Quentin Bertrand

The performance of the LassoCV seemed too good to be true. I might have catch a copy problem in the adaptive example: https://github.com/mathurinm/celer/commit/582c74eaeda7547985fbcedcee7da0ff965ddc34 I modified a little bit the example...

Thanks a lot for your answers, I tried on a larger example and I observed some nice speedups! (n=10, p=10_000) Time taken to solve the Lasso optimization problem 0.015 Time...

Since the generation might be time-consuming, we could also try to add a multi-GPU support. I took a look at this script https://github.com/NVlabs/edm/blob/main/generate.py but it is unclear how to do...

Hello @sujay-pandit , Thanks a lot for the incoming PR. My plan would be: - 1 First, add the `sample_weights` to the `Quadratic` class, note that you will also have...

From a user point of view, I would see a function estimator.get_alpha_max(X, y), with ``` def get_alpha_max(X, y): return self.penalty.dual(X.T @ self.fdatafit.rawgrad(y, 0)) ``` To me, this is unclear how...

Quick design question: since almost no code is shared in the altered functions, would it be worth it to add a class `WeightedQuadratic` inheriting from `Quadratic`, and overwriting only the...

Hello, Thanks a lot for your interest! - adding a group Lasso with positive weights should be "easy", we should be able to do it quickly (i.e. this week if...

In the meantime, I will open a separate PR for the positive weights

Do we want a `GroupLasso` estimator [as well](https://github.com/mathurinm/celer/blob/c1b564c3c0c8fc214deb67b57f4a4dd95f66121f/celer/dropin_sklearn.py#L994) ?

Thanks for the report, this seems to show that the `positive=True` feature introduced in https://github.com/scikit-learn-contrib/skglm/commit/1dcbfffa91999c68e23ab76c8b5dc00456766705 was not correct. I will investigate this week.