scrapinghub / python-crfsuite

A python binding for crfsuite

python-crfsuite

python-crfsuite is a python binding to CRFsuite.

Installation

Using pip:

pip install python-crfsuite

Using conda:

conda install -c conda-forge python-crfsuite

Usage

See docs and an example.

See Also

sklearn-crfsuite is a python-crfsuite wrapper which provides API similar to scikit-learn.

Contributing

• Source code: https://github.com/scrapinghub/python-crfsuite
• Issue tracker: https://github.com/scrapinghub/python-crfsuite/issues

Feel free to submit ideas, bugs reports, pull requests or regular patches.

In order to run tests, install Cython (> 0.24.1) and tox, then type

./update_cpp.sh; tox

from the source checkout.

Please don't commit generated cpp files in the same commit as other files.

Authors and Contributors

Original authors are Terry Peng <[e-mail]> and Mikhail Korobov <[e-mail]>. Many other people contributed; some of them can be found at github Contributors page.

Bundled CRFSuite C/C++ library is by Naoaki Okazaki & contributors.

License

python-crfsuite is licensed under MIT license. CRFsuite library is licensed under BSD license.

Alternatives

• https://github.com/chokkan/crfsuite/tree/master/swig/python - official Python wrapper, exposes C++ API using SWIG.
• https://github.com/jakevdp/pyCRFsuite - uses C API instead of C++ API; allows to use scipy sparse matrices as an input. At the time of writing it is unmaintained.
• https://github.com/bosondata/crfsuite-rs - uses a Rust wrapper with CFFI instead of C++ API; allows to tag with GIL released for better performance.

This package (python-crfsuite) wraps CRFsuite C++ API using Cython. It is faster than official SWIG wrapper and has a simpler codebase than a more advanced pyCRFsuite. python-crfsuite works in Python 2 and Python 3, doesn't have external dependencies (CRFsuite is bundled, numpy/scipy stack is not needed) and workarounds some of the issues with C++ CRFsuite library.