How to create pipelines from AutoSKlearn output configurations?

[alinedelvalle]

Hello,
I use AutoSKlearn to find a pipeline with a multi-label classifier (ensemble size is 1). I have the pipeline configuration as the response. I would like to use this configuration to train pipelines with different training and testing sets. I used the fit_pipeline method passing the training and testing data, but regardless of the data I pass, the results are always the same. Is it correct to use fit_pipeline? If so, what could be the error?

[simonprovost]

Hi there!

I am far from maintaining Auto-Sklearn nor part of the Auto-Sklearn team, but I was passing by and hope this might help.

I believe you are looking for refit following the discovery of your champion model/ensemble: https://automl.github.io/auto-sklearn/master/api.html?highlight=refit#autosklearn.classification.AutoSklearnClassifier.refit

Hope this helps,

Cheers