ControlledHiddenMarkovModels.jl

Welcome to the documentation of ControlledHiddenMarkovModels.jl, a package for Hidden Markov Models with exogenous control variables.

Why would you need it?

This package focuses on discrete-time HMMs with a finite state space, but it's not the only one! A few alternatives that I am aware of:

• HMMBase.jl
• HMMGradients.jl
• MarkovModels.jl
• Mitosis.jl

I started my own package because I needed specific features that were not simultaneously available elsewhere (to the best of my knowledge):

• Control variables (obviously)
• Compatibility with generic emissions (beyond Distributions.jl)
• Numerical stability thanks to log-scale computations
• Compatibility with automatic differentiation of parameters
  • in forward mode
  • in reverse mode (WIP)

Getting started

To install the package, open a Julia Pkg REPL and run

pkg> add https://github.com/gdalle/ControlledHiddenMarkovModels.jl

Mathematical background

To understand the algorithms implemented here, check out the following literature:

A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition, Lawrence R. Rabiner (1989)

An Input Output HMM Architecture, Yoshua Bengio and Paolo Frasconi (1994)