In hiatus. Currently evaluating best way to approach the DTM model. Using the Pólya-Gamma augmentation and the original DTM formulation is complicated and might not give better performance than simpler models (e.g., using truncated Pitman-Yor Processes).
NOTE: The implementation of LDA has been broken out (and refined) into lda.
NOTE: If you're interested in implementing the dynamic topic model using Pólya-Gamma, most of the hard work has been done: https://github.com/HIPS/pgmult
horizont implements a number of topic models. Conventions from scikit-learn are followed.
The following models are implemented using Gibbs sampling.
- Latent Dirichlet allocation (Blei et al., 2003; Pritchard et al., 2000)
- (Coming soon) Logistic normal topic model
- (Coming soon) Dynamic topic model (Blei and Lafferty, 2006)
horizont.LDA
implements latent Dirichlet allocation (LDA) using Gibbs
sampling. The interface follows conventions in scikit-learn.
>>> import numpy as np
>>> from horizont import LDA
>>> X = np.array([[1,1], [2, 1], [3, 1], [4, 1], [5, 8], [6, 1]])
>>> model = LDA(n_topics=2, random_state=0, n_iter=100)
>>> doc_topic = model.fit_transform(X) # estimate of document-topic distributions
>>> model.components_ # estimate of topic-word distributions
Python 2.7 or Python 3.3+ is required. The following packages are also required:
- numpy
- scipy
- scikit-learn
- futures (Python 2.7 only)
GSL is required for random number
generation inside the Pólya-Gamma random variate generator. On Debian-based
sytems, GSL may be installed with the command sudo apt-get install
libgsl0-dev
. horizont looks for GSL headers and libraries in /usr/include
and /usr/lib/
respectively.
Cython is needed if compiling from source.
- Documentation: http://pythonhosted.org/horizont
- Source code: https://github.com/ariddell/horizont/
- Issue tracker: https://github.com/ariddell/horizont/issues
horizont is licensed under Version 3.0 of the GNU General Public License. See LICENSE file for a text of the license or visit http://www.gnu.org/copyleft/gpl.html.