topic Re: GLMSELECT Suggestions? in Statistical Procedures

GLMSELECT Suggestions?

NKormanik — Tue, 03 Jul 2018 01:05:08 GMT

So many options, hard to decide what to do. Any bottom-line suggestions?

Tentatively the following looks promising:

/selection=lasso stop=none choose=cvex cvmethod=random (10) modelaverage refit showpvals;

Re: GLMSELECT Suggestions?

PaigeMiller — Tue, 03 Jul 2018 11:10:25 GMT

Use PROC PLS if you have lots of correlated X variables. PLS handles the correlation better than any method using OLS or variable selection methods.

But in response to your exact question, the options in GLMSELECT (and every other procedure) are all to be used in specific situations and not in other sitautions, and you've given us no information about your situation.

Re: GLMSELECT Suggestions?

Ksharp — Tue, 03 Jul 2018 12:31:38 GMT

Agree with Paige.

After running PROC PLS, you could get a table of variable importance ,pick up variables which VIP>0.8

Re: GLMSELECT Suggestions?

NKormanik — Thu, 05 Jul 2018 11:24:15 GMT

Suggestions please....

Before I head over to Proc PLS, I'd like to at least come up with some
results using GLMSELECT. Tried the following and received error:

ERROR: No sample yielded a valid model.

This was the code I used:

proc glmselect data = sas_3.combined_sorted
plot=all;
class i_Day
/ split
;
model i_50503 = i_Open--i_23604
/ selection=stepwise
select=cv
cvmethod=split(10)
cvdetails=all
choose=validate
stop=press
drop=competitive
hierarchy=none
showpvalues
stb
stats=(adjrsq aic aicc ase bic cp fvalue press rsquare sbc sl)
details=all
;
modelaverage
sampling=srs
tables=all
refit;
run;

Re: GLMSELECT Suggestions?

PaigeMiller — Thu, 05 Jul 2018 12:09:51 GMT

ERROR: No sample yielded a valid model.

How can we possibly help? You haven't explained what you are trying to do, and we don't have your data. Even worse, I have very little experience with GLMSELECT and lots of experience with PLS.

Re: GLMSELECT Suggestions?

Ksharp — Thu, 05 Jul 2018 12:51:52 GMT

Can you use the at most succinct code to test your data at first place ?

something like

proc glmselect data = sas_3.combined_sorted;
class i_Day;
model i_50503 = i_Open--i_23604 / selection=stepwise ;

run;

After that ,add some more option you need ,and see which one generate the ERROR information.

Re: GLMSELECT Suggestions?

NKormanik — Thu, 05 Jul 2018 23:56:47 GMT

Thanks Ksharp, and all others here. You folks are wonderful.