Add auxiliary particle filter #11
Conversation
| function update_ref!( | ||
| pc::ParticleContainer{T}, ref_state::Union{Nothing,AbstractVector{T}}, step::Integer=0 | ||
| pc::ParticleContainer{T}, | ||
| ref_state::Union{Nothing,AbstractVector{T}}, | ||
| ::AbstractFilter, | ||
| step::Integer=0, | ||
| ) where {T} |
There was a problem hiding this comment.
why do we need the filter passed through this function?
There was a problem hiding this comment.
We don't need it here, but we might need to dispatch on the filter type when updating the reference particle. That would be useful for ancestor resampling for example
src/resamplers.jl
Outdated
| ) | ||
| return reset_weights!(state, idxs, filter) |
There was a problem hiding this comment.
I prefer the reset_weights!(...) paired with logmarginal(...) as is done here. @THargreaves and I had a brief back and forth about this, but we never properly settled that discussion.
src/algorithms/apf.jl
Outdated
| predicted = map( | ||
| x -> SSMProblems.simulate(rng, model.dyn, step, x; kwargs...), | ||
| states.filtered.particles, | ||
| ) |
There was a problem hiding this comment.
shouldn't this be a deterministic function of states.filtered.particles?
| function step( | ||
| rng::AbstractRNG, | ||
| model::AbstractStateSpaceModel, | ||
| alg::AuxiliaryParticleFilter, | ||
| iter::Integer, | ||
| state, | ||
| observation; | ||
| kwargs..., | ||
| ) | ||
| proposed_state = predict(rng, model, alg, iter, state, observation; kwargs...) | ||
| filtered_state, ll = update(model, alg, iter, proposed_state, observation; kwargs...) | ||
|
|
||
| return filtered_state, ll | ||
| end |
There was a problem hiding this comment.
It should work without redefining step, as long as AuxiliaryParticleFilter is a subclass of AbstractFilter
There was a problem hiding this comment.
The current api doesn't forward observation to predict. I also tend to agree with #9 and splitting resample and predict.
There was a problem hiding this comment.
The current api doesn't forward observation to predict. I also tend to agree with #9 and splitting resample and predict.
I wonder if it should. For general proposal distributions other than forward simulation (i.e. bootstrap filter) that would be needed.
Or at least have a subtype for Guided, Independent and Auxiliary filters that has this modified step/predict method.
There was a problem hiding this comment.
The current api doesn't forward observation to predict
I completely forgot about that. We may need to think about defining some sort of AbstractProposal for more complex transition kernels.
There was a problem hiding this comment.
And yeah, I agree we should split up resample and predict
|
I can add a comparison, this is not really ready for prime time, just seeing if more general filters would fit into the API |
FredericWantiez
force-pushed
the
fred/auxiliary-particle-filter
branch
2 times, most recently
from
bd2e22f
to
0fadd2b
Compare
FredericWantiez
force-pushed
the
fred/auxiliary-particle-filter
branch
from
0fadd2b
to
c791095
Compare
Add a draft for an auxiliary particle filter. This is more of a proof of concept as we need to change the filter interface a tiny bit.
Looks like we might need to add some of these:
reset_weights(filter::AbstractFilter, ...)update_weights!(filter::AbstractFilter, ...)As well as passing the
observationinpredict.