一个极简的神经元网络助你入门(RELU)

 

import numpy as np
import matplotlib.pyplot as plt

#此示例神经网络的样本输入输出分别是二进制和十进制0-8

#输入样本
X=np.array([
    [0,0,0],
    [0,0,1],
    [0,1,0],
    [0,1,1],
    [1,0,0],
    [1,0,1],
    [1,1,0],
    [1,1,1]    
])

#输出样本
Y=np.array([
    [0],
    [1],
    [2],
    [3],
    [4],
    [5],
    [6],
    [7]
])


#构造神经网络 1个输入层3个神经元 1个输出层1个神经元 激活函数为ReLU
W=np.random.rand(1,3)

print("初始模型随机权重：",W)

def relu(x):
    x[x<0]=0
    return x

def relu_derivative(x):
    return (x>0).astype(int)

learning_rate=0.1
train_times=100
costs=[]

for i in range(train_times):
    for j in range(len(X)):
        y=relu(W.dot(X[j]))
        cost=Y[j]-y
        cost2=cost**2
        costs.append(cost2)
        gradient=-np.dot(cost*relu_derivative(W.dot(X[j])), X[j].reshape(1,3))