A simple Hidden Markov Model implementation in Julia. Intended mostly for educational purposes. Only supports discrete emission probabilities
I am developing HMM.jl for a more general-purpose module.
Pkg.clone("https://github.com/ahwillia/ToyHMM.jl.git")
using ToyHMM
n_states = 2
n_outputs = 3
hmm = dHMM(n_states,n_outputs)
println(hmm.A) # state-transition matrix (randomly initialized, rows sum to 1)
println(hmm.B) # emmission matrix (randomly initialized, rows sum to 1)
println(hmm.p) # initial state probabilities (randomly initialized)
o = [1,1,2,1,1,2,1,2,1,3,3,3,3,2,2,3,3,3] # example observation sequence
ch = baum_welch!(hmm,o) # fit model using Expectation-Maximization
println(ch) # log-likelihood values, convergence history
println(hmm.A) # fitted values of the hmm model
println(hmm.B)
println(hmm.p)
println(viterbi(hmm,o)) # most likely state sequence given hmm params
(also see test/runtests.jl
for some examples)
using ToyHMM
n_states = 2
n_outputs = 3
# create a very long output sequence
true_model = dHMM(n_states,n_outputs)
(s,o) = generate(true_model,100_000)
# try to recover similar params by fitting new model
fit_model = dHMM(n_states,n_outputs)
@time ch = baum_welch!(fit_model,o)
elapsed time: 7.958006041 seconds (4140814448 bytes allocated, 26.85% gc time)
Michael Hamilton's implementation (python): http://www.cs.colostate.edu/~hamiltom/code.html
Guy Zyskind's implementation (python): https://github.com/guyz/HMM
Rabiner, Lawrence R. "A tutorial on hidden Markov models and selected applications in speech recognition." Proceedings of the IEEE 77.2 (1989): 257-286.
MacKay DJC (1997). Ensemble Learning for Hidden Markov Models Technical report, University of Cambridge
MacKay DJC (1998). Choice of Basis for Laplace Approximation. Machine Learning. 33(1), 77-86.
Beal MJ (2003). Variational Algorithms for Approximate Bayesian Inference. PhD Thesis, University College London