probe_dataset¶
Scope: compute · Signature: probe_dataset(dataset_id: str, kind: {signal|forecast_signal|baseline_floor|temporal_split}, params: dict[str, Any] | None = None, refresh: bool = False) -> Any
Use for cheap pre-flight probes on a dataset, before proposing a run.
kind="signal": is there learnable signal for params.target? (verdict,
drivers, leakage smell, holdout tiles; task = auto | binary_classification
| multiclass_classification | regression; curves stay on its ui_url).
kind="forecast_signal": does a zero-shot model beat seasonal-naive?
(time_col, target_col, series_id_col, horizon, resample_to, resample_agg).
kind="baseline_floor": a trivial predictor's score (target, task, metric
from describe_platform("metrics"), positive_class). kind="temporal_split":
a temporal run's partition, in the plan's words (time_column +
experiment.validation's temporal_boundary_mode, holdout_frac,
validation_cutoff, outer_test_cutoff, embargo; validation_holdout_frac);
its launch block pastes into the plan. Signal checks are cached
(refresh=true recomputes); one still computing returns status="running"
and a wait_handle: wait once with the session tool await_operation
(operation_id=
Powered by: ml4 foundation signal check, zero-shot Chronos backtest, baseline floor and temporal partition (ml4/docs/guides/signal-check.md, ml4/docs/guides/forecasting.md).