Fit for AutoSklearn2Classifier fails whens input matrix is in CSR sparse Format

[JohnnyC08]

Describe the bug

When fitting a CSR matrix on the AutoSklearn2Classifier, the method fails and returns Trying to set illegal value 'gradient_boosting'

python version: 3.9
auto-sklearn version: 0.13.0

To Reproduce

Steps to reproduce the behavior:

import sklearn.datasets
from scipy import sparse
from autosklearn.experimental.askl2 import AutoSklearn2Classifier

X, y = sklearn.datasets.load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = \
    sklearn.model_selection.train_test_split(X, y, random_state=1)

X_train = sparse.csr_matrix(X_train)
automl = AutoSklearn2Classifier(
    time_left_for_this_task=120,
    per_run_time_limit=30
)
automl.fit(X_train, y_train, dataset_name='breast_cancer')

Note, the above works fine when using autosklearn.classification.AutoSklearnClassifier.

The use case for a CSR matrix is fitting text classification models after the text has been transformed into a TF-IDF format.

[eddiebergman]

Hi @JohnnyC08, thanks for the bug report and the reproducible code. I can confirm this causes the same error but I can't immediately tell why.

I will address it as soon as I can as sparsity is a frequency culprit of issues, but in the meantime I'm afraid you will have to go with the non-experimental version.

[mfeurer]

We just added support for sparse data via #1245 and it will be available in the upcoming release.