An Introduction to Bayesian Statistics

This is an introduction on how to use Bayesian Statistics to run a binary outcome A/B test. The notebook Intro_to_Bayesian_Stats.ipynb uses PyMC4 for generating samples and probability. Included in this notebook is some of the most commonly used visualizations for an A/B test.

Setup

1. Create virtual environment

Once the material is on your computer, you'll see that the repository for this course has a file called environment.yml that includes a list of all the packages you need to install. If you run:

conda env create -f environment.yml

This will create a new conda environment called bayes.

conda activate bayes

2. Setup and launch Jupyter

Next, make Jupyter aware of this new virtual environment. With the bayes environment activated, run:

python -m ipykernel install --user --name bayes

That will create what's called a kernel in Jupyter linguo (this is just a mirror of your virtual environment). Then, you can start Jupyter Lab or Notebook to access the materials:

jupyter lab

When you open any notebook, make sure it's using the right kernel, which will be named bayes. You can check this at the top-right of the Jupyter page. Select bayes if you're in another python kernel.

 

Figure 1: Kernel Selection in Jupyter Notebook

Note: The presentation mode for RISE is only available in Jupyter Notebook and not in Jupyter Lab.

 

Figure 2: Kernel Selection in Jupyter Lab

 

Your setup is done. You can start sampling!

References & Thanks

Many of the techniques and code came from the Causal Inference Book Club, https://ravinkumar.com/BookClub2022.html and https://github.com/canyon289/causal_inf_bookclub

Thanks to Ravin Kumar for his enthusiastic attitude towards causual inferencing. Check out his book with his co-authors: https://bayesiancomputationbook.com/welcome.html

And, one of many text books that inspired me is here: https://mixtape.scunning.com/ and https://github.com/scunning1975/mixtape