Sparse group lasso for multiresponse Poisson regression

[canhochoi]

I tried to find but didn't see manual to implement sparse group lasso for multiresponse Poisson regression.

The manual only has an example for group selection in case of Poisson regression.

The Software feature section of the manual only shows for group selection, but not selecting some elements within the selected groups.

I want to ask if sparse group lasso is implemented? If it's not, how to implement it with skscope?

[Mamba413]

@canhochoi ，thanks for your comment.

From your comment, I think you are interested in sparse group lasso for multi-response Poisson regression.

• For multi-response Poisson regression, you can refer to: https://skscope.readthedocs.io/en/latest/gallery/GeneralizedLinearModels/poisson-identity-link.html.

• for sparse group lasso, actually, it includes two sub-components: (i) group selection and (ii) selecting some elements within the selected groups. For (i), we already implement some L0 regularization methods but not lasso. We plan to implement this in future. From our experiments, we expect L0 based methods generally shows superior or competitive performance, compared with group lasso, so you can use these methods for instead. For (ii), sparse group selection can be easily modified with skscope, you can see the algorithm in: http://www.jmlr.org/papers/volume25/23-0653/23-0653.pdf. There also have code for this method: https://github.com/ADSIHT/ADSIHT, which is implemented in R. You can ask @Yanhang129 for detailed usage.