functional.svi.SVIFit

The result of fitting a forecasting model with SVI.

Usage

Source

functional.svi.SVIFit(
    guide,
    params,
    losses,
    data=None,
    covariates=None,
)

Attributes

guide: AutoGuide | Callable[…, None]

The fitted variational guide.

params: dict[str, Array]

The learned variational parameters.

losses: Array

The ELBO loss per SVI step (shape (num_steps,)).

data: Array | None

The in-sample data the model was fit on (needed to draw from hand-written guides and by ~numpyro_forecast.convert.to_datatree()). None for fits constructed without it.

covariates: Array | None

The in-sample covariates the model was fit on. None for fits constructed without it.