Latent Dirichlet Allocation

Code for HKUST MATH 5472 Final Project

Python implementation of LDA based on Latent Dirichlet Allocation

Usage

from preprocess import *
from main import *

corpus = preprocessing(M=200)

preprocess.py provides a text-preprocessing for American Press corpus, returns a list where each element represents a document coded by {0,1}. Preprocess the first M documents in AP corpus.

alpha, beta = LDA.parameter_estimation(corpus, k=10, tol=1e-6, max_iter=100)

LDA.parameter_estimation performs variantial inference EM to estimate Dirichlet parameter alpha, and word probability beta. Number of topics k should be given.

Examples

Check the notebooks sim_data.ipynb and ap_modeling.ipynb to play the examples in report lda.pdf.