functional.mcmc.resolve_kernel()

Normalize a kernel specification.

Usage

Source

functional.mcmc.resolve_kernel(
    kernel,
    model,
    kernel_kwargs,
)

None -> NUTS(model, **kernel_kwargs) (kwargs tune the default kernel without naming it); a kernel class -> kernel(model, **kernel_kwargs); a kernel instance -> returned unchanged, and combining an instance with non-empty kernel_kwargs raises ValueError (ambiguous). Anything else -> TypeError.

Parameters

kernel: KernelLike

The kernel specification (see ~numpyro_forecast.typing.KernelLike).

model: ForecastModel

The model the kernel is built against.

kernel_kwargs: Mapping[str, Any] | None

Extra keyword arguments forwarded to the kernel constructor (ignored, and rejected, for an already-constructed instance).

Returns

MCMCKernel

The resolved kernel.

Raises

KernelConfigError

If a kernel instance is combined with non-empty kernel_kwargs.

KernelResolutionError

If kernel is neither None, an MCMCKernel subclass, nor an MCMCKernel instance.