aimz.ImpactModel.sample_posterior_predictive#

ImpactModel.sample_posterior_predictive(X, *, intervention=None, rng_key=None, return_sites=None, shard_axis='obs', batch_size=None, output_dir=None, progress=True, **kwargs)[source]#

Draw samples from the posterior predictive distribution.

This method is a convenience alias for predict(), with in_sample automatically set to True.

Parameters:

  • X (ArrayLike | ArrayLoader) – Input data. If array-like, the leading axis is the observation axis. Alternatively, a data loader that holds all array-like objects and handles batching internally.

  • intervention (dict | None) – A dictionary mapping sample sites to their corresponding intervention values. Interventions enable counterfactual analysis by modifying the specified sample sites during prediction (posterior predictive sampling).

  • rng_key (Array | None) – A pseudo-random number generator key. By default, an internal key is used and split as needed.

  • return_sites (str | Iterable[str] | None) – Names of variables (sites) to return. If None, samples param_output and deterministic sites.

  • shard_axis (Literal['obs', 'draw']) – Multi-device sharding strategy; no effect on a single device. "obs" (default) shards the input across devices and replicates the posterior. "draw" shards the posterior across devices and replicates the input, which must be an array, not a data loader. If the model has no posterior samples, the data path is used regardless of shard_axis.

  • batch_size (int | None) – Size of each batch, taken from the input under shard_axis="obs" and from the draws under shard_axis="draw". Also used as the chunk size when storing results. If None, it is determined automatically from the input size and number of samples. Ignored if X is a data loader, in which case the data loader is expected to handle batching internally.

  • output_dir (str | Path | None) – The directory where the outputs will be saved. If the specified directory does not exist, it will be created automatically. If None, a model-owned temporary directory is used. A subdirectory is generated within this directory to store the outputs. The temporary directory is removed by cleanup(), cleanup_models(), or when the model is garbage-collected. Pass an explicit output_dir to keep results beyond the model’s lifetime.

  • progress (bool) – Whether to display a progress bar.

  • **kwargs (object) – Additional arguments passed to the model.

Returns:

Posterior predictive samples. Posterior samples are included if available.

Raises:

  • TypeError – If param_output is passed as an argument, or shard_axis="draw" is used with a data loader X.

  • ValueError – If shard_axis is not "obs" or "draw".

Return type:

xr.DataTree

See also

predict().