Variational Inference in Python
summary
Variational inference is presented as an optimization alternative to MCMC with concrete tooling and results: Edward (TensorFlow) and PyMC3 (Theano) run notebooks (with Docker) to do BBVI and ADVI, respectively; a beta–binomial example converges in ~10,000 VI iterations versus 50,000 MCMC samples; a congressional ideal‑points model cleanly separates Democrats and Republicans; a truncated stick‑breaking mixture (K≈10) fits heteroscedastic data and produces posterior predictive draws, demonstrating practical scalability and reproducible code.
AI-written summary · about
chapters (10)
- 0:00Variational inference intro
- 3:10Bayesian inference basics
- 7:27Variational inference fundamentals
- 13:22Automating VI in Python
- 15:47Blackbox VI with Edward
- 19:41Political modeling with Edward
- 24:16Automatic diff VI in PyMC3
- 26:22Bayesian nonparametrics in PyMC3
- 31:37Summary
- 32:39Q&A
related talks
35mVariational Inference and Python
01h 29mA Beginner's Guide to Variational Inference
21mProbabilistic Programming with Programmable Variational Inference
39mProbablistic Programming Data Science with PyMC3
38mBlack-Box Variational Inference for Probabilistic Programs
01h 22mDeep Probabilistic Methods with PyTorch
12mBayesian Neural Networks With Variational Inference in PyTorch
44mAn introduction to PyMC3