mathurinm

MNT sklearn 1.2dev breaks estimator fitting

1

Validation of parameters at fit time seems to be an issue, parameters which should be set by the parent class of, e.g., MCPRegressor, are not returned by `get_params()` and so...

Article by Damek Davis and Wotao Yin to solve min f + g + h, all 3 convex (generalizes Froward Backward and Douglas Rachford) also apparently work for non convex...

ENH warn/raiseError in WS solvers if penalty is L05 or L23 for ws_strategy='subdiff'

1

This will cause the solver not to move since 0 is always a critical point

It seems to work very well: https://github.com/scikit-learn/scikit-learn/pull/23507

ENH/BUG? check_estimator has some very adversarial cases for SVC

1

To reproduce, run this in `test_estimators.py`: ```python estimator_name = "SVC" clf = clone(dict_estimators_ours[estimator_name]) clf.verbose = 2 clf.tol = 1e-6 # failure in float32 computation otherwise # if isinstance(clf, WeightedLasso): #...

Probably calls for a non working set solver, or we can work something out with `is_penalized` (that could be called `yields_sparse_coefs` or something, ie if this is not true, we...

ot.solve uses GPU even though tensors are on CPU?

3

## Describe the bug Running `ot.solve` with tensors on the CPU allows memory on the GPU (thisis documented in `get_backend_list`) but also seems to use the GPU, as the Watts...

Improvements for WeightedLasso class

1

- [x] find new name for class as it will handle enet - [x] add support for generic triplet `penalty(w, alpha), prox(w[j], alpha, stepsize), kkt_violation(XtR, alpha)` to handle any penalty...