[BioC] memory problem with fitPLM in package affyPLM

Ben Bolstad bolstad at stat.berkeley.edu
Thu Jun 9 20:06:02 CEST 2005

Answers interpolated below.

On Thu, 2005-06-09 at 10:48 -0700, fhong at salk.edu wrote:
> Thanks, Ben. That helps a lot! But I still have some questions? Would you
> please also help me on this.
> > There were significant changes in the structure of the PLMset object
> > between 1.2.x and 1.3.x which is why you are having problems with the
> > boxplot(), Mbox() commands on your old PLMset using the new code.
> But why when I reload in to R 2.0.1 ( the on ei used to generate PLMset
> object), and tried boxplot ( suppose to produce NUSE plot), it gave me
> something strange (see attachment)

try something like

boxplot(Pset,ylim=c(0.9, 1.2))

though I am not really too sure why you have such extreme outliers on
your plot.

> >
> > Also, if you can live without the weights (or alternatively the
> > residuals) you could do
> >
> > Pset <- fitPLM(my.Data,output.param=list(varcov="none",weights=FALSE))
> >
> >
> > or
> >
> > Pset <- fitPLM(my.Data,output.param=list(varcov="none",residuals=FALSE))
> >
> > which would also reduce the memory overhead.
> Will those simplificaiton change the underlying model that is fitted to
> the data. e.g., weights=FALSE doesn't this mean it won't use iteratively
> reweighted least squares (IRLS)?

No the fitting procedure will be unchanged, ie it still uses IRLS. All
it means is that the weights aren't kept around after they have been
used. Otherwise given that there is a weight for every PM probe a lot of
memory gets used up.

More information about the Bioconductor mailing list