r-hsmm

Implementation of Recurrent Hidden Semi-Markov Model http://www.cc.gatech.edu/~lsong/papers/DaiDaiZhaLietal17.pdf

build

Get the source code

git clone https://github.com/Hanjun-Dai/r-hsmm

This code depends on an obsolete graphnn library:

cd code/graphnn-1.11
build the graphnn library with the instructions here:
https://github.com/Hanjun-Dai/graphnn/tree/c59a2dd15cf528bfe0ade5a5680466dfaf027c0a

Build the rnn-hsmm c++ code:

cd code/rnn_edhmm_viterbi
make

learning

lambda = 0

essentially this is equivalent to EM.

lambda = \infty

coming soon

inference

One can use either EM or variational auto-encoder to do inference.

EM:

use the viterbi decoder in code/rnn_edhmm_viterbi

VAE:

coming soon