Date:         Thu, 27 Jan 2000 03:28:24 GMT
Reply-To:     David Dickinson <dickinsn@NOSPAM.URREA.ORG>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David Dickinson <dickinsn@NOSPAM.URREA.ORG>
Organization: Ye 'Ol Disorganized NNTPCache groupie
Subject:      Re: proc mixed and memory (conclusions)
Content-Type: text/plain; charset=us-ascii

Thanks to all who wrote such detailed responses (Dale, Michael, James,
John).

There were two basic ideas about the problem: including mis-specified models
and improper syntax of my repeated statement.  I think that while
mis-specifying my model may be an item for consideration, it is the improper
syntax that was causing the memory drain:

1.     Our "repeated  patno / type=ar(1);" is at best confusing to proc
mixed (and, at least, wrong).  We get the same results with this as with
omitting the repeated statement altogether (with the exception of slight
differences in Schwartz's and Akaike's Criterion).  What I still don't
understand is that if it doesn't do anything, why does SAS let you do it?
We have used it in many other models without causing problems (though,
apparently getting the wrong answers).
2.  The syntax achieving the result we had hoped for would be "repeated
/subject=patno type=ar(1);"
3.  It was suggested that instead of just "subject=patno", we should use
"subject=patno(facid)".  Since I don't see documentation of this method in
the stat manual, I am assuming that it is just telling SAS that unique patno
are grouped within facid, and that this will improve performance.
Predictably, this change without the change in (1) did not solve the memory
drain.
4.  A suggestion was to include "repeated occasion /subject=patno..." where
occasion is a simple counter of the sequence of each measurement for each
patient.  I'm still reading manuals on this, and my limited experience with
mixed is apparent.



David Dickinson wrote:

> I am running into memory problems with proc mixed.  I am running a
> clustered model of repeated measurements, which is otherwise quite
> simple.  I am keeping only the four variables (numeric 8) involved, 9180
> patients with repeated measures for a total of 37000 obs, divided into
> 160 distinct clusters.  No other datasets are in the workspace.
>
> Before, I had gotten an out of memory message explicitly stating that a
> certain number of bytes were being requested and were unavailable,
> stopping the procedure.  So I increased the size of my pagefile to a
> total of 3.6 G of available memory.  Now when I run the proc mixed, I
> see a maximum memory use in the task manager of only 1.3G (about 1/3 of
> the memory used), yet sas stops the proc because of insufficient memory
> (this time, no indication of how much it needs).
>
> proc mixed data=test.all (where=(msr gt .z and cnt gt .z) keep=facid
> patno msr cnt);
>    class facid patno;
>    model msr=cnt / solution;
>    repeated  patno / type=ar(1);
>    random facid / type=cs;
> run;
>
> Any ideas?  Any help is greatly appreciated.
> Thanks
>
> --
> To Reply, Remove NOSPAM from my address
> David Dickinson, URREA

--
To Reply, Remove NOSPAM from my address
David Dickinson, URREA