python - Is PyMC3 useful for creating a latent dirichlet allocation model? -
i've spent last several weeks trying learn pymc whereby main task using build lda topic model. tried example pymc2.3 https://stats.stackexchange.com/questions/104771/latent-dirichlet-allocation-in-pymc simple model ran overnight , never made sampling step. i've switched pymc3.
is there fundamental limitations because random variables categorical? has ever succeeded in creating lda model pymc3? found partial implementation @ unable create lambda function in hierarchical pymc3 model couldn't work without container, , don't think original author able either. know of resources study in order figure out how build this?
tl;dr implementation given in link works , full code testing on inaugural speech corpus can seen at: https://github.com/napsternxg/ipython-notebooks/blob/master/pymc_lda.ipynb
i implemented solution specified in link mention using pymc2 , got work on inaugural speech dataset. not confident of correctness of solution provided @ link mention implementation works , gives topic distribution. however, getting interpret implementation more suited understands mathematical definition of lda better.
Comments
Post a Comment