R语言神经网络

library(nnet)

  

 

  
1
 

Warning message:
"package 'nnet' was built under R version 3.5.3"

  

 

  
1
  
2
 

x1 <- round(runif(2000,1,2000))
x2 <- round(runif(2000,1,2000))

  

 

  
1
  
2
 

x11 <- scale(x1[1:1900])
x12 <- scale(x2[1:1900])
x21 <- scale(x1[1901:2000])
x22 <- scale(x2[1901:2000])

  

 

  
1
  
2
  
3
  
4
 

y1 <- x11^2 +x12^2

  

 

  
1
 

y2 <- x21^2+ x22^2
train<-cbind(x11,x12,y1)
test <- cbind(x21,x22,y2)

  

 

  
1
  
2
  
3
 

nn <- nnet(train[,3]~.,train[,-3],size=6,decay = 0.01,maxit = 1000,linout = T)

  

 

  
1
 

# weights:  25
initial  value 10075.565825 
iter  10 value 2340.410983
iter  20 value 592.177348
iter  30 value 219.730823
iter  40 value 52.509799
iter  50 value 19.285671
iter  60 value 8.751054
iter  70 value 7.858170
iter  80 value 7.029086
iter  90 value 5.948757
iter 100 value 5.687481
iter 110 value 5.666755
iter 120 value 5.664340
iter 130 value 5.622939
iter 140 value 5.497017
iter 150 value 5.446309
iter 160 value 5.444145
iter 170 value 5.439014
iter 180 value 5.353202
iter 190 value 5.228943
iter 200 value 5.149531
iter 210 value 5.070757
iter 220 value 5.049548
iter 230 value 5.014714
iter 240 value 4.950743
iter 250 value 4.943060
iter 260 value 4.942651
final  value 4.942645 
converged

  

 

  
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
 

nn

  

 

  
1
 

a 2-6-1 network with 25 weights
inputs: V1 V2 
output(s): train[, 3] 
options were - linear output units  decay=0.01

  

 

  
1
  
2
  
3
  
4
 

pp <- predict(nn,test)

  

 

  
1
 

plot(pp,y2)

  

 

  
1
 

# AMORE实现BP网络
library('AMORE')

  

 

  
1
  
2
 

fivenum(x11)

  

 

  
1
 

1. -1.76085794812317
2. -0.862405626815202
3. 0.0506626652700281
4. 0.84938248009788
5. 1.66959636959817

fivenum(x12)

  

 

  
1
 

1. -1.74190485072
2. -0.867376734891097
3. 0.025316653373426
4. 0.866109263250477
5. 1.71468698988565

net <- newff(n.neurons=c(2,5,1),learning.rate.global=1e-3, momentum.global=0.4,error.criterium='LMS', Stao=NA,hidden.layer="tansig",output.layer="purelin",method="ADAPTgdwm")

  

 

  
1
  
2
  
3
 

res<- train(net,train[,-3],y1,error.criterium = "LMS", report = TRUE,show.step = 100,n.show=10)

  

 

  
1
  
2
 

index.show: 1 LMS 0.0340939331074392 
index.show: 2 LMS 0.0141267812308106 
index.show: 3 LMS 0.00582576799493561 
index.show: 4 LMS 0.00277297232823935 
index.show: 5 LMS 0.00136366703334354 
index.show: 6 LMS 0.000715656665439051 
index.show: 7 LMS 0.000427191082734316 
index.show: 8 LMS 0.000302547940536113 
index.show: 9 LMS 0.000249249579120589 
index.show: 10 LMS 0.00022584996446879 

  

 

  
1
  
2
  
3
  
4
  
5
  
6
  
7
  
8
  
9
  
10
 

x<- sim(res$net,test[,-3])

  

 

  
1
 

library(rgl)

  

 

  
1
 

plot3d(x11,x12,y1,col="blue")
plot3d(x21,x22,z,col="blue")

  

 

  
1
  
2
 

文章来源: maoli.blog.csdn.net，作者：刘润森！，版权归原作者所有，如需转载，请联系作者。

原文链接：maoli.blog.csdn.net/article/details/97945039