thinc: Learn super-sparse multi-class models
thinc is a Cython library for learning models with millions of parameters and dozens of classes. It drives http://honnibal.github.io/spaCy , a pipeline of very efficient NLP components. I've only used thinc from Cython; no real Python API is currently available.
Currently the only model implemented is the averaged perceptron, which is surprisingly competitive for these problems.
Despite the recent enthusiasm for deep learning, linear models can still perform very well, if the right feature engineering is applied. The key is adding good conjunction features --- e.g., "next_word=X && next_next_word=Y". For this, I have a helper-class thinc.features.Extractor, which you pass a list of templates, which then performs your feature extraction, given an array of atomic context items.
thinc was written as part of the development of spaCy, which is dual-licensed: GPL v3, or you can pay for a commercial license. thinc is licensed in the same way. For a commercial license, contact firstname.lastname@example.org
Copyright (C) 2014 Matthew Honnibal
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>.