README.md

crf_simple

This is my first GitHub project, really? Yes!

this python files are a simple implementation of line_chain crf to POS tagging. this project contains four python files currently. They are:

1) logspace.py

this class contains two method to calculate matrix product in log space.

2) crf.py

this file implements several methods to calculate joint probability distribution, matrginal probability distribution, inference, learning( max likelihood).

3) crf_main.py

it runs as a overall framework to train crf.

4) L_BFGS.py

when to learning parameters, we need to minimum the negative log_likelihood function of dataset, because it has large amount of parameters, so we need the limited-memery BFGS algorithm to optimize the likelihood function. this file implements the l-bfgs optimization algorithm.

5) train.txt and test.txt

this two file are train dataset and test dateset.