rungsted

Rungsted. An efficient HMM-based structured prediction model for sequential labeling tasks, with extras.


Keywords
hmm, perceptron, structured_model
License
MIT
Install
pip install rungsted==1.2.4

Documentation

Build Status

Rungsted structured perceptron sequential tagger

Install

The software is installable via PyPI, e.g. do

pip install rungsted

At the moment, Rungsted only works on Python 3.

Demo

The repository contains a subset of the part-of-speech tagged Brown corpus. To run the structured perceptron labeler on this dataset, execute:

rungsted --train data/brown.train --test data/brown.test.vw

Rungsted's input format is closely modeled on the powerful and flexible format of Vowpal Wabbit, with the exception that Rungsted is perfectly fine with labels that are not integers.

Datasets

Provided you have a working installation of NLTK, you can recreate the Brown dataset with this command.

python rungsted/datasets/cr_brown_pos_data.py data/brown.train.vw data/brown.test.vw

There is also a script rungsted/datasets/conll_to_vw.py to convert from CONLL-formatted input to Rungsted

Building and uploading to PyPI

First, run CYTHON=1 python setup.py sdist to generate a source distribution. Then upload the distribution files to PyPI with twine: twine upload dist/*.

To develop locally, use CYTHON=1 python setup.py develop.