Closed slawekslex closed 4 years ago
What is your use case? I think they should be passed self.dataset for the setup but want to be sure that's what you need.
Getting the dataset sounds good.
Just fixed this, the setup gets passed the full DataLoader
in the end. Pipelines are setup in the logical order: after_item, before_batch and after_batch.
The Transform.setup() is called before the TfmdDL super constructor here As the result the transform wont have access to the data. See for example: