topic Lasso Logistic Regression using GLMSELECT procedure in SAS Programming

Lasso Logistic Regression using GLMSELECT procedure

akanujia44 — Thu, 29 Aug 2019 13:42:00 GMT

Hi All,

I have been working on a Lasso Logistic regression with binary response and 20 predictor varaibles (a mix of categorical and continuous ) and have read a lot on using GLMSELECT procedure and coding the outcome ±1, and applying a cutoff (usually 0) to the predictions. I also tried using HPGENSELECT procedure and somehow it throws the Format error and no predictor variables are displayed in the results and hence want to try GLMSELECT proc.

My query is what does the line 'applying a cutoff (usually 0) to the predictions' mean? Can anyone throw more light on this approach, like any sample SAS code where this has been applied?

Thank you for your time and support.

Re: Lasso Logistic Regression using GLMSELECT procedure

PaigeMiller — Thu, 29 Aug 2019 15:10:19 GMT

If the outcomes are ±1 then a cutoff of 0 would be on the predicted values used to determine if the regression predicts an observation is a –1 or a +1.

Using binary responses in PROC GLMSELECT is not truly a logistic regression.

PROC HPGENSELECT does have the Lasso for use with logistic regression (and really for use with many generalized linear models).

I also tried using HPGENSELECT procedure and somehow it throws the Format error

We can't help with this minimal explanation. We need to see the log (not just the error message but the entire PROC HPGENSELECT part of the log). Please copy the relevant parts of the log as text, and then paste it into the window that appears when you click on the {i} icon.

Re: Lasso Logistic Regression using GLMSELECT procedure

akanujia44 — Tue, 03 Sep 2019 14:51:30 GMT

Hi,

The SAS code is:

proc hpgenselect data=mybinarydata ;
where dep ne . and treat =a;
class cov1 cov2 cov3 cov4..........c10;
model dep(event='1') = cov1 cov2 cov3 cov4..........c20 / dist=binary;
selection method=Lasso(choose=SBC) details=all;
performance details;

run;

and the SAS log has the following:

NOTE: The HPGENSELECT procedure is executing in single-machine mode.
ERROR: The analytical component failed to load a format.

Log/Listing results for thread 1.
ERROR: The format xyz could not be loaded for variable cov1.
NOTE: The SAS System stopped processing this step because of errors.

the procedure runs, but with this error and it has only intercept in the final results. Not sure what is gone wrong. Any help on this.

Thanks.

Re: Lasso Logistic Regression using GLMSELECT procedure

PaigeMiller — Tue, 03 Sep 2019 15:05:13 GMT

Run PROC CONTENTS on the data set mybinarydata. Is there a format listed for COV1? If so, then you have to make that format available in your code.