Bayesian posterior classifier (optional)

Combines a deterministic ClassificationResult with a spatial prior. The deterministic key remains authoritative – this function reports only an alternative probabilistic view useful for downstream uncertainty quantification.

Usage

posterior_classify(result, prior, epsilon = 0.001)

Arguments

result

A ClassificationResult from classify_wrb2022.

prior

A spatial-prior data.table (as returned by spatial_prior).

epsilon

Small smoothing constant added to all prior entries before normalising, so RSGs unseen by the prior do not receive zero posterior.

Value

A data.table with columns rsg_code, prior, likelihood, posterior.

Details

Posterior is computed under the simple model: $$P(rsg | site, evidence) \propto L(rsg | evidence) \times P(rsg | site)$$ where the likelihood L is concentrated on the deterministic assignment (delta-1 at that code) by default, optionally smoothed if key_passed_others is supplied.