Fix: Prevent AttributeError when checking for UnifiedSolver in transform function

[anandawolz]

The previous code accessed cebra.solver.UnifiedSolver directly in an isinstance() check.
In CEBRA versions where UnifiedSolver is not exposed as an attribute (e.g. sklearn models) of cebra.solver, this caused an AttributeError before the check could fall through to the next condition.

Change: Removed the direct attribute lookup of UnifiedSolver to prevent AttributeError in environments where UnifiedSolver is unavailable.

• UnifiedSolver check is replaced by BaseSolver, which is already imported in a stable manner from cebra.solver.base.
• The function now distinguishes between:

1. BaseSolver models (expects data and label)
2. SklearnCEBRA models (expects only data)

[CeliaBenquet]

same here, just use solver and cebra

[CeliaBenquet]

You can simplify naming here, just import them as they are called, no renaming :)

[CeliaBenquet]

lgtm! Just suggestion to remove the prints if not used.

[CeliaBenquet]

Mm actually @anandawolz let's discuss the change, it is not correct as not all solvers have the labels in the transformer (only Unified).

[CeliaBenquet]

@anandawolz, regarding my last comment, how did you handle it?

the def transform() in that wrapper takes a label, but that will not work for other solvers because they don't take a label. Have you tested that?

[MMathisLab]

@CeliaBenquet can you finalize this PR and merge?

[CeliaBenquet]

@anandawolz, regarding my last comment, how did you handle it?

the def transform() in that wrapper takes a label, but that will not work for other solvers because they don't take a label. Have you tested that?

this was addressed here: 08ec240