latent.stats.bayesian¶

Bayesian estimation for binary metrics.

Functions¶

beta_binomial_posterior¶

beta_binomial_posterior(successes: int, total: int, prior_alpha: float = 1.0, prior_beta: float = 1.0, confidence_level: float = 0.95) -> MetricResult

Beta-Binomial posterior with configurable prior.

Args: successes: Number of successes. total: Total trials. prior_alpha: Beta distribution alpha parameter (default 1 = uninformative). prior_beta: Beta distribution beta parameter (default 1 = uninformative). confidence_level: Credible interval level.

Returns: MetricResult with posterior mean and credible interval.

get_empirical_prior¶

get_empirical_prior(metric_name: str, experiment_name: str | None = None) -> tuple[float, float]

Retrieve prior from most recent MLflow run.

Searches MLflow for the most recent run containing the specified metric and converts it to Beta distribution parameters.

Args: metric_name: Name of the metric to look up. experiment_name: Optional experiment name to search within.

Returns: Tuple of (alpha, beta) for Beta distribution prior. Returns (1.0, 1.0) if no prior run found.