kleinschmidt / Particles.jl

nonparametric clustering with sequential monte carlo

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

integrate with OnlineStats?

kleinschmidt opened this issue · comments

Worth considering at least...although it might require some changes in the implementation in order to be really useful, since this currently is NOT O(1) with memory (the cluster assignments are tracked by the particles themselves). Maybe it would be possible to create a variant that doesn't actually keep track of the ancestor, since that's only needed for the read-out stuff right?

At a more practical level I think this just requires defining ParticleFilter{T} <: OnlineStat{T}, enforcing that data is of type T, and adding methods for

  • _fit!(stat::ParticleFilter{T}, value::T)
  • value(stat::ParticleFilter{T}, args...) (not sure what this would return actually...some kind of representation of the inferred clustering? just itself? a MixtureModel that captures the clusters?)
  • nobs(stat) (override default that looks in the .n field).