Open ravinkohli opened 3 years ago
Note that this would mean implementing functionalities of Auto-Sklearn into the traditional classification pipeline which might be redundant if there are plans towards merging APT and ASK or enabling running both in sequence (which should be doable since they use the same backend and logging, I think).
To match the performance in the Auto-Pytorch-Tabular, currently the preprocessing in Lucas' implementation is used. Later, to improve or get the same performance we should use the tabular preprocessing pipeline in Auto-PyTorch. This can be done by adding the imputer, encoder, scaler choice modules to the steps of traditional classification pipeline.