Class DirichletLearner

Online Dirichlet parameter learner for CPT updates.

Maintains Dirichlet priors for each CPT cell and updates them with new observations using closed-form moments (no MCMC needed).

• Prior: Dirichlet(α)
• After n observations: Dirichlet(α + counts)
• Mean: (α + count) / (α₀ + n)
• Variance: closed-form from Dirichlet moments

Usage:

const learner = new DirichletLearner({ alpha: 1 });
learner.update('A', '', true);   // root node A observed anomalous
learner.update('B', '1', false); // B observed normal when parent=1
const cpt = learner.getCPTs();   // get updated CPTs
Copy