How and when are timestamps used in Kumo?

When connecting tables with stable attributes about entities of interest, a column may be designated as the create date (representing the entity's creation date). For tables representing events corresponding to the previously mentioned table types, a timestamp column _must _exist to represent when the information in the row becomes known and true (i.e., when the event happened).

Timestamps are also used by the Kumo platform in several areas to perform feature selection and transformations to avoid data leakage.

For temporal pQueries (e.g., tasks that predict some type of aggregation of values over time), Kumo will generate a set of training/validation/test splits during training that cover the time range of the target table. Timestamps can also be used at batch prediction time to implement temporal filtering (i.e., setting a custom prediction anchor time). This is configured in a batch prediction job's settings—under "Prediction Anchor Time", you can set an optional starting date for your predictions in ISO 8601 format (e.g., 2024-02-27).