loo: Efficient Leave-One-Out Cross-Validation for Bayesian Models

Aki Vehtari developed an efficient approximate leave-one-out cross-validation method for Bayesian models. The method uses Pareto smoothed importance sampling (PSIS) to regularize importance weights and is described in a 2017 paper. The package also provides methods for model averaging via stacking and other weighting techniques.

Use Cases

• Perform efficient model validation for Bayesian models based on the described leave-one-out cross-validation approximation.
• Regularize importance weights for predictive error estimation using the Pareto smoothed importance sampling (PSIS) procedure.
• Compare predictive errors between different Bayesian models using the provided approximate standard errors.
• Average Bayesian predictive distributions using stacking and other model weighting techniques mentioned in the description.

Strengths

• Implements a method described in a peer-reviewed 2017 paper by Vehtari, Gelman, and Gabry.
• Provides approximate standard errors for estimated predictive errors and model comparisons as a byproduct.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Aki Vehtari

Collection Method

Methodological development for statistical computation, as described in the referenced academic paper.

Time Range

Method described in a 2017 publication.

Freshness

Last updated date is unknown.

Geography

Geographic coverage is not specified.