O'Reilly logo

Probabilistic Programming from Scratch 3: Performance and PyMC3

Mike Lee Williams reveals the flaw in ABC: Performance! He makes some fixes, uses it to solve a famous puzzler, then offers up a faster system: PyMC3

Fast Forward Labs' Mike Lee Williams built a probabilistic programming system from scratch using the Approximate Bayesian Computation (ABC) algorithm and Python. ABC is a brute force algorithm: It works, but it's slow. In this segment, he speeds it up, shows how it can solve the famous "German tank problem", then cracks the tank puzzle more quickly with an off-the-shelf probabilistic programming system called PyMC3.

What you will learn:

  • Solidify your understanding of what probabilistic programming truly means
  • Deepen your awareness of the Bayesian method's value in probabilistic programming
  • Review how ABC quantifies risk, inserts institutional knowledge, and handles on-line learning
  • Compare the qualities of a custom-built ABC based system vs. off-the-shelf PyMC3