How to see the selected features when ensemble size = 1?

[zgm0407]

Hi,
I applied the codes in #719 and #524. It worked well when the ensemble size was 50. However, there was some thing wrong when the ensemble size was 1.

automl = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=120,
    include_preprocessors=['select_percentile_classification'],
    include_estimators=['random forest],
    ensemble_size=1,
    resampling_strategy='cv',
    resampling_strategy_arguments={'folds':5},
)

automl.fit(X_train,y_train)

for weight, model in automl.get_models_with_weights():
    print(tuple(model.steps[-2]))
    print(tuple(model.steps[-2])[-1].choice.preprocessor.scores_)
    print(tuple(model.steps[-2])[-1].choice.preprocessor.percentile)


('feature_preprocessor', <autosklearn.pipeline.components.feature_preprocessing.FeaturePreprocessorChoice object at 0x7ff0e3123630>)



AttributeError                            Traceback (most recent call last)
<ipython-input-154-6172a5b19273> in <module>
      5     print(tuple(model.steps[-2]))
      6     # Obtain the scores of the current feature selector
----> 7     print(tuple(model.steps[-2])[-1].choice.preprocessor.scores_)
      8     # Obtain the percentile configured by Auto-sklearn
      9     print(tuple(model.steps[-2])[-1].choice.preprocessor.percentile)

AttributeError: 'SelectPercentileClassification' object has no attribute 'preprocessor'

[mfeurer]

From a brief check the issue is rather the examples don't work with cross-validation. Could you kindly check if the error goes away if you disable cross-validation?

[zgm0407]

Yeah that can fix the problem. It works out if I choose hold-out strategy.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs for the next 7 days. Thank you for your contributions.

[eddiebergman]

Update:

We should document that when using CV, the models returned are actually VotingClassifier/Regressor which means that issue #719 #524 don't work as expected.