kmeansbasic

#Kmeans
i=data.frame(iris$Petal.Length,iris$Sepal.Length)
x=i
c1=kmeans(i,centers=3,nstart=5)
i$class=c1$cluster
plot(i$iris.Petal.Length,i$iris.Sepal.Length,col=i$class,pch=20)
init=sample(3,nrow(i),replace=TRUE) #initial class assignment

plot(x,col=init)
max.iter=20;threshold=.01;n.iter=1
delta=threshold+1;d0=0 
ss=data.frame(matrix(NA,nrow=1,ncol=max.iter))
#Calculate the centre of each subgroup as the average position of all observations is that subgroup
center=t(sapply(unique(init),function(y) colMeans(x[init==y,])))
c0=data.frame(matrix(NA, nrow = nrow(center) , ncol = ncol(center)))
points(center,pch = 10,col=1:3)
while(delta > threshold && n.iter < max.iter && delta!=d0){
  c0=center
  tmp=dist(rbind(center,x))
  tmp=as.matrix(tmp)[-(1:3),1:3]
  k.iter=apply(tmp,1,which.min)       
  center=t(sapply(unique(init),function(y) colMeans(x[k.iter==y,])))
  d0=delta
  delta=norm(center-c0,"2") #2 norm is essentially the euclidean distance
  print(n.iter)
  print(delta)
  n.iter = n.iter+1
  
}
plot(x,col=k.iter)
points(center,pch = 8,col=unique(k.iter))