matlab memory problems

[annib]

Running Matlab2006b on Fedora Core 6 linux on a 64-bit 3.0GHz Intel
Xeon dual-core processors and 8GB RAM and I'm getting an out of Memory
error when performing step 8 permutations on 125 images (~3.4MB each).
What's the largest data set you guys have run so?

thanks
Anna

[Christian Habeck]

Hi anna,

that might be quite a steep memory requirement. I have dealt wich that
many images before, but there were 2MB each, i.e. 79 by 95 by 69 voxels.
Your are quite a bit bigger...


I am sorry I can't offer an immediate solution. Tony is away, but we
could see what could be stripped from the workspace in a reduced version
of the program that only keeps the bare bones information. Otherwise I
only have these 2 'lame' suggestions that both aim at data reduction:

(1) Threshold the images more conservatively
(2) Split the sample in half and have a 'bagged' covariance pattern
derivation. Imagine you have two samples s1 and s2. Derive the substrate
of the allele in each sample to produce a pattern 1 and 2. Then cross
validate and apply pattern 1 to s2 and pattern 2 to s1 and see how well
the correlation holds up. Get a weighted sum of both patterns, according
to the goodness of fit. Obviously this would be complex to beyond just
the point estimate and incorporate inferential statistics. I could help
with that and do this off line if you really haveproblems with the
memory issue.


Bests - Chris

--
Christian Habeck
ch...@columbia.edu
http://www.cumc.columbia.edu/dept/sergievsky/cnd/habeck.html

[annib]

Thanks Chris,
I was really only posting to let you know the limits of the program
really. I think I'll go for the splitting the group in half.

Anna