PyMC4: Exploiting Coroutines for Implementing a Probabilistic Programming FrameworkDownload PDF

Sep 16, 2019 (edited Sep 16, 2019)NeurIPS 2019 Workshop Program Transformations SubmissionReaders: Everyone
  • Keywords: python, statistical-analysis, bayesian inference, mcmc, variational-inference, probabilistic-programming
  • TL;DR: PyMC4: A probalilistic programming language on top of dynamic graphs and utilizing coroutines
  • Abstract: PyMC4 is an open-source probabilistic programming library whose goal is to give users access to cutting-edge algorithms in Bayesian statistical computing while being extensible enough to help researchers to implement novel algorithms. Like its predecessor PyMC3 and the C++ library Stan, the aim of PyMC4 is to provide users with a high-level API to specify probabilistic models. Our focus is on Bayesian models where inference can be performed using (dynamic) Hamiltonian Monte Carlo and variational inference, both of which require automatic differentiation of user-defined joint density functions.
3 Replies