Feature: Adds MCP penalty as option for regularization of GEE model. #9445
Conversation
|
Hello @josrpowe208! Thanks for opening this PR. We checked the lines you've touched for PEP 8 issues, and found:
|
| @@ -1432,7 +1441,7 @@ def _regularized_covmat(self, mean_params): | |||
|
|
|||
| return ma | |||
|
|
|||
| def fit_regularized(self, pen_wt, scad_param=3.7, maxiter=100, | |||
| def fit_regularized(self, pen_wt, gamma=3.7, method='scad', maxiter=100, | |||
There was a problem hiding this comment.
this breaks backwards compatibility
There was a problem hiding this comment.
I can revert gamma to scad_param for the purpose of backward compatibility. However, both SCAD and MCP use gamma as the parameter in the function, so this should likely be changed in future releases to avoid confusion if both penalty methods are made available in the same fit call.
|
PR looks fine, but I'm not familiar with the details here and need to read up to review this. I worked a lot with SCAD in statsmodels, but never looked closely at MCP. |
NumPy's guide.
Notes:
Added the option for selecting the Minimax Concave Penalty (MCP), Zhang 2010a; DOI: 10.1214/09-AOS729, for fitting a regularized version of the GEE model.