You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently our implementation (encoder_decoder, dataset class and utils) support daily data. To support hourly input data:
The encoder_decoder and all model components remain unchanged. The architecture is completely agnostic to whether T=31 or T=744 (24*31). it just sees a temporal dimension to patch. The only model instantiation change is adjusting patch_size (e.g., patch_size=(24, 4, 4) for hourly) and max_days (e.g., max_days=744).
The dataset class need changes.
The util function add_month_day_dims doesnot work for hourly data. We need another function for hourly data.
Currently our implementation (encoder_decoder, dataset class and utils) support daily data. To support hourly input data:
add_month_day_dimsdoesnot work for hourly data. We need another function for hourly data.