Skip to main content

Module hmm

qualia_core_db::solvers::learning::sequential

Module hmm 

Source
Expand description

Discrete Hidden Markov Model (PRML ch 13.2) — the standard estimators over a sequence of discrete observations: the scaled forward algorithm for the sequence log-likelihood, Viterbi for the most-likely state path, and Baum-Welch (EM) to learn the parameters. Mission note: time-indexed provenance / life-record reasoning is temporal; this is the canonical model over censored temporal evidence. Kernel-class Reduction (the message passes).

Structs§

Hmm
A discrete HMM: k hidden states, m observation symbols.

Functions§

baum_welch
Learn HMM parameters from one observation sequence by Baum-Welch (EM). Returns (model, final_log_likelihood). Initialised randomly (seeded). Fails closed on bad shapes / out-of-range symbols.