MATLAB神经网络（1）之R练习

MATLAB神经网络（1）之R练习

将在MATLAB神经网络中学到的知识用R进行适当地重构，再写一遍，一方面可以加深理解和记忆，另一方面练习R，比较R和MATLAB的不同。
如要在R中使用之前的数据，应首先在MATLAB中用writetable函数将原本的由mat文件读入的数据写到csv文件中,以备R读入。

writetable(T,filename) writes to a file with the name and extension specified by filename.

writetable determines the file format based on the specified extension. The extension must be one of the following:

1. .txt, .dat, or .csv for delimited text files
2. .xls, .xlsm, or .xlsx for Excel® spreadsheet files
3. .xlsb for Excel spreadsheet files supported on systems with Excel for Windows® See doc writetable.

writetable(table(c1),"data1.csv");
writetable(table(c2),"data2.csv");
writetable(table(c3),"data3.csv");
writetable(table(c4),"data4.csv");

---

 这里我们使用R中十分经典的鸢尾花数据集iris（在dplyr包中）。

library(dplyr)

## Warning: package 'dplyr' was built under R version 3.5.3

##
## Attaching package: 'dplyr'

## The following objects are masked from 'package:stats':
##
## filter, lag

## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union

dim(iris)

## [1] 150 5

str(iris)

## 'data.frame': 150 obs. of 5 variables:
## Sepal.Length: num 5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
## Sepal.Width : num 3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
## Petal.Length: num 1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
## Petal.Width : num 0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
## Species : Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...

可以看到该数据集共有150组数据，4个自变量，1个因变量（factor），鸢尾花有3类。

sort: sort a vector or factor (partially) into ascending or descending order.

order: returns a permutation which rearranges its first argument into ascending or descending order, breaking ties by further arguments.

k<-rnorm(150)
n<-order(k)
input<-iris[,1:4]
output1<-iris[,5]
output1<-as.integer(output1)
output<-matrix(0,150,3)
output<-as.data.frame(output)
#把输出从1维变成3维
for(i in 1:150)
{
if(output1[i]==1)
output[i,1]=1
else if(output1[i]==2)
output[i,2]=1
else
output[i,3]=1
}
input_train=input[n[1:120],]
output_train=output[n[1:120],]
input_test=input[n[121:150],]
input_test=input[n[121:150],]
me<-apply(input_train,2,mean)
va<-apply(input_train,2,var)
me

## Sepal.Length Sepal.Width Petal.Length Petal.Width
## 5.875000 3.030000 3.849167 1.239167

va

## Sepal.Length Sepal.Width Petal.Length Petal.Width
## 0.6707983 0.1876639 3.0223522 0.5622346

#输入数据归一化
inputn<-scale(input_train)

下面进行神经网络初始化。

结构4-5-3

innum<-4
midnum<-5
outnum<-3
#权值初始化
w1<-matrix(rnorm(innum*midnum),midnum,innum)
b1<-rnorm(midnum)
w2<-matrix(rnorm(outnum*midnum),midnum,outnum)
b2<-rnorm(outnum)
#学习率
xite<-0.1
loopNumber<-50
I<-rep(0,midnum)
Iout<-rep(0,midnum)
FI<-rep(0,midnum)
dw1<-matrix(0,innum,midnum)
db1<-rep(0,midnum)

神经网络训练

E<-rep(0,loopNumber)
for(ii in 1:loopNumber)
{
for(i in 1:120)
{
x=inputn[i,]
for(j in 1:midnum)
{
I[j]<-sum(inputn[i,]*w1[j,])+b1[j]
Iout[j]<-1/(1+exp(-I[j]))
}
yn<-t(w2)%*%Iout+b2

e<-output_train[i,]-yn
E[ii]<-E[ii]+sum(abs(e))

dw2<-t(e)%*%Iout
db2<-e

for(j in 1:midnum)
{
S<-1/(1+exp(-I[j]));
FI[j]<-S*(1-S);
}

for(k in 1:innum)
{
for(j in 1:midnum)
{
dw1[k,j]<-FI[j]*x[k]*sum(e*w2[j,])
db1[j]<-FI[j]*sum(e*w2[j,])
}
}

w1<-w1+xite*t(dw1)
b1<-b1+xite*t(db1)
w2<-w2+xite*t(dw2)
b2<-b2+xite*t(db2)
}
}

分类预测

inputn_test<-input_test
for(i in 1:120)
{
inputn_test[i,]<-(input_test[i,]-me)/va^0.5
}
fore=matrix(0,3,30);
for(i in 1:30)
{
for(j in 1:midnum)
{
I[j]=sum(inputn_test[i,]*w1[j,])+b1[j]
I<-unlist(I)
Iout[j]=1/(1+exp(-I[j]))
}
fore[,i]=t(w2)%*%Iout+b2
}

结果分析

output_fore=rep(0,30)
for(i in 1:30)
{
output_fore[i]<-which.max(fore[,i])
}
error=output_fore-output1[n[121:150]]
t<-table(output_fore,output1[n[121:150]])
t

##
## output_fore 1 2 3
## 1 13 0 0
## 2 0 9 0
## 3 0 0 8

#正确率
options(digits=3)
rightridio<-(t[1,1]+t[2,2]+t[3,3])/30
result<-paste("正确率是 ",round(rightridio*100,digits=3),"%")
result

## [1] "正确率是 100 %"