Date: Wed, 8 Oct 2008 17:37:00 -0700
Reply-To: "yoonsup@gmail.com" <yoonsup@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "yoonsup@gmail.com" <yoonsup@GMAIL.COM>
Organization: http://groups.google.com
Subject: Delayed run time leading to out of memory when LSMEANS is added
in proc mixed
Content-Type: text/plain; charset=ISO-8859-1
Hi all,
I'm having a trouble with out of memory warning when I run a series of
models in proc mixed.
Initially, I had total of 24 different models which are only different
in the type of covariance structure
such as CS, FA0(n), FA1(n), FA(n) and UN.
I fitted the 24 models to a data set in the past without problem. I
saved some ods outputs including
covparms, solutionr, InfoCrit, asycov, along with outp from model
statement. At that point, I didn't have
LSMEANS statement and didn't output solution for fixed effects and
lsmeans.
Today I went back to my code and modified it so that I can save
LSMEANS and solution for fixed effects.
So I included lsmeans statement and solution option in model statement
and fitted the models.
After about 4 models, SAS gave me the out of memory warning and
stopped.
So I tried it on different computer and experienced the same problem.
I then did a bit of experiment
on which would cause excessive computation to get the result between
LSmeans or solution for fixed effects
by having only one of the code.
Having looked at run and cpu time in a log file, either of them took
pretty much same time.
I even eliminated some effects that I don't need in LSMEANS but the
result were the same.
But I ran the original code without LSmeans statment and saving
lsmeans and solution for fixed effect,
the analyses seemed to go so smooth, taking way way less time, e.g.
30 seconds without Lsmeans vs 30 minutes with lsmeans.
My guess is that getting solution for fixed effects (3 fixed effect
terms I had --- 5 environemts,
4 reps within each environemtns and 14 incomplete blocks within each
rep and environemtns)
takes quite a lot of memory and it adds up when a series of models are
being analyzed.
Is there any way to fix this problem? I in fact have 51 seperate
datasets to which all 24 models should be fitted.
Thanks.
Yoon
