[41] Intro to Probabilistic Programming with PyMC (Austin Rochford)

Upcoming Events
Join our Meetup group for more events!
https://www.meetup.com/data-umbrella

Austin Rochford: Introduction to Probabilistic Programming with PyMC

Key Links
GitHub repo: https://pymc-data-umbrella.xyz/en/lat...

Resources
Jupyter Notebook: https://github.com/pymc-devs/pymc-dat...
Austin's website: https://austinrochford.com/talks.html
ArviZ: https://github.com/arviz-devs/arviz
PyMC Series of events: https://pymc-data-umbrella.xyz

Topics Covered
Probabilistic programming from two perspectives
-- Philosophical: storytelling with data
-- Mathematical: Monte Carlo methods
Probabilistic programming with PyMC
-- The Monty Hall problem
-- Robust regression
Hamiltonian Monte Carlo
-- Aesara
Lego example
Next Steps

Agenda
00:00 Reshama introduces Data Umbrella
04:40 Austin begins talk
06:15 Talk agenda
08:08 Probabilistic programming from two perspectives
08:53 What is probabilistic programming?
10:15 Mathematical: Monte Carlo Methods
13:55 Monty Hall Problem (game: Let's Make a Deal)
16:15 Solve Monty Hall Problem using PyMC (solution)
18:42 Using Aesara
21:00 Doing inference with sampling
24:00 What is Aesara? (It is based on Theano.) PyMC's tensor computational backend, fills niche such as PyTorch or TensorFlow.
25:20 Using PyMC to do robust regression: with example Anscombe's Quartet
28:10 Using ArviZ (library with pre-built visualizations and statistical routines that will help you understand the results of your inference with PyMC.
33:08 What is Ridge Regression? (normal priors on your coefficients)
36:05 Student-T Distribution
39:00 Why are we using Aesara? To do Hamiltonian Monte Carlo.
43:10 Bayesian Analysis of Lego Prices
49:00 Recommended books
50:37 Meenal talks about upcoming PyMC sprint
56:30 Q&A with Austin

Event

In the last ten years, there have been a number of advancements in the study of Hamiltonian Monte Carlo and variational inference algorithms that have enabled effective Bayesian statistical computation for much more complicated models than were previously feasible. These algorithmic advancements have been accompanied by a number of open source probabilistic programming packages that make them accessible to the general engineering, statistics, and data science communities. PyMC is one such package written in Python and supported by NumFOCUS. This talk gives an introduction to probabilistic programming with PyMC, with a particular emphasis on the how open source probabilistic programming makes Bayesian inference algorithms near the frontier of academic research accessible to a wide audience.

About the Speaker

Austin Rochford is the Chief Data Scientist at Kibo Commerce. He is a recovering mathematician and is passionate about math education, Bayesian statistics, and machine learning.

LinkedIn:   / austin-rochford  
Twitter:   / austinrochford  
GitHub: https://github.com/AustinRochford/