bigcorPar.r

bigcorPar <- function(x, nblocks = 10, verbose = TRUE, ncore="all", ...){
  library(ff, quietly = TRUE)
  require(doMC)
	if(ncore=="all"){
		ncore = multicore:::detectCores()
		registerDoMC(cores = ncore)
	} else{
		registerDoMC(cores = ncore)
	}

	NCOL <- ncol(x)

	## test if ncol(x) %% nblocks gives remainder 0
	if (NCOL %% nblocks != 0){stop("Choose different 'nblocks' so that ncol(x) %% nblocks = 0!")}

	## preallocate square matrix of dimension
	## ncol(x) in 'ff' single format
	corMAT <- ff(vmode = "single", dim = c(NCOL, NCOL))

	## split column numbers into 'nblocks' groups
	SPLIT <- split(1:NCOL, rep(1:nblocks, each = NCOL/nblocks))

	## create all unique combinations of blocks
	COMBS <- expand.grid(1:length(SPLIT), 1:length(SPLIT))
	COMBS <- t(apply(COMBS, 1, sort))
	COMBS <- unique(COMBS)

	## iterate through each block combination, calculate correlation matrix
	## between blocks and store them in the preallocated matrix on both
	## symmetric sides of the diagonal
	results <- foreach(i = 1:nrow(COMBS)) %dopar% {
		COMB <- COMBS[i, ]
		G1 <- SPLIT[[COMB[1]]]
		G2 <- SPLIT[[COMB[2]]]
		if (verbose) cat("Block", COMB[1], "with Block", COMB[2], "\n")
		flush.console()
		COR <- cor(x[, G1], x[, G2], ...)
		corMAT[G1, G2] <- COR
		corMAT[G2, G1] <- t(COR)
		COR <- NULL
	}

	gc()
	return(corMAT)
}

'multicore' has been deprecated. Replaced with parallel

I have the following dimensions in the data. Kindly let me know how many cores should I assign for the calculation as I repeatedly gives me the error Choose different 'nblocks' so that ncol(x) %% nblocks = 0!
dim(data)
[1] 514 26346

Thanks Bob for the code. The only problem is the size limitation of ff matrix is about 45,000. In addition, converting ff matrix to ffdf and then writing to file takes a long time. I made a fork to modify the code to handle ~120,000 columns (unlimited number in theory) and print flatten correlation to a file. I run the script for 24 chromosomes separately and it took about 15 hours to complete using 128 CPUs and 4Gb of memory.

Thanks Bob for the code. The only problem is the size limitation of ff matrix is about 45,000. In addition, converting ff matrix to ffdf and then writing to file takes a long time. I made a fork to modify the code to handle ~120,000 columns (unlimited number in theory) and print flatten correlation to a file. I run the script for 24 chromosomes separately and it took about 15 hours to complete using 128 CPUs and 4Gb of memory.

Hi! can you share your code? It would be very useful, thanks in advance!