BP神经网络实现（python）

BP神经网络的计算

最近学习了浙江大学的机器学习课程，又学习了一些内容，用python实现了BP神经网络
代码如下：

#定义sigmoid和它的导数
import numpy as np 
import random,math
def make_matrix(m,n):
    return np.random.random(size=(m,n))
def sigmoid(x):
    return 1.0/(1.0+np.exp(-x))
def simgmoid_derivate(x):
    return x*(1-x)
#定义类
class BPNeuralNetwork:
    def __init__(self):
        #输入层
        self.input_n = 0
        self.hidden_n = 0
        self.output_n = 0
        self.input_cells = []
        self.hidden_cells = []
        self.output_cells = []
        self.input_weights = []
        self.output_weights = []
        self.input_correction = []
        self.output_correction = []
    def set_up(self,ni,nh,no):
        self.input_n= ni
        self.hidden_n = nh
        self.output_n = no
        #初始化cell
        self.input_cells = [1.0] * self.input_n
        self.hidden_cells = [1.0] * self.hidden_n
        self.output_cells = [1.0] * self.output_n
        # 初始化权重
        self.input_weights = make_matrix(self.input_n,self.hidden_n)
        self.output_weights = make_matrix(self.hidden_n,self.output_n)
        
        #激活
        #矫正矩阵
        self.input_correction = make_matrix(self.input_n,self.hidden_n)
        self.output_correction = make_matrix(self.hidden_n,self.output_n)
    #第一次前馈
    def predict(self,inputs):
        for i in range(self.input_n):
            self.input_cells[i]  = inputs[i]
        for j in range(self.hidden_n):
            total = 0.0
            for i  in range(self.input_n):
                total+=self.input_cells[i]* self.input_weights[i][j]
            #若是连续的值需要修改激活函数
            self.hidden_cells[j] = sigmoid(total)
            
        #输出层计算
        for k in range(self.output_n):
            total = 0.0
            for j in range(self.output_n):
                total += self.hidden_cells[k]*self.output_weights[k][j]
            self.output_cells[k] = sigmoid(total)
        return self.output_cells[:]
    #反向传播  学习率λ， 矫正率μ
    def update(self,x,y,learn,correct):
        self.predict(x)
        #计算误差
        output_deltas = [0.0]*self.output_n
        for  o in range(self.output_n):
            error = y[0] -self.output_cells[0]
            #如果是连续值则无需计算
            output_deltas[o] = simgmoid_derivate(self.output_cells[o]) * error
        #计算隐藏层的误差
        hidden_deltas = [0.0] * self.hidden_n
        for h in range(self.hidden_n):
            for o in range(self.output_n):
                change = output_deltas[o] * self.hidden_cells[h]
                #更新权重
                self.output_weights[h][o] += learn * change + correct * self.output_correction[h][o]
                self.output_correction[h][o] = change
        #更新权重
        for i in range(self.input_n):
            for h in range(self.hidden_n):
                change=hidden_deltas[h] * self.input_cells[i]
                self.input_weights[i][h] += learn * change + correct * self.input_correction[i][h]
                self.input_correction[i][h] = change
        #获得全局的误差
        error = 0.0
        for o in range(len(y)):
            error += 0.5 * (y[o] - self.output_cells[o]) ** 2
        return error
    #开始训练设置迭代次数 学习率和矫正率
    def train(self,x,y,limit=1000000,learn =0.05,correct =0.1):
        for m in range(limit):
            error =0.0
            for i in range(len(x)):
                #每次计算一个
                x_item =x[i]
                y_item = y[i]
                error+= self.update(x_item,y_item,learn,correct)

最后测试一下

cases = [[1,1],
        [2,2],
        [2,4],
        [1,2],
        ]
labels = [[0],[0],[1],[1]]
t= BPNeuralNetwork()
t.set_up(2,5,1)
t.train(cases,labels,10000,0.05,0.1)
for case in cases:
    print(t.predict(case))

[0.5285172149874504]
[0.5343332825010273]
[0.5379687945071842]
[0.5336526221969785]
效果并不是很好，可能哪里需要完善。