[Question] Rebuilding Auto Sklearn pipelines with the parameter dictionary returned by .cv_results_

[Muhammad-Hassan1000]

Is there a way to rebuild an auto sklearn pipeline after I have imported the params dictionary from the .cv_results into a csv file?

My objective is to get the feature importance of every pipeline auto-sklearn tries. I know some models do not support feature importance so I have used permutation importance for them instead. I did get feature importance for the best pipeline ranked first in the leaderboard but I want to get the importance for every combination of hyperparameter it tries.

What I currently have is this csv file.

So, I'm looking for a method to rebuild the pipeline using these dictionaries.

I'd like to mention one more thing. Is there a direct way while training the AutoSklearnClassifier to access all the pipelines it tried with every hyperparameter combination to achieve my goal? I have looked into the documentation but I have failed to find anything accurate that would help me in this task. The documentation here lists a function .show_models() but the models it shows does not match the models shown in the .leaderboard().

Am I missing some working of auto-sklearn?