Installer crf++ avec Homebrew

Liens du projet

• Page d'accueil
• Dépôt

Commandes et alias

• crf_learn
• crf_test

Historique du projet

The official CRF++ page lists the first release, version 0.1, on 2005-05-28. The release history shows a steady 2005-2007 development period with API work, parallel training, MIRA training, language bindings, L1 regularization, and a license change to LGPL/BSD dual licensing.

Historique d'adoption

CRF++ became a familiar package for NLP users because it combined a command-line learner and tester, a template-based feature system, and bindings for several languages. Later releases focused on compatibility and maintenance, including GCC, libtool, C++11, Windows, and model-loading fixes.

Modes d'utilisation

Users prepare training and test files plus feature templates, train models with crf_learn, and decode with crf_test. The package is especially associated with sequence-labeling experiments where researchers want a compact C++ CRF implementation available from source or OS packages.

Pourquoi les passionnés de paquets s'y intéressent

For package maintainers, CRF++ is a classic NLP C++ library/CLI: old enough to need build-system care, small enough to package directly, and still recognizable because many downstream tutorials and research pipelines refer to crf_learn and crf_test.

Chronologie

• 2005-05-28: CRF++ 0.1 released
• 2006-03-30: Version 0.41 added parallel training
• 2006-11-26: Version 0.45 added 1-best MIRA training
• 2007-02-12: Version 0.46 changed license to LGPL/BSD dual license and added Perl, Ruby, Python, and Java bindings
• 2007-07-07: Version 0.48 added L1-CRF support
• 2013-02-13: CRF++ 0.58 released

Related projects

• CRF++ is related to later CRF sequence-labeling packages such as CRFsuite, which explicitly compares its data-format flexibility with CRF++.

Sources

• https://taku910.github.io/crfpp
• https://github.com/taku910/crfpp