By Tim Hopper: tdhopper.com

Understanding Probabilistic Topic Models By Simulation

Overview

Description

Latent Dirichlet Allocation and related topic models are often presented in the form of complicated equations and confusing diagrams. I will present LDA as a generative model through probabilistic simulation in simple Python. Simulation will help data scientists to understand the model assumptions and limitations and more effectively use black box LDA implementations.

Abstract

Those without training in probabilistic graphical models and measure theory, data scientist may have a hard time understanding Latent Dirichlet Allocation and other probabilistic topic models. However, because LDA is a generative model, we can write Python code to generated data based on the model assumptions.

The talk will progress as follows:

• Introduction to mixture models
• Simulation of mixture models
• Introduction to grouped data
• Simulation of latent Dirichlet allocation
• Fitting and visualizing LDA with Python

Setup Conda Environment and Launch Notebook

With Conda installed, run

$ git clone https://github.com/tdhopper/pydata-nyc-2015.git understanding-lda
$ cd understanding-lda
$ make install
$ source activate understanding-lda

To view the notebook, run

$ make

To view the notebook as a slideshow, run

make slides