Tutorial: Up and down the PyData Stats Stack

Or Lies damned lies and statistics

Peadar Coyle

I work as a Senior Data Scientist at Channel 4

We're a media company, and we leverage data for targeted advertising, customer analytics and recommendation engines

We're hiring, so have a chat with me if interested.

Who am I?

What else do I do?

• Open Source Contributor to PyMC3
• Mathematics and Physics background
• Fellow of the Royal Statistical Society and member of NumFOCUS
• Author of interviews with data scientists book :) 

Stats is everywhere...

What was wrong with this?

http://www.bbc.co.uk/programmes/b075thgl

More or Less debunked this - the sample was wrong!

Outcomes of this tutorial

• Have three tools to attack the same problem
• Some tricks and tips like hypothesis testing and feature selection
• Understand how to interpret and debug three versions of Logistic Regression. ML, Frequentist Stats and Bayesian

http://bit.ly/pydataldnstats2

Different schools of data 

HT: Vincent D. Warmerdam

Firstly hypothesis testing

• Or how to do t-tests and all that stuff

Three schools

• ScikitLearn 
• Statsmodels
• PyMC3

Bayesian tooling

PyMC3

https://pymc-devs.github.io/pymc3/Bayesian_LogReg/ 

How likely am I to make more than $50K?

Statsmodels

Frequentist Logistic Regression