Probabilistic Machine Learning with PyMC3


Abstract: Probabilistic Programming is a powerful set of tools which is barely sampled from by the traditional Machine Learning toolbox. PyMC3 is a popular Python library which leverages this power. Learn more about the library – and Probabilistic Programming – from one of the project’s core contributors.

Introduction: Dr. Thomas Wiecki of Quantopian spoke at the recent ODSC UK conference compared traditional Machine Learning with Probabilistic Programming. He first drew a broad comparison of the two fields across points such as model specification, evaluation metrics, regularization, and the structure of solutions. This portion of the presentation highlighted the advantage produced by the flexibility of Probabilistic Programming, primarily due to its intrinsic appreciation of uncertainty.

After this overview, Dr. Wiecki’s focused on Probabilistic Programming with PyMC3. The mature Python library contains a wide range of powerful sampling algorithms and variational inference methods. The general code samples that followed concentrated on variational inference methods, and served as a warm up for a Quantitative Finance use case.

The crowd responded enthusiastically to to the results produced within this application. It was a perfect way to cap off a riveting display of the power of Probabilistic Programming.