资 源 简 介
Introduction
This is an implementation of hidden Markov model (HMM) training and classification for NVIDIA CUDA platform. A serial implementation in C is also included for comparison.
The implementation of HMM follows the tutorial paper by Rabiner. The three problem for HMM defined in the paper are:
1. compute the probability of the observation sequence
1. compute the most probable sequence
1. train hidden Markov mode parameters
This implementation supports all the three problems. However there is no support for continuous densities.
Usage
The command line usage is as follows.
$ ./hmm -hhmm [-hnt] [-c config] [-p(1|2|3)]usage: -h help -c configuration file -t output computation time -p1 compute the probability of the observation sequence -p2 compute the most probable sequence (Viterbi) -p3 train hidden Markov mode parameters (Baum-Welch) -n number of iterations
Configuration
The config
