Multi-Armed Bandit Simulator

This repository contains a notebook that visualizes different Multi-Armed Bandit algorithms in the context of online advertising. You can play a game and manually optimize a virtual campaign. Your results are compared to the different MAB algorithms.

Content

You can find an interactive explanation of the most popular Multi-Armed Bandit algorithms $\epsilon$-greedy, Thompson Sampling, Upper Confidence Bounds) in the notebook multi-armed-bandit-algorithms.ipynb. In optimization_simulator.ipynb, you can manually optimize a campaign and compete with the implemented algorithms. beta-distribution-plotter.ipynb contains an interactive plotter based on the bokeh library to visualize the $Beta$ distribution, which is the conjugate prior to the Binomial distribution. It visualizes the impact on clicks & bounces on the posterior distribution over click rates.

Citation

@article{Grigo2021,
title = "Multi-Armed Bandit Simulator",
year = "2021",
url = "https://github.com/congriUQ/multi-armed-bandits",
author = "Constantin Grigo",
keywords = "Multi-Armed Bandits, Bayesian Inference, Thompson Sampling, Upper Confidence Bounds, Reinforcement Learning",
}