score_batch

Scope: builder · Signature: score_batch(experiment_id: str, action: {submit|status|cancel|list|evaluation|ranking|rows}, job_id: str | None = None, params: dict[str, Any] | None = None, idempotency_key: str | None = None) -> Any

Use to score a dataset with a trained experiment, and read the results.

action="submit": params dataset_id (same workspace), target_column (when labels exist: the evaluation then compares), calibrated, include_proba, include_reason_codes, chunk_size, reasons_per_item → a job and its wait_handle ("job:/": in a session ParleML runs, await_operation on it; elsewhere poll action="status" with job_id). "list" the experiment's jobs; "cancel" one. For a finished supervised job: "evaluation" (the held-out-style metrics on the new rows; curves as at most 50 points, like get_model_evaluation's), "ranking" (lift / gains). All families support "rows" (offset, limit ≤100). Anomaly and clustering runs score too; recommendation and ltr runs score each row into its ranked list, a recommendation run with each product's holding and top reasons_per_item (default 3) drivers (get_playbook("schema-recommendation")). Returns its idempotency_key.