It would be nice if the TimeSeriesDataset allowed for using a different strategy to fill missing timesteps aside from fill forward.
E.g. a good addition would be linear fill between the two boundary timesteps.

Thanks!