[Question] How to know the data and feature preprocessing used in the ensemble?

[TuanDTr]

Hi, the method AutoSklearnClassifier().show_models() displays the models found in the ensemble. I wonder if it is possible to know exactly which data or feature preprocessing steps have been done before training the model. The method show_models only gives only the object:

{'model_id': 2, 
'rank': 1, 
'cost': 0.04255319148936165, 
'ensemble_weight': 0.04, 
'data_preprocessor': <autosklearn.pipeline.components.data_preprocessing.DataPreprocessorChoice object at 0x7f704fb2dee0>, 
'balancing': Balancing(random_state=1), 
'feature_preprocessor': <autosklearn.pipeline.components.feature_preprocessing.FeaturePreprocessorChoice object at 0x7f70a7e4e7f0>,
'classifier': <autosklearn.pipeline.components.classification.ClassifierChoice object at 0x7f70a7e4e1c0>, 
'sklearn_classifier': RandomForestClassifier(max_features=5, n_estimators=512, n_jobs=1,
                       random_state=1, warm_start=True)}

and it is not clear to know which steps they are. Is it possible to get the preprocessing steps in such a case?

Many thanks!

[eddiebergman]

Sorry for the delay,

You could also use estimator.leaderboard(detailed=True) to get a pandas version which gives the str of the choices made. However data_preprocessor encompassing many sub things that are chosen is kind of a pain point.

You can use this most likely:

models = estimator.show_models()
model = ...  # Select one

model_id = model["model_id"]

runhistory = estimator.automl_.runhistory_
full_config = runhistory.ids_config[model_id]

[bonfire666666]

Sorry for the delay,

You could also use estimator.leaderboard(detailed=True) to get a pandas version which gives the str of the choices made. However data_preprocessor encompassing many sub things that are chosen is kind of a pain point.

You can use this most likely:

models = estimator.show_models()
model = ...  # Select one

model_id = model["model_id"]

runhistory = estimator.automl_.runhistory_
full_config = runhistory.ids_config[model_id]

I dont think the model_id matches the id in ids_config