require(stream)
require(streamMOA)
library(MASS)
library(tictoc)
####
stream<-DSD_Gaussians(k=3,d=2)
dstream<-DSC_DStream(gridsize=.1)
update(dstream,stream, n=1500)
km<-DSC_Kmeans(k=3)
recluster(km,dstream)
plot(km,stream, type="both")
###
# data from scratch
xx<-mvrnorm(500000,mu=c(0,0),Sigma=diag(c(1,1)))
xx<-rbind(xx,mvrnorm(500000,mu=c(3,3),Sigma=diag(c(1,1))))
xx<-rbind(xx,mvrnorm(500000,mu=c(0,5),Sigma=diag(c(1,1))))
xx<-xx[sample(seq(1,dim(xx)[1]),dim(xx)[1]),]
#
dstream<-DSC_DStream(gridsize=.1)
update(dstream,xx[1:500,])
km<-DSC_Kmeans(k=3)
recluster(km,dstream)

plot(km,type="both")
points(xx[1:500,],col=rgb(0,0,0,alpha=0.1))

update(dstream,xx[600:1000,])
recluster(km,dstream)

plot(km,type="both")
points(xx[600:1000,],col=rgb(0,0,0,alpha=0.1))

#### using DSD_memory
# data with drift...
xx<-mvrnorm(1500000,mu=c(0,0),Sigma=diag(c(1,1)))
xx<-rbind(xx,mvrnorm(1500000,mu=c(3,3),Sigma=diag(c(1,1))))
xx<-rbind(xx,mvrnorm(1500000,mu=c(0,5),Sigma=diag(c(1,1))))
xx<-xx[sample(seq(1,dim(xx)[1]),dim(xx)[1]),]
xx[,1]<-xx[,1]+sort(rnorm(seq(1,dim(xx)[1])))
tic()
kk<-kmeans(xx,3)
toc()


stream<-DSD_Memory(xx)
# initialize
tic()
dstream<-DSC_DStream(gridsize=.1,lambda=0.01) #lambda is a fading factor 
# to handle concept drift...
update(dstream,stream,n=50000)
km<-DSC_Kmeans(k=3)
recluster(km,dstream)
plot(km,stream, xlim=c(-6,6),ylim=c(-3,6),type="both")
#
zz<-0
while((stream$state$counter+50000)<=dim(xx)[1]) {
  zz<-zz+1
  update(dstream,stream,n=50000)
  recluster(km,dstream)
  if (round(zz/10)==zz/10) {
    plot(km,stream, xlim=c(-6,6),ylim=c(-3,6), type="both")}
}
recluster(km,dstream)
toc()
kk<-kmeans(xx,3)
points(kk$centers,pch=6,cex=5,col=3)