pmcmc and pfilter now have the capability of saving filtered trajectories. These can be extracted using the new method filter.traj.
The principal use will be in conjunction with pmcmc, where, upon convergence to the posterior, samples from the filtered trajectories will be draws from the posterior P[x[1:T] | y[1:T]], i.e., the smoothing distribution.
Thanks to Sebastian Funk for initiating this development!
pomp: statistical inference for
partially-observed
Markov processes