现
 hidden_inputs = np.dot(self.wih,inputs)

激活函数，如sigmoid： $O_{h i d d e n} = s i g m o i d (X_{h i d d e n})$ scipy库里有sigmoid函数，称为expit()
须在初始化时就指定激活函数的种类:

#scipy.special for the sigmoid function expit()
import scipy.special

# activation(激活) fun is the sigmoid function
#激活函数命名为了self.activation_function()
self.activation_function = lambda x : scipy.special.expit(x) #匿名函数，放到__init__中

#应用到准备进入隐藏层节点的信号

#calculate(计算) the signals(信号) emerging(出现) from hidden layer
hidden_outputs =self.activation_function(hidden_inputs)

整合查询网络部分的代码

#整合：输入层到隐藏层
  # calculate signals into hidden layer
  hidden_inputs = np.dot(self.wih,inputs)

  # calculate the signals emerging from hidden layer
  hidden_outputs =self.activation_function(hidden_inputs)

  #整合：隐藏层到输出层，最终输出

  # calculate signals into final output layer
  final_inputs = np.dot(self.who,hidden_outputs)

  # calculate the signals emerging from final output layer
  final_outputs =self.activation_function(final_inputs)

整合

In [44]:

 #代码框架（神经网络框架）
class neuralNetwork:
    
    #initialise(初始化) the neural network
    def __init__(self,inputnodes,hiddennodes,outputnodes,learningrate):

        #set number of nodes(节点) in each input,hidden,output layer
        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes

        # link weight matrices , wih and who
        # weights inside the arrays are w_i_j, where link is from node i to node j in the next layer

        #w11 w21
        #w12 w22 etc
        self.wih = (np.random.rand(self.hnodes,self.inodes)-0.5)
        self.who = (np.random.rand(self.onodes,self.inodes)-0.5)

        #learning rate
        self.lr = learningrate

        # activation(激活) fun is the sigmoid function
        #激活函数命名为了self.activation_function()
        self.activation_function = lambda x : scipy.special.expit(x) #匿名函数，放到__init__中
        
        pass

    #train the neural network
    def train():
        pass

    #query(询问) the neural network
    def query(self,inputs_list):

        # convert(转换) inputs list to 2d array
        inputs = np.array(inputs_list,ndmin=2).T
    
        #输入层到隐藏层
        # calculate signals into hidden layer
        hidden_inputs = np.dot(self.wih,inputs)
        # calculate the signals emerging from hidden layer
        hidden_outputs =self.activation_function(hidden_inputs)

        #隐藏层到输出层，最终输出

        # calculate signals into final output layer
        final_inputs = np.dot(self.who,hidden_outputs)
        # calculate the signals emerging from final output layer
        final_outputs =self.activation_function(final_inputs)
        
        return final_outputs

In [45]:

#测试
#此处为参数例子：创建一个每层3个节点、学习率为0.5的小型神经网络对象

#number of input，hidden and output nodes
input_nodes = 3
hidden_nodes = 3
output_nodes = 3

#learning rate is 0.3
learning_rate = 0.3

#create instance(实例) of neural network:
n= neuralNetwork(input_nodes,hidden_nodes,output_nodes,learning_rate)

#该实例还没什么用，因为训练函数还没设定

In [46]:

n.query([1.0,0.5,-1.5])
#该询问函数还没什么用，因为网络还没训练，此时输出的是初始随机矩阵处理的结果

Out[46]:

array([[0.44087252],
       [0.45725042],
       [0.4933395 ]])

5.训练网络¶

与query()函数一样：输入样本，得到输出
将计算所得输出与所需输出对比，使用插值来指导网络
第一部分：

In [47]:

#模仿query()函数，得到第一部分
#train the neural network
def train(self , inputs_list , targets_list):
    # convert(转换) inputs list to 2d array
    inputs = np.array(inputs_list,ndmin=2).T#将列表转为numpy数组

    #导入训练集的y真实值
    targets = np.array(targets_list,ndim=2).T#将列表转为numpy数组

    #输入层到隐藏层
    # calculate signals into hidden layer
    hidden_inputs = np.dot(self.wih,inputs)
    # calculate the signals emerging from hidden layer
    hidden_outputs =self.activation_function(hidden_inputs)

    #隐藏层到输出层，最终输出

    # calculate signals into final output layer
    final_inputs = np.dot(self.who,hidden_outputs)
    # calculate the signals emerging from final output layer
    final_outputs =self.activation_function(final_inputs)
    
    pass

第二部分：求误差， $e r r o r s_{o u t p u t} = y - \hat{y}$

#output layer error is the (target - actual)
output_errors = targets - final_outputs

