An Introduction to Bayesian Statistics for the Python Programmer (no data science experience needed)
Christian Stucchio (~christian) |
The legendary economist Paul Samuelson once said, “When my information changes, I alter my conclusions. What do you do, sir?” When new data comes along, a smart man will change his opinion in response to it, rather than clinging to his old opinion when it no longer fits the facts.
Bayesian statistics is the mathematically optimal way to change one’s opinions in response to new data.
In this talk I’ll provide a pythonic, rather than mathematical, introduction to Bayesian statistics. The target audience is programmers, not data scientists. I’ll show how a numpy array can be used to represent our opinion about the world. I’ll then show how we can use Bayesian statistics to change our opinion when we gain new information.
A number of examples will be used for illustration, including disease prediction, compass calibration and detecting errors in production systems.
- Listener should be familiar with python and arrays. Maybe a little numpy.
- Understand the general idea of a mathematical model.
- In the second half of the talk, it might be helpful (though not required) if the listener remembers the big picture of integrals from Calculus 2.
I have not yet written slides for the talk. The level of simplicity will be along these lines: https://www.chrisstucchio.com/pubs/slides/helpshift_2014/slides.html#1
I plan to draw examples (most likely with a LOT of simplification) from previous blog posts I've written:
Chris Stucchio is the Director of Data Science at Wingify (the parent company of Visual Website Optimizer). He’s had a variety of careers, including physicist, stock trader, call center worker, bodyguard and demolition worker. He’s been using python for scientific computing since the Numarray days (anyone else remember life before Numpy?). He’s a big proponent of careful reasoning and wants to teach the world about the best tool to do that.