VinDsl.jl: Fast and Furious Variational Inference

VinDsl.jl is a Julia package that aims to provide a fast, flexible, thoroughly hackable domain-specific language (DSL) for variational Bayesian inference.

In particular VinDsl features:

  • Intelligent index handling: you define the model structure, VinDsl handles the sum over indices automatically
  • A set of macros for coding conjugate models and updates
  • Limited support for automatic expectation-taking
  • built-in support for Hidden Markov Models

Contents:

Indices and tables