[R] problem with generalized singular value decomposition using LAPACK
Altuna Akalin
altuna.akalin at bccs.uib.no
Sat Sep 22 11:32:32 CEST 2007
Quoting elw at stderr.org:
>
> > I'm trying to run generalized singular value decomposition (GSVD)
> > function from LAPACK library. Basically my problem is that I can not run
> > it for large matrices, I get a memory error. I'm using R 2.5.1. I tried
> > this on intel centos5 machines with 2 GB memory and 8 GB memory. I have
> > unlimited max memory,cpu time and virtual memory.
> >
> >> res=GSVD( matrix(1:35000,5000,6), matrix(1:35000,5000,6) ) #runs
> >> without
> > problems
> >> res=GSVD( matrix(1:36000,6000,6), matrix(1:36000,6000,6) )
> > Error: cannot allocate vector of size 274.7 Mb
>
>
> Did you examine R's internal memory limits, e.g. ?memory for help...
>
> that should help. [You should, most likely, be able to allocate a ~274MB
> vector on a 2GB machine... but you might have to coerce R into taking that
> much memory from the system, rather than trying to be conservative.]
>
> --elijah
>
Hi Elijah,
Thanks for your reply. I checked internal memory limits and it seems there is no
maxima set for consecutive cells or vector heap so i guess only limit is machine
resources. Also I set the "--max-ppsize" to the maximum value allowed with the
hope that it might help, it didn't change, still can not allocate memory. Below
I pasted ulimit -a output, maybe it helps for diagnostics. Further more, I
changed the stack size to "unlimited", that didn't help either.
core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
max nice (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 32759
max locked memory (kbytes, -l) 32
max memory size (kbytes, -m) unlimited
open files (-n) 1024
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
max rt priority (-r) 0
stack size (kbytes, -s) 10240
cpu time (seconds, -t) unlimited
max user processes (-u) 32759
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
cheers
altuna
More information about the R-help
mailing list