Changelog • msmbayes

Version 0.4 (2026/10/24)

Faster and more memory-efficient model fitting, by using analytic derivatives of the matrix exponential.
HMMs can optionally use parallel processing to evaluate the likelihood (grainsize argument to msm).
Default number of phases in the phase-type approximation model changed to 3.
soj_haz() function to calculate the hazard of the sojourn distribution for a set of times, for semi-Markov models.
soj_quantile() function to calculate quantiles of the sojourn distribution for semi-Markov models.
Utility sim_2state_smm() to simulate from a reversible 2-state semi-Markov model with generic sojourn distributions.
Miscellaneous fixes and improvements to nphase distribution functions.

msmbayes() now supports semi-Markov models with phase-type approximations to Weibull and Gamma sojourn distributions.
obstype and deathexact supported for exact transition times, as in msm.
Censored states supported through censor_states.
Misclassification models with fixed misclassification probabilities now supported.
Bayesian computation now uses rstan instead of cmdstanr. cmdstanr is now only used for Pathfinder, and is no longer a dependency.
Functions msmbayes_prior_sample() and msmbayes_priorpred_sample() to simulate from prior and prior predictive distributions.
Data summary function, statetable(), similar to its msm counterpart.
New output functions, including soj_prob() to return the CDF of the fitted sojourn distribution at arbitrary points, and pnext() to return the probability of the next state.
summary() has an improved interface for selecting model parameters.

Friendlier interface for specifying prior distributions.
Prior distributions now summarised in summary.msmbayes() output.
New function msmhist() for illustrating multi-state data over time.
New function standardise_to() for computing outputs standardised over the distribution of a given population.