PROC SUMMARY and MULTILABEL formats.

data_null__ — Mon, 13 Jan 2014 17:16:14 GMT

I was trying to see how many 2 way tables I could get out of PROC SUMMARY and I noticed a performance issue with multi-lable formats and the MLF option. I think my test must be flawed but I can't figure a way to get the same performance without MLF even if the formats don't actually have multi-lables but just use the option as in my example below.

17         data sample;
18            array d[1500];
19            do id = 1 to 100; 
20               trt = rantbl(8,.3,.3);
21               do i = 1 to dim(d);
22                  d = rantbl(8,.1,.05,.3,.4);
23                  end;
24               output;
25               end;
26            run;

NOTE: The data set WORK.SAMPLE has 100 observations and 1503 variables.
NOTE: DATA statement used (Total process time):
      real time           0.03 seconds
      user cpu time       0.02 seconds
      system cpu time     0.01 seconds
      

27         proc format;
28            value dgrp(notsorted) 2='Group A' 1='Group C' 5='Group D' 4='Group E' 3='Group B'; 
29            value trt(notsorted)  1='Placebo' 2='Active 1' 3='Active 2'; 
29       !                                                                ;
2                                                          The SAS System                             08:41 Monday, January 13, 2014

30            run;
NOTE: PROCEDURE FORMAT used (Total process time):
      real time           0.00 seconds
      user cpu time       0.00 seconds
      system cpu time     0.00 seconds
      

31         proc summary data=sample chartype completetypes missing;
32            class d: trt / preloadfmt order=data;
33            format d: dgrp. trt trt.; 
34            types trt (d:)*trt;
35            output out=_null_;* / levels ways;
36            run;

NOTE: Multiple concurrent threads will be used to summarize data.
NOTE: There were 100 observations read from the data set WORK.SAMPLE.
NOTE: PROCEDURE SUMMARY used (Total process time):
      real time           30.20 seconds
      user cpu time       29.84 seconds
      system cpu time     0.36 seconds
      memory              186330.71k
      OS Memory           207000.00k
      Timestamp           01/13/2014 08:58:08 AM
      Page Faults                       1
      Page Reclaims                     0
      Page Swaps                        0
      Voluntary Context Switches        136
      Involuntary Context Switches      234
      Block Input Operations            1
      Block Output Operations           0
      

37         
38         proc format;
39            value dgrp(notsorted multilabel) 2='Group A' 1='Group C' 5='Group D' 4='Group E' 3='Group B'; 
40            value trt(notsorted multilabel)  1='Placebo' 2='Active 1' 3='Active 2' /*1,2,3='Total'*/; 
41            run;

NOTE: PROCEDURE FORMAT used (Total process time):
      real time           0.00 seconds
      user cpu time       0.00 seconds
      system cpu time     0.00 seconds
      

42         proc summary data=sample chartype completetypes missing;
43            class d: trt / preloadfmt mlf order=data;
44            format d: dgrp. trt trt.; 
45            types trt (d:)*trt;
46            output out=_null_;* / levels ways;
47            run;

NOTE: Multiple concurrent threads will be used to summarize data.
NOTE: There were 100 observations read from the data set WORK.SAMPLE.
NOTE: PROCEDURE SUMMARY used (Total process time):
      real time           3.42 seconds
      user cpu time       4.14 seconds
      system cpu time     0.50 seconds
      memory              257941.45k
      OS Memory           278520.00k
      Timestamp           01/13/2014 08:58:11 AM
      Page Faults                       0
      Page Reclaims                     0
      Page Swaps                        0
      Voluntary Context Switches        157
      Involuntary Context Switches      43
      Block Input Operations            0
      Block Output Operations           0
      

Re: PROC SUMMARY and MULTILABEL formats.

data_null__ — Tue, 14 Jan 2014 11:53:05 GMT

I was thinking that someone might comment. I guess I should have made it a question.

Re: PROC SUMMARY and MULTILABEL formats.

Tom — Tue, 14 Jan 2014 14:29:51 GMT

What is the question? The second summary appears to use less time and more memory. More memory might be attributed to the MLF option. Less time might just be disk caching of you input data set.

Re: PROC SUMMARY and MULTILABEL formats.

data_null__ — Tue, 14 Jan 2014 17:24:34 GMT

It's not disk caching.

As best I can tell the performance different is related to the use of MLF. I just think it is interesting and wondered if anyone had noticed it before or maybe someone from SAS had an explanation. It does take a lot of variables before the performance different becomes noticeable.

Re: PROC SUMMARY and MULTILABEL formats.

Cynthia_sas — Tue, 14 Jan 2014 20:33:38 GMT

Hi:

I didn't comment because you're not using MULTILABEL and MLF the way they are intended. Your user-defined format does not have any "overlapping" values...so I'm not sure what your test shows. The usual way to specify the MLF would be if you had 5 years in the data (1998 - 2002) and you wanted the year categories to be every year "by itself" and then the 1998-1999 as one category, and the 2000-2002 years as a separate category. So essentially, you are "double counting" each year.:

1998

1999

2000

2001

2002

1998 and 1999

2000 through 2002

You might have the same results, you might not. As we say in the Advanced Programming class -- Your Mileage May Vary.

cynthia

Re: PROC SUMMARY and MULTILABEL formats.

data_null__ — Tue, 14 Jan 2014 20:58:04 GMT

I removed the "true" MULTILABEL from my example to make the two steps equivalent in the ouput they produce, same number of observations from the same number of variables and crossings.

PROC FORMAT and PROC SUMMARY don't seem to care if you used MULTILABEL as it was intended or not.

topic Re: PROC SUMMARY and MULTILABEL formats. in SAS Procedures

PROC SUMMARY and MULTILABEL formats.

Re: PROC SUMMARY and MULTILABEL formats.

Re: PROC SUMMARY and MULTILABEL formats.

Re: PROC SUMMARY and MULTILABEL formats.

Re: PROC SUMMARY and MULTILABEL formats.

Re: PROC SUMMARY and MULTILABEL formats.