mathurinm/celer

ENH add weights to GroupLasso to provide support for an Adaptive GroupLasso

sehoff opened this issue · 5 comments

sehoff commented

Hi,
thanks for this great package! Does the group-lasso method allow for different penalizations for different groups, such that one could fit an adaptive group lasso, i.e., applying different weights in the second call to the method? (reference for adapative group lasso: https://www.sciencedirect.com/science/article/abs/pii/S0167947308002582)

If not, would it be possible to extend the method (easily)?

Thank you.

mathurinm commented

Hi !
Thanks for the interest. The feature is not currently supported, but it's quite easy to implement a weights attribute that can be changed after each fit ad affects the strength of each penalty (as is currently done in the Lasso class). I should be able to implement it next week

mathurinm commented

(as long as you don't need some of these weights to be 0, because celer won't support that, see #185)

sehoff commented

I really appreciate your fast answer and implementation. All my weights will be positive!

cruyffturn commented

Hi ! Thanks for the interest. The feature is not currently supported, but it's quite easy to implement a weights attribute that can be changed after each fit ad affects the strength of each penalty (as is currently done in the Lasso class). I should be able to implement it next week

I thought the celer_path allows group weights as it's indicated in here.

mathurinm commented

@cruyffturn thanks for pointing this out; it is a mistake in the docstring as weights are not passed to the solver inside the loop over alphas here: https://github.com/mathurinm/celer/blob/main/celer/homotopy.py#L296
#223 will fix this soon.