实现一个四层的神经网络
import numpy as np
//定义一个激活函数
def sigmoid(x,deriv=False):
if (deriv == True):
return x*(1-x)
return 1/(1+np.exp(-x))
//构造样本
X = np.array([
[1,0,1,0,1,1],
[1,1,1,0,1,1],
[1,0,1,0,0,1],
[1,0,0,0,0,1],
[1,1,1,1,1,1],
[0,0,1,0,1,0],
[0,0,1,1,1,1]
])
//构造标签
y = np.array([
[1],
[0],
[0],
[1],
[1],
[1],
[0]
])
//设定一个种子
np.random.seed(1)
//随机化初始权重值,高斯初始化,或者随机0-1初始化
w0 = 2*np.random.random((6,5))
w1 = 2*np.random.random((5,7))
w2 = 2*np.random.random((7,1))
//构造网络,这里迭代6000次
for i in range(60000):
# 前向传播
L0 = X
L1 = sigmoid(L0.dot(w0))
L2 = sigmoid(L1.dot(w1))
L3 = sigmoid(L2.dot(w2))
#计算错误
L3_error = L3 - y
if(i%10000 == 0): #每10000次打印出来一个error
print("Error: "+np.str(np.mean(np.abs(L3_error))))
#进行反向传播
L3_delta = L3_error * sigmoid(L3,deriv = True)
L2_error = L3_delta.dot(w2.T)
L2_delta = L2_error * sigmoid(L2,deriv = True)
L1_error = L2_delta.dot(w1.T)
L1_delta = L1_error * sigmoid(L1,deriv = True)
L0_error = L1_delta.dot(w0.T)
L0_delta = L0_error * sigmoid(L0,deriv = True)
#更新梯度
w2 -= L2.T.dot(L3_delta)
w1 -= L1.T.dot(L2_delta)
w0 -= L0.T.dot(L1_delta)
print(w2)
print(w1)
print(w0)
结果: