kyamagu / lda-matlab

Latent Dirichlet Allocation for Matlab

Latent Dirichlet Allocation for Matlab

This is a Matlab version of the David Blei's original implementation of Latent Dirichlet Allocation written in C.

http://www.cs.columbia.edu/~blei/lda-c/

The package includes a few API functions that internally calls the original C implementation by mex interface.

Compile the code with lda.make function in Matlab before using the package. Here is a quick usage example:

corpus = lda.load_corpus('ap/ap.dat');
[model, distribution] = lda.estimate(corpus, 'seeded', ...
                                     'num_topics', 100, ...
                                     'initial_alpha', 0.05);
distribution = lda.infer(corpus, model);

The corpus is sparse row vectors of word count. corpus(d, w) is a count of word w in document d. The resulting distribution contains row vectors of topic weights for each document at each row.

The same example is in the main.m demo function.

API

All functions are scoped under lda namespace.

estimate     Run the EM algorithm to estimate topics and distribution.
infer        Run inference on new samples using a learned LDA model.
load_corpus  Load dataset in lda-c file format into a sparse matrix.
save_corpus  Save dataset in lda-c file format from a sparse matrix.
make         Build a mex file.

Check help of each function for details.

License

The code may be redistributed under LGPL v2.1 license.