For training with multiple timestep sequences, both data and labels should be series. Currently, when num_steps for PrimateReaching dataset is greater than 1, the data is a series but the label is still only the last timestep label. We should add label series functionality.
For training with multiple timestep sequences, both data and labels should be series. Currently, when num_steps for PrimateReaching dataset is greater than 1, the data is a series but the label is still only the last timestep label. We should add label series functionality.