When creating a new attribute query, it's crucial to apply the appropriate filters.
This can help avoid data leakage, where the model might train on future data that will be unavailable in real-world scenarios.
Pecan assists in preventing this leakage by implementing a date filter. This supplies the model with entity data only prior to the Marker (the prediction date for each entity).
How to apply the date filter?
If your attribute query includes a date column, activate the "multiple rows per entity" toggle in the form of "Add attribute". Then, specify the relevant Date column and the historical duration for training.
By following this method, you're ensuring that the model is trained with pertinent data, eliminating the risk of utilizing future information.