forecaster.PathfinderForecaster

Fit a forecasting model with BlackJAX Pathfinder variational inference.

Usage

Source

forecaster.PathfinderForecaster(
    rng_key,
    model,
    data,
    covariates,
    *,
    num_elbo_samples=200,
    ftol=1e-05,
    maxiter=30
)

A thin shim over numpyro_forecast.contrib.blackjax.fit_pathfinder(). BlackJAX is an optional dependency (pip install numpyro_forecast[blackjax]) imported lazily here, so constructing this class is the opt-in that pulls it in; importing numpyro_forecast never does. The constructor mirrors Forecaster without guide/optim (Pathfinder has neither) and adds num_elbo_samples/ftol.

Parameters

rng_key: Array

PRNG key for inference.

model: ForecastModel

The forecasting model to fit (OOP instance or functional model).

data: Array

In-sample data with time at axis -2.

covariates: Array

Covariates with time at axis -2 and the same duration as data.

num_elbo_samples: int = 200

Number of Monte Carlo samples used to estimate the ELBO along the L-BFGS path.

ftol: float = 1e-05

L-BFGS relative function-value tolerance (convergence criterion).

maxiter: int = 30

Maximum number of L-BFGS iterations (see ~numpyro_forecast.contrib.blackjax.fit_pathfinder(); models with per-step time latents typically need far more than the default 30).