topic Re: Partitioning Data in SAS Procedures

Partitioning Data

mahler_ji — Mon, 24 Nov 2014 12:33:42 GMT

Hello All,

Looking for a quick way to randomly partition a dataset into three different subsets for model training, testing and validation. I would like to be able to vary the sizes of each set (50% test, 30% training, etc etc) and make sure that the sets are randomly generated.

Thanks!

John

Re: Partitioning Data

Ksharp — Mon, 24 Nov 2014 13:12:16 GMT

Or you should check proc surveyselect ;

data shoes;
 set sashelp.shoes;
 r=ranuni(-1);
run;
proc rank data=shoes out=have groups=100;
 var r;
 ranks rank;
run;
data test training valid;
 set have;
 select;
  when(rank lt 50) output test;
  when(rank lt 80) output training;
  otherwise output valid;
 end;
run;

Xia Keshan

Re: Partitioning Data

slchen — Mon, 24 Nov 2014 14:12:44 GMT

also try to use proc surveryselect

Re: Partitioning Data

stat_sas — Mon, 24 Nov 2014 15:22:35 GMT

data test training valid;

set have;

if ranuni(2345)<=0.5 then output test;

else if ranuni(2345)>0.5 and ranuni(2345)<=0.8 then output training;

else output valid;

run;