每个隐藏层节点重组误差： $e r r o r s_{h i d d e n} = w e i g h t s_{h i d d e n_o u t p u t}^{T} ∙ e r r o r s_{o u t p u t}$

# hidden layer error is the output_errors, split by weights,recombind at hidden nodes
hidden_errors = np.dot(self.who.T,output_errors)

更新节点j与下一层节点k之间的链接权重的矩阵表达式： $△ W_{j, k} = α * E_{k} * s i g m o i d (O_{k}) * (1 - s i g m o i d (O_{k})) ∙ O_{j}^{T}$ $W_{k} = W_{j} + △ W_{j, k}$ $α$ 为学习率， $*$ 为数乘， $∙$ 为点积

# update the weights for the links between the hidden and output layers
self.who += self.lr * np.dot((output_errors * final_outputs * (1.0 - final_outputs)) , np.transpose(hidden_outputs))

#同理

# update the weights for the links between the input and hidden layers
self.wih += self.lr * np.dot((hidden_errors * hidden_outputs * (1.0 - hidden_outputs)) , np.transpose(inputs))

整合：神经网络完全体

In [48]:

 #代码框架（神经网络框架）
class neuralNetwork:
    
    #initialise(初始化) the neural network
    def __init__(self,inputnodes,hiddennodes,outputnodes,learningrate):

        #set number of nodes(节点) in each input,hidden,output layer
        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes

        # link weight matrices , wih and who
        # weights inside the arrays are w_i_j, where link is from node i to node j in the next layer

        #w11 w21
        #w12 w22 etc
        self.wih = (np.random.rand(self.hnodes,self.inodes)-0.5)
        self.who = (np.random.rand(self.onodes,self.inodes)-0.5)

        #learning rate
        self.lr = learningrate

        # activation(激活) fun is the sigmoid function
        #激活函数命名为了self.activation_function()
        self.activation_function = lambda x : scipy.special.expit(x) #匿名函数，放到__init__中
        
        pass

    #train the neural network
    def train(self , inputs_list , targets_list):
        # convert(转换) inputs list to 2d array
        inputs = np.array(inputs_list,ndmin=2).T#将列表转为numpy数组
        #导入训练集的y真实值
        targets = np.array(targets_list,ndim=2).T#将列表转为numpy数组

        #输入层到隐藏层
        # calculate signals into hidden layer
        hidden_inputs = np.dot(self.wih,inputs)
        # calculate the signals emerging from hidden layer
        hidden_outputs =self.activation_function(hidden_inputs)

        #隐藏层到输出层，最终输出

        # calculate signals into final output layer
        final_inputs = np.dot(self.who,hidden_outputs)
        # calculate the signals emerging from final output layer
        final_outputs =self.activation_function(final_inputs)

        #output layer error is the (target - actual)
        output_errors = targets - final_outputs

        # hidden layer error is the output_errors, split by weights,recombind at hidden nodes
        hidden_errors = np.dot(self.who.T,output_errors)

        # update the weights for the links between the hidden and output layers
        self.who += self.lr * np.dot((output_errors * final_outputs * (1.0 - final_outputs)) , np.transpose(hidden_outputs))

        #同理

        # update the weights for the links between the input and hidden layers
        self.wih += self.lr * np.dot((hidden_errors * hidden_outputs * (1.0 - hidden_outputs)) , np.transpose(inputs))
        
        pass

    #query(询问) the neural network
    def query(self,inputs_list):

        # convert(转换) inputs list to 2d array
        inputs = np.array(inputs_list,ndmin=2).T
    
        #输入层到隐藏层
        # calculate signals into hidden layer
        hidden_inputs = np.dot(self.wih,inputs)
        # calculate the signals emerging from hidden layer
        hidden_outputs =self.activation_function(hidden_inputs)

        #隐藏层到输出层，最终输出

        # calculate signals into final output layer
        final_inputs = np.dot(self.who,hidden_outputs)
        # calculate the signals emerging from final output layer
        final_outputs =self.activation_function(final_inputs)
        
        return final_outputs

posted on 2022-09-25 00:26 zz_BooM 阅读(32) 评论(0) 编辑收藏举报

刷新页面返回顶部

登录后才能查看或发表评论，立即登录或者逛逛博客园首页

相关博文：

· 科学文库网下载书

· 关与python面向对象的认识

· 根据经纬度计算距离半径内网点数量

· pytorch基础学习二

· Python 计算HHI指数

阅读排行：
· 阿里最新开源QwQ-32B，效果媲美deepseek-r1满血版，部署成本又又又降低了！
· 单线程的Redis速度为什么快？
· SQL Server 2025 AI相关能力初探
· AI编程工具终极对决：字节Trae VS Cursor，谁才是开发者新宠？
· 展开说说关于C#中ORM框架的用法！

zz-BooM

公告