Probabilistic Programming with Python

Erik Sjoberg

Audience level:
Libraries and Extensions / ライブラリや拡張


Discussion of the basics and applications of probabilistic programming systems (PPS), as well as the capabilities and limitations of PPS with Python libraries such as PyMC, STAN, and Venture. 確率プログラミングの紹介と応用、およびPythonライブラリの現状


I will introduce the central concepts and ideas behind probabilistic programming systems (PPS), and introduce several Python packages that can be used to write probabilistic programs. PPS seek to create a programming environment that allows non-ML-experts to apply Bayesian statistics to solve complex statistics and machine-learning problems. By removing the need to write custom inference code, PPS systems have the potential to empower experts and non-experts alike to perform machine learning and statistical modeling with vastly simpler code. The current state of the art allows for rapid prototyping of complex models on reasonably sized datasets.