Cross Validation and custom pipeline in Autogluon.

Hello,
I have a problem with data leakage and wanted to avoid it so I created a custom pipeline to cleanse the data before train/validation. However, when the cross validation is performed across the train dataset, splitting it into train and val in each fold, does the pair go through the _fit_transform within the pipeline? Is there any extra configuration that I'll have to enable first?

Start by tracing AutoGluon's custom pipeline and cross-validation entry points, focusing on how _fit_transform is invoked for each train/validation fold. Check the relevant pipeline and validation documentation or tests, then establish whether transformations are fit only on each training split and whether any configuration is required; done means the behavior and configuration are clearly documented or reproduced.

Written by the indexing model from the issue text.