Date: Tue, 24 Jun 2008 16:40:00 +0200 (CEST)
From: lgautier at altern.org
Subject: Re: [Bioc-devel] normalize.loess effizient method
To: schmidb at ibe.med.uni-muenchen.de
Cc: bioc-devel at stat.math.ethz.ch
Message-ID: <57511.90.27.82.241.1214318400.squirrel at www.altern.org>
Content-Type: text/plain;charset=iso-8859-1
Thanks a lot Markus.
To facilitate our work (review and inclusion), could you directly post a
patch for your changes ?
(Something like svn diff > fixnormalizeloess.patch should make it).
Thanks,
L.
Hello,
I parallelized the normalize loess function (see affyPara package).
Thereby I found a memory inefficient implementation of normalize.loess.
The matrices fs and newdata are of the size of the complete intensity
matrix and not necessary. I removed this two matrices and the oldfs
matrix. Attached the changed code.
Best
Markus
normalize.loess <- function(mat, subset=sample(1:(dim(mat)[1]),
min(c(5000, nrow(mat)))),
epsilon=10^-2, maxit=1, log.it=TRUE, verbose=TRUE, span=2/3,
family.loess="symmetric"){
J <- dim(mat)[2]
II <- dim(mat)[1]
if(log.it)
mat <- log2(mat)
change <- epsilon +1
iter <- 0
w <- c(0, rep(1,length(subset)), 0) ##this way we give 0 weight to the
##extremes added so that we can interpolate
while(iter < maxit){
iter <- iter + 1
means <- matrix(0,II,J) ##contains temp of what we substract
for (j in 1:(J-1)){
for (k in (j+1):J){
y <- mat[,j] - mat[,k]
x <- (mat[,j] + mat[,k]) / 2
index <- c(order(x)[1], subset, order(-x)[1])
##put endpoints in so we can interpolate
xx <- x[index]
yy <- y[index]
aux <-loess(yy~xx, span=span, degree=1, weights=w,
family=family.loess)
aux <- predict(aux, data.frame(xx=x)) / J
means[, j] <- means[, j] + aux
means[, k] <- means[, k] - aux
if (verbose)
cat("Done with",j,"vs",k," in iteration ",iter,"\n")
}
}
mat <- mat - means
change <- max(colMeans((means[subset,])^2))
if(verbose)
cat(iter, change,"\n")
}
if ((change > epsilon) & (maxit > 1))
warning(paste("No convergence after", maxit, "iterations.\n"))
if(log.it) {
return(2^mat)
} else
return(mat)
}
--
Dipl.-Tech. Math. Markus Schmidberger
Ludwig-Maximilians-Universit?t M?nchen
IBE - Institut f?r medizinische Informationsverarbeitung,
Biometrie und Epidemiologie
Marchioninistr. 15, D-81377 Muenchen
URL: http://ibe.web.med.uni-muenchen.de
Mail: Markus.Schmidberger [at] ibe.med.uni-muenchen.de