Overview

This package implements Latent Dirichlet Allocation fitted via loopy belief propagation as described in Zeng et al. Learning topic models using belief propagation. 2012, IEEE Transations on pattern analysis and machine intelligence.

The package is available under a GPL-v3 license.

Installation

You can install the package using

pip install https://github.com/wkopp/bplda/archive/master.zip

Getting started

import numpy as np
from bplda import BeliefProbLDA

# toy data (10 documents, vocabulary size=5)
X = np.zeros((5, 10))
X[:3,:5]=1
X[-3:,-5:]=1

model = LDA(3, niter=10, seed=10)

# fit and get document-topic matrix
doc_top = model.fit_transform(minitest)

# access word-topic matrix
model.word_topic_

# compute the log-likelihood score
model.score(minitest)

model.loglikeli_