Particle Filtering

Particle Filtering

state = initialize_particle_filter(model::GenerativeFunction, model_args::Tuple,
    observations::ChoiceMap, proposal::GenerativeFunction, proposal_args::Tuple,
    num_particles::Int)

Initialize the state of a particle filter using a custom proposal for the initial latent state.

source
state = initialize_particle_filter(model::GenerativeFunction, model_args::Tuple,
    observations::ChoiceMap, num_particles::Int)

Initialize the state of a particle filter, using the default proposal for the initial latent state.

source
particle_filter_step!(state::ParticleFilterState, new_args::Tuple, argdiffs,
    observations::ChoiceMap, proposal::GenerativeFunction, proposal_args::Tuple)

Perform a particle filter update, where the model arguments are adjusted, new observations are added, and some combination of a custom proposal and the model's internal proposal is used for proposing new latent state. That is, for each particle,

  • The proposal function proposal is evaluated with arguments Tuple(t_old, proposal_args...) (where t_old is the old model trace), and produces its own trace (call it proposal_trace); and
  • The old model trace is replaced by a new model trace (call it t_new).

The choicemap of t_new satisfies the following conditions:

  1. get_choices(t_old) is a subset of get_choices(t_new);
  2. observations is a subset of get_choices(t_new);
  3. get_choices(proposal_trace) is a subset of get_choices(t_new).

Here, when we say one choicemap a is a "subset" of another choicemap b, we mean that all keys that occur in a also occur in b, and the values at those addresses are equal.

It is an error if no trace t_new satisfying the above conditions exists in the support of the model (with the new arguments). If such a trace exists, then the random choices not determined by the above requirements are sampled using the internal proposal.

source
particle_filter_step!(state::ParticleFilterState, new_args::Tuple, argdiffs,
    observations::ChoiceMap)

Perform a particle filter update, where the model arguments are adjusted, new observations are added, and the default proposal is used for new latent state.

source
Gen.maybe_resample!Function.
did_resample::Bool = maybe_resample!(state::ParticleFilterState;
    ess_threshold::Float64=length(state.traces)/2, verbose=false)

Do a resampling step if the effective sample size is below the given threshold. Return true if a resample thus occurred, false otherwise.

source
Gen.log_ml_estimateFunction.
estimate = log_ml_estimate(state::ParticleFilterState)

Return the particle filter's current estimate of the log marginal likelihood.

source
Gen.get_tracesFunction.
traces = get_traces(state::ParticleFilterState)

Return the vector of traces in the current state, one for each particle.

source
Gen.get_log_weightsFunction.
log_weights = get_log_weights(state::ParticleFilterState)

Return the vector of log weights for the current state, one for each particle.

The weights are not normalized, and are in log-space.

source
traces::Vector = sample_unweighted_traces(state::ParticleFilterState, num_samples::Int)

Sample a vector of num_samples traces from the weighted collection of traces in the given particle filter state.

source