Thompson Sampling

Thompson Sampling is a Bayesian approach to multi-armed bandits. This notebook reviews the theory walks through my implementation and some experiments. The experiments should give you some good understanding of the behaviour of Thompson Sampling in comparison to epsilon-greedy and UCB. To run the notebook online, click this link and open with Colab.

For a more extensive review of the theory, checkout A Tutorial on Thompson Sampling by Russo et al., 2017.