Has "group lasso" been implemented in Stata?

[Bruce Weaver]

Hello everyone. I have been reading up on LASSO recently, and anticipate that when I have a go at it, I will likely have some categorical variables with 3 or more levels to contend with. I gather that "group lasso" is the correct way to deal with that.* Has anyone has implemented "group lasso" for Stata yet? To this point, my searches have not turned up anything.

Also, to those of you who are knowledgeable about LASSO, what do you think about Simon & Tibshirani's (2012) standardized group lasso?

Simon N, Tibshirani R. STANDARDIZATION AND THE GROUP LASSO PENALTY. Stat Sin. 2012 Jul;22(3):983-1001. doi: 10.5705/ss.2011.075. PMID: 26257503; PMCID: PMC4527185.


Cheers,
Bruce


* For example, here is an excerpt from a 2008 article by Meier et al.

Already for the special case in linear regression when not only continuous but also categorical
predictors (factors) are present, the lasso solution is not satisfactory as it only selects individ-
ual dummy variables instead of whole factors. Moreover, the lasso solution depends on how
the dummy variables are encoded. Choosing different contrasts for a categorical predictor will
produce different solutions in general. The group lasso (Yuan and Lin, 2006; Bakin, 1999; Cai,
2001; Antoniadis and Fan, 2001) overcomes these problems by introducing a suitable extension
of the lasso penalty.

[Bruce Weaver]

I contacted the developers of -lassopack- to ask if they have any plans to implement group lasso. Unfortunately, they are not aware of any implementations of group lasso in Stata, nor is it on their current agenda to add it to -lassopack-.

One of them did offer the following tentative suggestion:

I don’t think sklearn has a group lasso implementation, but this discussion

https://stackoverflow.com/questions/...ve-group-lasso

points to what looks like a Python implementation of group lasso that is compatible with sklearn:

https://github.com/mathurinm/celer

and which might therefore be accessible in Stata via pystacked.

But alas, I am not a Pythonista, so it might be easier for me to just become a (temporary) useR if/when I want to use group lasso. Meanwhile, I'll add my wish for group lasso to the Wishlist thread and hope to see it in a future release.