Train/test splits for forecasting tasks ?

[hassony2]

Dear authors,

Thank you for this very interesting benchmark contribution.
Currently the splits are programatically defined (with a random split depending on a generator). Would you be able to share the list of train and test ids that were generated by this random_split call ?

Have a great day,

Yana

[hassony2]

In particular, for the forecasting task, it looks like the provided indices [ConvLSTM_logs/lightning_logs/version_8/validation_indices.txt](in https://github.com/kitamoto-lab/benchmarks/blob/1bdbefd7c570cb1bdbdf9e09f9b63f7c22bbdb27/forecasting/ConvLSTM_logs/lightning_logs/version_8/validation_indices.txt) contain some samples (such as 92) with < 24 frames, which I would have expected to be filtered.

Would you be willing to share the list of indices which was used for the results reported in the paper ?

Have a great day,

Yana