Date:         Wed, 12 Sep 2001 09:52:55 -0400
Reply-To:     Howard Schreier <howard_schreier@ITA.DOC.GOV>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Howard Schreier <howard_schreier@ITA.DOC.GOV>
Subject:      Re: Proc Summary

Let's consider what this PROC SUMMARY would accomplish.

This can be argued analytically, but here's some simulation code:

   %let nclass = 20;
   %let obscount = 100000;

   data presumm;
   array c(*) c1-c&NCLASS;
   keep       c1-c&NCLASS;
   do n = 1 to &OBSCOUNT;
      do nn = 1 to &NCLASS;
         c(nn) = round(ranuni(1));
         end;
      output;
      end;
   run;

   proc summary data=presumm nway;
   class       c1-c&NCLASS;
   output out=postsumm;
   run;

Basically, this generates binary (zero/one) CLASS variables, uniformly
distributed.

The output from PROC SUMMARY contains 95,308 observations. When I
increase &NCLASS to 25, I get 99,840 observations. At &NCLASS=27, there
are 99,963. Can you spell "asymptote"?

So even if the step could run, it would achieve little or no data
reduction unless the inital data volume were vastly larger, or the value
distributions greatly different, than in this simulation.

So my advice is to skip this step. Just process the uncollapsed data. If
crosstabs over small subsets of the categorical variables will be needed
often, it may be worthwhile to prepare corresponding summarized data
sets.

On Wed, 12 Sep 2001 19:25:20 +1000, Jan Mie Cheah <Janmie.Cheah@SAS.COM>
wrote:

>Dear All,
>
>I would like to know if there are any known limitation to the PROC SUMMARY
>procedure. The situation I encountered is that when I summarized using 32
>class variables and NWAY option(i.e. having a total of 32 variables in the
>class statement of my proc summary), the _TYPE_ will have a value of
>4294967295; however when I used 33 variables (the same 32 variables with
>additional one variable), the _TYPE_ value is
>111111111111111111111111111111111; and again when I used 34
>variables, the _TYPE_ value is
>1111111111111111111111111111111111.
>
>My question is, should this be the case? I understand that when you use the
>NWAY option, PROC SUMMARY/PROC MEANS may encounter insufficient memory to
>the summarization all the class variables. Is this tested and proven? Are
>there any work-arounds for this? What is the maximum number of class
>variables that can be processed?
>
>Thanks in advance!!!
>Janmie