R语言13行代码实现神经网络

参考自11行python
以及吴恩达老师的机器学习
http://blog.csdn.net/sadfasdgaaaasdfa/article/details/47185199

还是有些地方有些疑惑。例如误差反向传递回去的推倒还不太明白。

x <- data.frame(c(0,0,1),c(0,1,1),c(1,0,1),c(1,1,1))y <- c(0,1,1,0)x <- as.matrix(t(x))syn0 <- matrix(rnorm(3 * col ),ncol=3,nrow=3)syn1 <- matrix(rnorm(3 * 1 ),ncol=3,nrow=1)for (i in 1:10000){  l1 <- 1/(1+exp(-(x %*% t(syn0) )))  l2 <- 1/(1+exp(-(l1)  %*% t(syn1) ))  errorOutput <- l2 - y  errorHidder <- errorOutput %*% syn1 * l1  * (1-l1 )  syn1 <- syn1 -  t(0.01 * t(l1) %*% errorOutput )  syn0 <- syn0 - t(0.01 * t(x) %*% errorHidder) }
1
2
3
4
5
6
7
8
9
10
11
12
13

运行结果

#nnetOneLayer <- function(x,y,numberofUnits = 3,alpha = 0.01,iter=10000){  x <- as.matrix(t(x))  # x的特征数量  col <- ncol(x)  #0-1的权重初始化 nrow 表示特征数  syn0 <- matrix(rnorm(numberofUnits * col ),ncol=col,nrow=numberofUnits)  #1-2的权重初始化  syn1 <- matrix(rnorm(numberofUnits * 1 ),ncol=(numberofUnits),nrow=1)  cost <- c(NA)  length(cost) <- iter  for (i in 1:iter){    #输入层到隐层的值    l1 <- 1/(1+exp(-(x %*% t(syn0) )))    #隐层到输出层的值    l2 <- 1/(1+exp(-(l1)  %*% t(syn1) ))    #输出层误差    errorOutput <- l2 - y    #隐层误差    errorHidder <- errorOutput %*% syn1 * l1  * (1-l1 )    #    syn1 <- syn1 -  t(alpha * t(l1) %*% errorOutput )    syn0 <- syn0 - t(alpha * t(x) %*% errorHidder)     cost[i] <- errorOutput  }  plot(cost)  cbind(y,l2)}x <- data.frame(c(0,0,1),c(0,1,1),c(1,0,1),c(1,1,1))y <- c(0,1,1,0)nnetOneLayer(x,y)l2
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

真实结果与预测结果

 0 0.01942965 1 0.95005097 1 0.96147338 0 0.06028846
1
2
3
4

cost函数随着迭代的变化