Add support for new __sklearn_tags__

[stes]

Fix #204

sklearn 1.6.0 was released on Dec 9, 24 and introduced a new mechanism for specifying estimator tags (https://scikit-learn.org/dev/developers/develop.html). This PR adopts CEBRA to comply with this new notation. Older sklearn variants will fall back to the more_tags() functions as recommended in this comment.

Indepedently, I spotted a bug in the inheritance order in the CEBRA class, which was fixed now, as described here.

Finally, since the code is now version dependent and there might be users rolling older sklearn version, I extended the test suite by one case checking with a legacy sklearn version (version 1.4.2 which is roughly one year old) -- this will hopefully cover the most important cases. The majority of tests are with sklearn latest (1.6.0 as of Dec 16, 24).

[MMathisLab]

Lgtm; but did not directly test

[Gunnar-Stunnar]

Was this ever fixed, I am using Cebra v0.4.0 and now receiving this error:

[<ipython-input-13-2e6de4decd48>](https://localhost:8080/#) in train(self, neural_session, continous_sessions)
     44 
     45         # fit decoder
---> 46         emb = self.cebra_posOnly_model.transform(nStack_train)
     47         fullKin = trainAllKin(emb, cStack_train)
     48 

[/usr/local/lib/python3.10/dist-packages/sklearn/utils/_set_output.py](https://localhost:8080/#) in wrapped(self, X, *args, **kwargs)
    317     @wraps(f)
    318     def wrapped(self, X, *args, **kwargs):
--> 319         data_to_wrap = f(self, X, *args, **kwargs)
    320         if isinstance(data_to_wrap, tuple):
    321             # only wrap the first output for cross decomposition

[/usr/local/lib/python3.10/dist-packages/cebra/integrations/sklearn/cebra.py](https://localhost:8080/#) in transform(self, X, session_id)
   1224         """
   1225 
-> 1226         sklearn_utils_validation.check_is_fitted(self, "n_features_")
   1227         model, offset = self._select_model(X, session_id)
   1228 

[/usr/local/lib/python3.10/dist-packages/sklearn/utils/validation.py](https://localhost:8080/#) in check_is_fitted(estimator, attributes, msg, all_or_any)
   1749         raise TypeError("%s is not an estimator instance." % (estimator))
   1750 
-> 1751     tags = get_tags(estimator)
   1752 
   1753     if not tags.requires_fit and attributes is None:

[/usr/local/lib/python3.10/dist-packages/sklearn/utils/_tags.py](https://localhost:8080/#) in get_tags(estimator)
    403         for klass in reversed(type(estimator).mro()):
    404             if "__sklearn_tags__" in vars(klass):
--> 405                 sklearn_tags_provider[klass] = klass.__sklearn_tags__(estimator)  # type: ignore[attr-defined]
    406                 class_order.append(klass)
    407             elif "_more_tags" in vars(klass):

[/usr/local/lib/python3.10/dist-packages/sklearn/base.py](https://localhost:8080/#) in __sklearn_tags__(self)
    857 
    858     def __sklearn_tags__(self):
--> 859         tags = super().__sklearn_tags__()
    860         tags.transformer_tags = TransformerTags()
    861         return tags

AttributeError: 'super' object has no attribute '__sklearn_tags__'

This is after installing the latest Cebra package with the scikit-learn v1.6.0.

[Gunnar-Stunnar]

rolling back scikit-learn back to v1.5.2 worked

[stes]

Hi @Gunnar-Stunnar , this was merged after the cebra 0.4.0 release. If you install the latest version from git,

pip install git+https://github.com/AdaptiveMotorControlLab/CEBRA.git

the error should disappear even with sklearn > 1.6.0. In case you give that a try, please let me know if it works!

[Gunnar-Stunnar]

Looks like it worked!

My logs are now being filed with this error:

/usr/local/lib/python3.10/dist-packages/sklearn/utils/deprecation.py:151: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
  warnings.warn(
/usr/local/lib/python3.10/dist-packages/sklearn/utils/deprecation.py:151: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
  warnings.warn(
pos: -9.4474 neg:  15.6202 total:  6.1728 temperature:  0.1000: 100%|██████████| 1000/1000 [00:16<00:00, 61.28it/s]
/usr/local/lib/python3.10/dist-packages/sklearn/utils/deprecation.py:151: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
  warnings.warn(

[juliagorman]

I recently installed CEBRA and am still gettitng the first error. I rolled back to scikit-learn back to v1.5.2 and now am getting this log error described above

[MMathisLab]

You would need to pull from git, but a new version is coming soon!

[juliagorman]

if i re-install from git, will it change the PyTorch version I currently have installed in my conda environment? Should I just make a new environment to install from git?

[stes]

I recently installed CEBRA and am still gettitng the first error. I rolled back to scikit-learn back to v1.5.2 and now am getting this log error described above

are you referring to this log output? This is just a warning message which is safe to ignore, there is no effect with respect to model fitting.

You can configure the warnings package if you want to get rid of the message.

Otherwise, as @MMathisLab noted, we will soon also release a new version of CEBRA properly handling this.