-
Notifications
You must be signed in to change notification settings - Fork 9
Open
Description
I'd like to propose adding a Latent Dirichlet Allocation (LDA) example to the codebase. Looking at the current functionality, it seems we already have all the necessary building blocks in place - specifically the DirichletCollection
and MultinomialPolya
nodes!
LDA would make an excellent addition to our examples for several reasons:
- It's a fundamental topic model widely used in text analysis and machine learning
- It would demonstrate RxInfer's capability with hierarchical Bayesian models
- It exercises multiple probability distributions in combination
- The results are intuitive and visually presentable
The generative model is straightforward:
- Topic-word distributions:
φₖ ~ Dirichlet(β)
- Document-topic distributions:
θₗ ~ Dirichlet(α)
- Topic assignments:
zₗₙ ~ Multinomial(θₗ)
- Words:
wₗₙ ~ Multinomial(φzₗₙ)
This seems like a relatively easy yet powerful example to implement that would showcase RxInfer.jl's strengths. It could also serve as a benchmark against other probabilistic programming frameworks.
Metadata
Metadata
Assignees
Labels
No labels