HMMRBM.jl

HMMRBM.jl couples Hidden Markov Models with Restricted Boltzmann Machine (RBM) emission models. It extends HiddenMarkovModels.jl by providing multi-sequence training utilities and RBM-based observation families sourced from RestrictedBoltzmannMachines.jl.

It implements hidden Markov models trained on multiple sequences of observations obtained from distinct emission distributions that share the emission parameter. The result of training is a transition matrix for each sequence and one emission parameter for all sequences.

It also implements emission distributions that use a pre-trained restricted Boltzmann machine as parametric emission, and hidden vector as emission parameter. This allows training HMMs with latent-aligned RBMs as the emission distribution.

Trained HMMs can be persisted with HDF5.jl.

Installation

HMMRBM.jl is registered in the BrainRegistry.jl registry.

To activate the registry, do

using Pkg
pkg"registry add https://github.com/gfayebedrin/BrainRegistry.jl"

This only needs to be done once per Julia installation.

To install the package, do

using Pkg
Pkg.add("HMMRBM")
Pkg.add("HDF5") # Optional