Here is some code, intended to take a dataset, apply a marker variable called "selected" which identifies records lacking a sale price as test data, and then use the rest of the data to create a model to predict saleprice values. This much seems to work. I also would like the procedure to use the test data to perform a test on the model thus created, but this does not seem to work, and I can't figure out why. data dt; set miout2 (obs=1670); if saleprice=. then selected=0; else selected=1; run; ods graphics on; proc glmselect data=dt plots=all; partition rolevar=selected(test='0' train='1'); class [lots of variables]; model saleprice = [even more variables] / selection =lar(choose=cv stop=none) cvmethod=random(10) details=all; run;
... View more