Custom Transform Steps

transform_step(
    name: str,
    func: Callable[..., Any],
    *,
    store_key: str | None = None,
    **kwargs: Any,
) -> MCStep

transform_step creates a custom per-replication transform operation. It lives under SymbolicDSGE.monte_carlo.operations.transforms.

The callable receives the normalized operation arguments:

func(
    *,
    context: MCContext,
    reference: SolvedModel,
    dgp: SolvedModel | None,
    rep_idx: int,
    **kwargs,
) -> Any

If the callable returns an MCData object, the pipeline replaces context.data with that return value. The return value is stored in context.payloads[store_key or name].

Inputs:

Name	Description
name	Runtime step name and default payload key.
func	Callable to execute for each replication.
store_key	Optional override for the payload key.
kwargs	Extra keyword arguments forwarded to func.

Bundled custom operations

In-process pipelines may use any callable. A pipeline written to an .sdsge bundle requires the callable to be a NumpyCustomFunc, usually via the custom_operation decorator. The bundle stores the callable as a side-channel resource and the portable spec references it by key.