SklearnClassifier does not support sklearn-like estimators (e.g. ThunderSVM)

[zacps]

Is your feature request related to a problem? Please describe. Scikitlearn SVMs can be slow on some datasets. GPU accelerated SVMs, such as the one in ThunderSVM, can perform better.

Currently if you pass thundersvm.SVC into SklearnClassifier an error is raised:

  File "...\python\python38\lib\site-packages\art\estimators\classification\scikitlearn.py", line 76, in SklearnClassifier
    raise TypeError("Model is not an sklearn model. Received '%s'" % model.__class__)
TypeError: Model is not an sklearn model. Received '<class 'thundersvm.thundersvmScikit.SVC'>'

Describe the solution you'd like I'd like SklearnClassifier to support sklearn-like classifiers.

Describe alternatives you've considered Specific support for thundersvm could be added.

[zacps]

I've since realised the interface is really only 'like' sklearn's. It might be better to improve the interface of thundersvm before adding support here.

[beat-buesser]

Hi @zacps Thank you very much for sharing this idea for new features! I hadn't thought of it yet, but I think it could be useful in many applications to support scikitlearn-like libraries. In case you continue exploring thundersvm with ART, please let us know.