R Language 13 line code implementation Neural network

Reference from 11 lines python
And Wunda Teacher's machine learning.
http://blog.csdn.net/sadfasdgaaaasdfa/article/details/47185199

There are some doubts in some places. For example, the error of reverse transfer back to the demolition is not quite clear.

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
  Err Orhidder <-erroroutput%*% syn1 * L1  * (1-L1)
  syn1 <-syn1-  t (0.01 * t (L1)%*% erroroutput)
  Syn0 <-syn0-t (0.01 * t (x)%*% errorhidder) 
}
  

Run Results

#


Nnetonelayer <-function (x,y,numberofunits = 3,alpha = 0.01,iter=10000) {
  x <-As.matrix (t (x))
  # The number of features of X
  col <-ncol (x)
  #0-1 weight initialization nrow representing the number of features
  syn0 <-matrix (rnorm (numberofunits * col), ncol=col,nrow= numberofunits)
  #1-weight initialization of 2
  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 C19/>errorhidder <-erroroutput%*% syn1 * L1  * (1-L1)

    #
    syn1 <-syn1-  t (Alpha * t (L1)%*% err Oroutput)
    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

real results and predicted results

0 0.01942965
 1 0.95005097
 1 0.96147338
 0 0.06028846

the cost function changes with the iteration