seiichiinoue / hmm

implementation of simple hmm; hidden markov model with viterbi algorithm for part-of-speech tagging

Hidden Markov Model

the implementation of hidden markov model and viterbi algorithm with C++

explanation of this implementation(myblog): The implementation of Hidden Markov Model

Environment

• C++ 11+
• clang++ 10.0
• boost 1.71.0
• glog 0.4.0
• gflag 2.2.2

Usage

• if you want to use your additional data for training(optional, NOTE: if you want to use text processing script, you must install mecab-python3 package.)

$ python utils/text.py --tar_path data/[filename].txt --wakati_save_path data/[filename]-wakati.txt --pos_save_path data/[filename]-pos.txt

• training and predict pos with viterbi algorithm

$ make
$ ./hmm -ITER=200 > data/wiki-result.txt

• valuation (mapping minimizing error rate)

you can choose perl(original) script or python(program myself) script

$ ./utils/grade-bayes-hmm.pl data/wiki-pos.txt data/wiki-result.txt

or

$ python utils/valuation.py --ref data/wiki-pos.txt --test data/wiki-result.txt

• accuracy = 83.53%

Reference

• A Revealing Introduction to Hidden Markov Models
• neural networks for nlp
• position tagging using hidden markov model
• introduction to nonparametric bayes