LDA using collapsed gibbs sampling is implemented.
To use the LDA as a module, please use the TopicModels julia package.
Steps to use TopicModels package:
git clone #clone the repo
cd handson_julia #cd to parent directory of TopicModels folder
Pkg.activate("TopicModels") #activate the package in julia REPL
corpus = TopicModels.documentset_readData("news.txt") #import the document collection
wordPrior = TopicModels.Dirichlet(corpus.vocab_count, 0.01) #dirichlet word prior
M = 3 #number of topics
alpha = [0.01 for i in 1:M]
topicPrior = TopicModels.Dirichlet(alpha); #dirichlet topic prior
lda = TopicModels.LDA(topicPrior, wordPrior) #build LDA struct using word prior and topic prior
samples = TopicModels.lda_sample(corpus.documents, lda) #run LDA with collapsed gibbs sampling
words, proportions = TopicModels.lda_topicN(1, 10, corpus, lda) #top 10 words and topic proportions of Topic 1
LDA on dummy news dataset as well as on NIPS paper dataset is applied. Implementation and results can be viewed in LDA_with_package.ipynb notebook.
To view the complete implementation in jupyter notebook, please have a look at try_LDA.ipynb