#########################################################
##
## kdd data mining competition: 
##   try nearest neighbor logistic regression 
##
## use Weiliang's functions to get the distance matrix
##
#########################################################

source("separation.R")
source("matrix_position.R")


y<-read.table("http://hajek.stat.ubc.ca/~fyuan/kdd/kdd_blocks_weiliang.dat",header=T)
cl<-as.vector(y[,1])
x<-as.matrix(y[,-c(1:3)])
cl.u<-unique(cl)
names(cl.u)<-1:k0
tmpcl<-cl
k0<-length(cl.u)
for(i in 1:k0)
{ tmpcl[cl==cl.u[i]]<-i }

unix.time(tmp<-GetJMatrix(x, tmpcl, th.Jstar=0.1, alpha=0.05))
names(tmp)
tmp$Jmat

tmp2<-tmp$Jmat
diag(tmp2)<-1
min.row.col(tmp2)