Diagnostics Functions

birdman.diagnostics.ess(inference_object: InferenceData, params: Optional[List[str]] = None, **kwargs) → Dataset

Estimate effective sample size for parameters.

Wrapper function for az.ess.

See https://arviz-devs.github.io/arviz/api/generated/arviz.ess.html for details.

See Stan docs for more information on ESS.

Parameters

  • inference_object (az.InferenceData) – Inference object with posterior draws

  • params (List[str]) – Variables to include, defaults to all

  • kwargs – Keyword arguments to pass to az.ess

Returns

Estimated effective sample sizes

Return type

xr.Dataset

birdman.diagnostics.loo(inference_object: InferenceData, **kwargs) → ELPDData

Compute Pareto-smoothed importance sampling LOO CV.

Wrapper function for az.loo.

See https://arviz-devs.github.io/arviz/api/generated/arviz.loo.html for details.

Note

This function requires that the inference object has a log_likelihood group.

Parameters

  • inference_object (az.InferenceData) – Inference object with posterior draws

  • kwargs – Keyword arguments to pass to az.loo

Returns

Estimated log pointwise predictive density

Return type

az.ELPDData

birdman.diagnostics.r2_score(inference_object: InferenceData) → Series

Compute Bayesian \(R^2\).

Wrapper function for az.r2_score.

Note

This function requires that the inference object has a posterior_predictive group.

Parameters

inference_object (az.InferenceData) – Inference object with posterior draws

Returns

Bayesian \(R^2\) & standard deviation

Return type

pd.Series

birdman.diagnostics.rhat(inference_object: InferenceData, params: Optional[List[str]] = None, **kwargs) → Dataset

Estimate Gelman-Rubin statistic of chain convergence.

Wrapper function for az.rhat.

See https://arviz-devs.github.io/arviz/api/generated/arviz.rhat.html for details.

Rhat values should be very close to 1.0.

Parameters

  • inference_object (az.InferenceData) – Inference object with posterior draws

  • params (List[str]) – Variables to include, defaults to all

  • kwargs – Keyword arguments to pass to az.rhat

Returns

Estimated Rhat values

Return type

xr.Dataset