### First generate some data as described in Chapter 2

library(MASS)
Ident = matrix(c(1,0,0,1),nrow=2)

m1 = mvrnorm(10,mu=c(1,0),Sigma = Ident)
m2 = mvrnorm(10, mu=c(0,1),Sigma = Ident)


x1 = matrix(0,100,2)
x2=x1

for(i in 1:100){x1[i,]=mvrnorm(1,mu=m1[sample(1:10,1),],Sigma = Ident/5)
		  x2[i,]=mvrnorm(1,mu=m2[sample(1:10,1),],Sigma = Ident/5)}
simmix = rbind(x1,x2)
dim(simmix)

simmix=cbind(simmix,rep(1:2,each=100))
simmix = data.frame(simmix)
plot(simmix[,-3],pch="0", col=c("blue","orange")[simmix[,3]])

## Now the data is in the data frame simmix 
## (200 rows, 3 columns, 3rd column indicates group)

library(nnet)

## following the iris example in the help file carefully, this finally worked

targets = class.ind(simmix[,3])  #

sim.nnet = nnet(simmix[,1:2],targets,size=10,decay=0.02)

# now trying to duplicate Figure 11.4

gridlines = expand.grid(xx1 = seq(min(simmix[,1]),max(simmix[,1]),by=.05), xx2 = seq(min(simmix[,2]),max(simmix[,2]),by=.05))

dim(gridlines)


# expand.grid is a very handy function

fitted = predict(sim.nnet,gridlines)

dim(fitted)

fitted[1:3,]


points(gridlines, pch=".",col=c("blue","orange")[(fitted[,1]<=fitted[,2])+1])


## this plot is for weight decay =0.02, a similar construction with no weight decay
## imitates Figure 11.4 (top)