BUG - attribute `alpha` is not available for `penalty=SLOPE()` for string representation

[jolars]

Describe the bug

I am getting test errors from skglm (https://github.com/benchopt/benchmark_slope/actions/runs/15496605501/job/43634699241#step:7:471):

self       = <[AttributeError("'SLOPE' object has no attribute 'alpha'") raised in repr()] GeneralizedLinearEstimator object at 0x13454a440>

The problem, I think, is that on the following line you expect alpha to exist:

skglm/skglm/estimators.py

Line 208 in d57e361

self.penalty.alpha))

But for SLOPE it is alphas:

skglm/skglm/penalties/non_separable.py

Line 27 in d57e361

self.alphas = alphas

Steps to reproduce

from skglm import GeneralizedLinearEstimator, SLOPE

from skglm.penalties import SLOPE

alphas = [0.1, 0.5, 1.0, 2.0, 5.0]

model = GeneralizedLinearEstimator(penalty = SLOPE(alphas))

repr(model)

Additional context

[jolars]

I can submit a PR if you tell me what you would prefer would be the solution to this? Should SLOPE use alpha too? That would be a breaking change. Perhaps just rewrite the __repr__ function to either exclude alpha altogether, or to check for it and omit it in case it is missing. Or one could also check for alphas and use, say, the first and last values?

One could also add alpha to SLOPE but keep alphas for backward compatibility, maybe.

[Badr-MOUFAD]

Thanks @jolars for reporting this issue.
I'm in favor of adding an alpha to the SLOPE penalty, setting it to 1.0 and scaling alphas by it.