## File: StBr.r
## Description: screens meaningful variables and performns
##              data transformation on clustering 

# rm(list = ls())


## Description: Steinley & Brusco (2006) data transform to cluster 
StBrtransform <- function(X){
  apply(X, 2, function(x) 12 * var(x) / (max(x) - min(x))^2 )  
}


## Description: Clustering index (Steinley & Brusco (2006))
CI <- function(X, B = 1000) { # B : number of boostrap replications

  n <- nrow(X)
  p <- ncol(X)
  
  #ci <- apply(X, 2, function(x) 12 * var(x) / (max(x) - min(x))^2 )
  ci <- StBrtransform(X)
  
  rc    <- ci / min(ci)
  minV  <- which.min(rc)
  
  Xstar    <- scale(X)
  newRange <- apply(Xstar, 2, function(x) max(x) - min(x))
  
  rmin <- newRange[minV]

  datat <- array(0.0, dim = dim(X))
  
  # Reweighting X into datat
  for(i in 1:p){
    v          <- Xstar[, i]
    temp       <- rc[i] * (rmin / newRange[i])^2    
    datat[, i] <- sqrt(temp) * v
  }
  
  xboot  <- matrix(rnorm(n * B), nrow = n)
  #cinorm <- apply(xboot, 2, function(x) 12 * var(x) / (max(x) - min(x))^2 )
  cinorm <- StBrtransform(xboot)
  ci95   <- median(cinorm)
  
  #ciStar <- apply(datat, 2, function(x) 12 * var(x) / (max(x) - min(x))^2 )
  ciStar  <- StBrtransform(datat)
  selectv <- which(ciStar > ci95)
  
  return(list(selectv = selectv, 
              tdata  = datat) )  
}




#test <- matrix(rnorm(200), 40, 5)
#CI(test)$selectv