In the XGBoost and site-level predictions, taking the PV history from an hour behind and 24 hour lag is useful. Currently the datapipes is more setup for grabbing continuous history, and having a selecting one that can history in larger jumps the further in the past we go could be very helpful.
Context
Its been useful for the tree models and might be useful for other ML models
Possible Implementation
Make one in the select that selects the history more spaced out. This could be exponential drop off as you go further back. Or option to give list of timestamps that is wanted.
Detailed Description
In the XGBoost and site-level predictions, taking the PV history from an hour behind and 24 hour lag is useful. Currently the datapipes is more setup for grabbing continuous history, and having a selecting one that can history in larger jumps the further in the past we go could be very helpful.
Context
Its been useful for the tree models and might be useful for other ML models
Possible Implementation
Make one in the
selectthat selects the history more spaced out. This could be exponential drop off as you go further back. Or option to give list of timestamps that is wanted.