arviz_stats.base.array_stats.loo

array_stats.loo(ary, chain_axis=-2, draw_axis=-1, r_eff=1.0, log_weights=None, pareto_k=None, log_jacobian=None)

Compute Pareto-smoothed importance sampling leave-one-out cross-validation (PSIS-LOO-CV).

Parameters:

aryarray_like

Log-likelihood values.

chain_axisint, default -2

Axis for chains.

draw_axisint, default -1

Axis for draws.

r_efffloat, default 1.0

Relative effective sample size.

log_weightsarray_like, optional

Pre-computed PSIS log weights.

pareto_karray_like, optional

Pre-computed Pareto k-hat diagnostic values.

log_jacobianarray_like, optional

Log-Jacobian adjustment for variable transformations.

Returns:

elpd_iarray_like

Pointwise expected log predictive density.

pareto_karray_like

Pareto k-hat diagnostic values.

p_loo_iarray_like

Pointwise effective number of parameters.