models.md

Model Objects

Model objects describe transition and measurement capabilities independently of a concrete filter implementation.

This initial model layer focuses on likelihood- and sampling-based capabilities for particle filters, grid filters, and other callback-based estimators.

Measurement likelihoods

Use LikelihoodMeasurementModel when a measurement update is represented by a likelihood callback:

from pyrecest.models import LikelihoodMeasurementModel

measurement_model = LikelihoodMeasurementModel(likelihood)
weight = measurement_model.likelihood(measurement, state)

The callback convention is likelihood(measurement, state).

A log_likelihood callback may also be supplied when the log-domain form is preferable.

State-conditioned distributions

LikelihoodMeasurementModel.from_distribution_factory(...) builds a likelihood model from a callable that returns a conditional measurement distribution for a state. The returned distribution must expose a density method such as pdf.

Sampleable transitions

Use SampleableTransitionModel when the transition model can draw next-state samples:

from pyrecest.models import SampleableTransitionModel

transition_model = SampleableTransitionModel(sample_next)
samples = transition_model.sample_next(state, n=100)

The callback convention is sample_next(state, n=1).

Density-based transitions

Use DensityTransitionModel when the transition model can evaluate a transition density:

from pyrecest.models import DensityTransitionModel

transition_model = DensityTransitionModel(transition_density)
density = transition_model.transition_density(state_next, state_previous)

The callback convention is transition_density(state_next, state_previous).

Scope

These model objects are additive infrastructure. They do not deprecate existing filter APIs and do not modify filter behavior by themselves. Filter-specific adapters can consume these objects in later changes.