saved_states: Saved states

In pomp: Statistical Inference for Partially Observed Markov Processes

Saved states

Description

Retrieve latent state trajectories from a particle filter calculation.

Usage

## S4 method for signature 'pfilterd_pomp'
saved_states(object, ..., format = c("list", "data.frame"))

## S4 method for signature 'pfilterList'
saved_states(object, ..., format = c("list", "data.frame"))

Arguments

object	result of a filtering computation
...	ignored
format	character; format of the returned object (see below).

Details

When one calls pfilter with save.states="filter" or save.states="prediction", the latent state vector associated with each particle is saved. This can be extracted by calling saved_states on the ‘pfilterd.pomp’ object. If the filtered particles are saved, these particles are unweighted, saved after resampling using their normalized weights. If the argument save.states="prediction" was used, the particles correspond to simulations from rprocess, and their corresponding unnormalized weights are included in the output.

Value

According to the format argument, the saved states are returned either as a list or a data frame.

If format="data.frame", then the returned data frame holds the state variables and (optionally) the unnormalized log weight of each particle at each observation time. The .id variable distinguishes particles.

If format="list" and pfilter was called with save.states="unweighted" or save.states="TRUE", the returned list contains one element per observation time. Each element consists of a matrix, with one row for each state variable and one column for each particle. If pfilter was called with save.states="weighted", the list itself contains two lists: the first holds the particles as above, the second holds the corresponding unnormalized log weights. In particular, it has one element per observation time; each element is the vector of per-particle log weights.

See Also

More on sequential Monte Carlo methods: bsmc2(), cond_logLik(), eff_sample_size(), filter_mean(), filter_traj(), kalman, mif2(), pfilter(), pmcmc(), pred_mean(), pred_var(), wpfilter()

Other extraction methods: coef(), cond_logLik(), covmat(), eff_sample_size(), filter_mean(), filter_traj(), forecast(), logLik, obs(), pred_mean(), pred_var(), spy(), states(), summary(), time(), timezero(), traces()

pomp documentation built on June 8, 2025, 10:20 a.m.