L2 regularization Negative Binomial

[Marco De Nadai]

Hello,
Since I don't find any information about it, is the L2 regularization implemented in fit_regularized?

Thanks

[josef...@gmail.com]

On Fri, Jan 13, 2017 at 12:41 PM, Marco De Nadai <m...@marcodena.it> wrote:

Hello,
Since I don't find any information about it, is the L2 regularization implemented in fit_regularized?

In the discrete_model like NegativeBinomial fit_regularized is only L1 penalization.

OLS and GLM in master have fit_regularized with elastic net regularization.

So, GLM with negative binomial family would work, however it takes the Negative Binomial dispersion parameter as given and it has to be provided by the user, in contrast to the discrete model that estimates it at the same time as the other parameters.

(This area is still work in progress and a bit stalled at the moment.)

Josef

 

Thanks