Returning a SparseTensor from preprocessing_fn results in ValueError

[joelvogt]

Dear TFT team,

I'm trying to transform records of varying lengths without using a predefined length but receive a value error if index[0] is not set. Using tft.sparse_tensor_left_align works but changes the index. Could you please advise me on how to handle optional features correctly? I created a Jupyter notebook to document my issue:
https://drive.google.com/file/d/1O3yhB2mM28ha8GhbQjOFFgqsYhsZhMAr/view?usp=drivesdk
Thank you very much for your help!

Kind regards,

Joël