pypomp.Pomp.pfilter

Pomp.pfilter(J: int, key: Array | None = None, theta: PompParameters | None = None, thresh: float = 0.0, reps: int = 1, CLL: bool = False, ESS: bool = False, filter_mean: bool = False, prediction_mean: bool = False, track_time: bool = True) → None

Evaluate the log-likelihood via the bootstrap particle filter.

Propagates a swarm of J particles through the latent state space using Sequential Monte Carlo (bootstrap filter) to estimate the marginal log-likelihood of the observed data. Optionally computes conditional log-likelihoods, effective sample size, filtered means, and prediction means.

JAX vectorises the computation across all parameter sets in theta simultaneously.

Parameters:

  • J (int) – Number of particles.

  • key (jax.Array or None, optional) – JAX random key. Defaults to fresh_key.

  • theta (PompParameters or None, optional) – Parameter set(s) to evaluate. Defaults to theta.

  • thresh (float, optional) – ESS-based resampling threshold in the interval \([0, 1]\). Defaults to 0.0 (resample at every step).

  • reps (int, optional) – Number of independent filter replicates per parameter set. Defaults to 1.

  • CLL (bool, optional) – Whether to compute and store conditional log-likelihoods at each observation time. Defaults to False.

  • ESS (bool, optional) – Whether to compute and store the effective sample size at each observation time. Defaults to False.

  • filter_mean (bool, optional) – Whether to compute and store the filtered state mean at each observation time. Defaults to False.

  • prediction_mean (bool, optional) – Whether to compute and store the predicted state mean at each observation time. Defaults to False.

  • track_time (bool, optional) – Whether to record wall-clock execution time. Defaults to True.

Returns:

A Result is appended to results_history. Retrieve a dataframe summary with results() or the log-likelihoods directly via model.theta.logLik.

Return type:

None

See also

pypomp.functional.pfilter

Pure-functional JAX particle filter.

Examples

>>> model.fresh_key = jax.random.key(0)
>>> model.pfilter(J=1000)
>>> model.results()  # DataFrame with logLik and parameter columns