R/remCorlowIntensity.R

In interzoneboy/MS-prep: MSprep does basic parsing of sieve data

#' remCor.lowintensity
#'
#' Filter the columns of data frame d such that the less median-intense (or other user
#' specifiable column, correlated at a value higher than cutoff, is removed.
#'
#' @param d Input data frame
#' @param cutoff Correlation threshold at which to do removal
#' @param measure Column measure used to determine column to remove
#' @return a vector of column indices to keep.
#' @export
remCor.lowintensity <- function(d, cutoff=0.8, measure=median) {
	#This function returns indices of columns that should be retained, after dropping correlated columns (that have lower intensity).
	#After getting list of indices that are correlated, 
	# it checks their median values, and drops the one with lower value.
	#You can supply your own measure - maybe mean, sd or var in the function call.


	if(cutoff > 1 || cutoff <= 0) {
		stop("Vasant says cutoff should satisfy 0 < cutoff <= 1")
	}
	if(!is.matrix(d) && !is.data.frame(d)){
		stop("Please supply a data.frame or a matrix.")
	}
	r2cut = sqrt(cutoff)
	cormat <- cor(d)
	measure.index <-apply(d,2,measure) #calculate a measure for all columns
	bad.id <- which(abs(cormat)>r2cut, arr.ind=T);
	bad.id <- matrix(bad.id[bad.id[,1] > bad.id[,2]], ncol=2);
	drop.id <-ifelse(measure.index[bad.id[,1]]<measure.index[bad.id[,2]],bad.id[,1],bad.id[,2]) #check measure for both columns in bad.id and assign to drop.id the one with higher value
	if(length(drop.id)==0){
		1:ncol(d)
	}else{
		(1:ncol(d))[-unique(drop.id)]
	}
}