arviz_stats.psense

arviz_stats.psense(data, var_names=None, filter_vars=None, group='prior', coords=None, sample_dims=None, alphas=(0.99, 1.01), group_var_names=None, group_coords=None)

Compute power-scaling sensitivity values.

Parameters:

dataxarray.DataTree or InferenceData

Input data. It should contain the posterior and the log_likelihood and/or log_prior groups.

var_nameslist of str, optional

Names of posterior variables to include in the power scaling sensitivity diagnostic

filter_vars: {None, “like”, “regex”}, default None

Used for var_names only. If None (default), interpret var_names as the real variables names. If “like”, interpret var_names as substrings of the real variables names. If “regex”, interpret var_names as regular expressions on the real variables names.

group{“prior”, “likelihood”}, default “prior”

If “likelihood”, the pointsize log likelihood values are retrieved from the log_likelihood group and added together. If “prior”, the log prior values are retrieved from the log_prior group.

coordsdict, optional

Coordinates defining a subset over the posterior. Only these variables will be used when computing the prior sensitivity.

sample_dimsstr or sequence of hashable, optional

Dimensions to reduce. Defaults to rcParams["data.sample_dims"]

alphastuple

Lower and upper alpha values for gradient calculation. Defaults to (0.99, 1.01).

group_var_namesstr, optional

Name of the prior or log likelihood variables to use

group_coordsdict, optional

Coordinates defining a subset over the group element for which to compute the prior sensitivity diagnostic.

Returns:

xarray.DataTree

Returns dataTree of power-scaling sensitivity diagnostic values. Higher sensitivity values indicate greater sensitivity. Prior sensitivity above 0.05 indicates informative prior. Likelihood sensitivity below 0.05 indicates weak or non-informative likelihood.

Notes

The diagnostic is computed by power-scaling either the prior or likelihood and determining the degree to which the posterior changes as described in [1]. It uses Pareto-smoothed importance sampling to avoid refitting the model.

References

[1]

Kallioinen et al, Detecting and diagnosing prior and likelihood sensitivity with power-scaling, Stat Comput 34, 57 (2024), https://doi.org/10.1007/s11222-023-10366-5