dataframe
dataframe
¶
DataFrame-stage checks that inspect the training split's columns.
Classes:
| Name | Description |
|---|---|
DatasetSizeCheck |
Block training when the training split is unusably small. |
GroupbyColumnCheck |
Validate group-by column existence and integrity. |
OrderbyColumnCheck |
Validate order-by column existence. |
PseudoColumnCheck |
Detect collision with the reserved pseudo group column. |
ConstantColumnCheck |
Warn about columns with a single unique value. |
TimestampColumnCheck |
Validate time-series timestamp column presence and integrity. |
DatasetSizeCheck
¶
Bases: DataFrameCheck
Block training when the training split is unusably small.
An empty / near-empty split would produce cascading failures in
downstream token-budget and metadata checks; emitting the error here
lets the orchestrator gate them cleanly via requires.