Markov Chain Monte Carlo Simulator App

Background:

I was inspired by this excellent Medium article by Fortunato Nucera on the subject of using Metropolis Hastings for Bayesian Regression and set about trying to implement a variation of this myself. Additionally, I thought it would make for an interesting challenge to animate the results in streamlit!

In this project I wanted to:

• implement the Metropolis Hastings algorithm for a Simple Linear Regression model.
• create a streamlit app.
• test and deploy this application.

TODOS:

• ✨Style✨ the main page.
• Make parameters such as the number of burn in samples configurable in the side bar.
• Extend this to Bayesian Multiple Linear Regression.
• Deploy this on a Kubernetes instance (e.g. Minikube).

Installation:

# Setup for development
pip install pipenv
pipenv shell 
pipenv install --dev

Usage:

pipenv shell 
streamlit run main.py

If you want to generate new data, run the following command:

python scripts/generate_data.py

Testing:

This package uses Tox to manage testing across multiple environments - see the tox.ini file. Testing of the MCMC logic uses pytest and hypothesis.

# Example tox commands
tox # run all
tox -e py310 # test in python 3.10
tox -e lint

# Run tests alone
pytest