rllm.preprocessing.FillNAConfig

class rllm.preprocessing.FillNAConfig(numerical_strategy: Literal['mean', 'median', 'mode', 'constant'] = 'mean', numerical_fill_value: float = 0.0, categorical_fill_value: int | str = -1, text_fill_value: str = '', timestamp_strategy: Literal['ffill', 'bfill', 'median', 'constant'] = 'ffill', timestamp_fill_value: object | None = None)[source]

Bases: object

Configuration for missing-value imputation by column type. It centralizes fill strategies and fallback values for numerical, categorical, text, and timestamp columns. These options are consumed by fillna_by_coltype() and related helpers during preprocessing.

Parameters:

  • numerical_strategy (Literal["mean", "median", "mode", "constant"]) – Strategy for numerical columns.

  • numerical_fill_value (float) – Constant fallback for numerical columns.

  • categorical_fill_value (Union[int, str]) – Fill value for categorical columns.

  • text_fill_value (str) – Fill value for text columns.

  • timestamp_strategy (Literal["ffill", "bfill", "median", "constant"]) – Strategy for timestamp columns.

  • timestamp_fill_value – Constant fallback for timestamp columns when timestamp_strategy="constant".