Utilities

Numerical helpers shared by every setting.

Tie-breaking selection

utils.randmax	Index of a maximum of A, breaking ties uniformly at random.
utils.randmin	Index of a minimum of A, breaking ties uniformly at random.
utils.allmax	Maximum of a together with every index attaining it.

Divergences

Parametric divergences for the usual one-parameter exponential families, plus the non-parametric \(K_{\inf}\) used by the batched IMED variants. Pick the one matching your rewards — an index algorithm is asymptotically optimal only for the right divergence.

utils.klBern	Kullback-Leibler divergence between two Bernoulli distributions.
utils.klGauss	Kullback-Leibler divergence between two Gaussians of equal variance.
utils.klPoisson	Kullback-Leibler divergence between two Poisson distributions.
utils.klExp	Kullback-Leibler divergence between two exponential distributions.
utils.KLinf_threshold	Non-parametric divergence \(K_{\inf}\) between an empirical distribution and the set of distributions with mean above a threshold, evaluated using the concave dual.

Sampling

utils.categorical_sample	Draw one index from a categorical distribution.
utils.Dirac	Deterministic distribution concentrated on a single value.

Setting validation

Checks that a setting folder follows the layout described in Contributing.

validator.validate_setting	Validate a complete setting folder.
validator.validate_structure	Check required files and folders.
validator.validate_environment	Check Environment class and API.
validator.validate_agent	Check Agent class.
validator.validate_interaction	Check Interaction class and API.
validator.validate_environment_yaml	Validate envs/environments.yaml files.
validator.load_module	Import a python module from a file path.
validator.report	Print one check result and pass its verdict through.