arviz_stats.ELPDDataLOOSubsample

class arviz_stats.ELPDDataLOOSubsample(*, elpd, se, p, n_samples, n_data_points, scale, warning, good_k, elpd_i=None, pareto_k=None, approx_posterior=False, log_weights=None, log_jacobian=None, p_loo_i=None, influence_pareto_k=None, n_eff_i=None, subsample_size, subsampling_se, loo_subsample_observations, elpd_loo_approx, log_p=None, log_q=None, thin_factor=None)

Subsampled PSIS-LOO-CV results returned by loo_subsample.

Also returned by update_subsample. Inherits the attributes of ELPDDataLOO. With pointwise=True the log_weights attribute holds a Dataset with one variable named after the log likelihood variable, otherwise a DataArray restricted to the subsampled observations.

Attributes:

subsample_sizeint

Number of observations in the subsample.

subsampling_sefloat

Standard error due to subsampling uncertainty only.

loo_subsample_observationsndarray

Indices of the subsampled observations.

elpd_loo_approxxarray.DataArray

Approximate pointwise ELPD for every observation.

log_pxarray.DataArray or ndarray, optional

Log density of the target posterior.

log_qxarray.DataArray or ndarray, optional

Log density of the proposal posterior.

thin_factorint or str, optional

Thinning factor applied to the posterior draws, an integer or "auto".

__init__(*, elpd, se, p, n_samples, n_data_points, scale, warning, good_k, elpd_i=None, pareto_k=None, approx_posterior=False, log_weights=None, log_jacobian=None, p_loo_i=None, influence_pareto_k=None, n_eff_i=None, subsample_size, subsampling_se, loo_subsample_observations, elpd_loo_approx, log_p=None, log_q=None, thin_factor=None)

Methods

__init__(*, elpd, se, p, n_samples, ...[, ...])

Attributes

approx_posterior
elpd_i
influence_pareto_k
kind	Name of the estimator that produced the result.
log_jacobian
log_p
log_q
log_weights
n_eff_i
p_loo_i
pareto_k
thin_factor
subsample_size
subsampling_se
loo_subsample_observations
elpd_loo_approx
elpd
se
p
n_samples
n_data_points
scale
warning
good_k