The initial release: the full Markov-like-process inference family behind a decidable engine dispatch.
- HMM — forward–backward (
smooth), Viterbi (decode),loglik, Baum–Welch (fit); homogeneous or time-varying transitions. - Semi-Markov — the standard right-censored explicit-duration (EDHMM) Viterbi + per-frame posteriors, with shifted-negative-binomial dwell.
- Linear-Gaussian — the Kalman
filterand the RTSsmoothsmoother, expressed through the same forward recursion with a Gaussian belief. - Particle filter — a bootstrap (SIR) filter for nonlinear models; the first approximate engine, deterministic given its seed.
- Decidable dispatch —
resolve_engine→Exact | Approximate | Intractable.
numpy/scipy only, mypy --strict, 100% coverage; every engine validated against an independent oracle.
Install: pip install marmo then import markovlib (the name markovlib was taken on PyPI).