- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Posted 07-12-2016 06:55 AM
(2025 views)
I was running GLMSELECT adaptive LASSO and obtained the following:
WARNING: The adaptive weights for the LASSO method are not uniquely determined because the full least squares model is singular.
Been searching but it seems no one has posted anything on this. I imagine this is due to formation of singular matrix problem (struggling to understand what that is..).
Does this mean adaptive LASSO is not 'valid'? Would appreciate if anyone can guide me through this.
Thank you in advance.
Saiful.
code:
proc glmselect data=WORK.SD outdesign(addinputvars)=Work.Glmselect_Design
plots=(ASEplot criterionpanel coefficientpanel) seed=123456;
partition fraction(validate=0.3);
class Phase_at_admission Phase_at_SD Lethargy Bleed Multiple_comorbids /
param=glm;
model Outcome=Age_yr_ PR Plt Sr_Creat AST ALT HCO3 Lactate Phase_at_admission
Phase_at_SD Lethargy Bleed Multiple_comorbids AST*ALT HCO3*Lactate
Phase_at_admission*Phase_at_SD Age_yr_*Multiple_comorbids / stb
selection=lasso
(adaptive stop=none choose=sbc) stats=all;
run;
plots=(ASEplot criterionpanel coefficientpanel) seed=123456;
partition fraction(validate=0.3);
class Phase_at_admission Phase_at_SD Lethargy Bleed Multiple_comorbids /
param=glm;
model Outcome=Age_yr_ PR Plt Sr_Creat AST ALT HCO3 Lactate Phase_at_admission
Phase_at_SD Lethargy Bleed Multiple_comorbids AST*ALT HCO3*Lactate
Phase_at_admission*Phase_at_SD Age_yr_*Multiple_comorbids / stb
selection=lasso
(adaptive stop=none choose=sbc) stats=all;
run;
2 REPLIES 2
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Collinearity will more likely occur between CLASS variables when some class combinations are empty (i.e. if all males are right-handed in your data there is no way to estimate the left-handed male effect). If you can't find where the problem is, you can get collinearity diagnosis with proc reg after expanding your design matrix (create dummy variables for class effects) with proc glmmod.
PG
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Not sure, but you could try SELECTION option:
L1=
L1CHOICE=