SplitterTags¶

@dataclass
class SplitterTags:
    """Tags specific to cross-validation splitters.

    Parameters
    ----------
    splitter_type : {"expanding", "sliding", "gap"} or None, default=None
        Type of cross-validation splitter:
        - "expanding": Expanding window (train set grows over time)
        - "sliding": Sliding window (fixed train set size)
        - "gap": Reserved for a future gap splitter (wraps another splitter to
          add a gap between train and test); not used by any shipped splitter
        - None: Not determined or not applicable
    supports_panel_data : bool, default=False
        Whether the splitter can handle panel data (multiple time series
        with prefixed column names).
    produces_non_overlapping_tests : bool, default=True
        Whether test sets from different splits are guaranteed not to overlap.
        True for expanding/sliding windows, may be False for custom splitters.
    stateful : bool, default=False
        Whether the splitter maintains state across split() calls.

    """

    splitter_type: Literal["expanding", "sliding", "gap"] | None = None
    supports_panel_data: bool = False
    produces_non_overlapping_tests: bool = True
    stateful: bool = False