Santa competition 2020
Over the last few years, a sort of tradition has emerged on Kaggle: in early December, there is a Santa-themed competition. The actual algorithmic side varies from year to year, but for our purposes, the 2020 competition is an interesting case: https://www.kaggle.com/c/santa-2020.
The setup was a classical multi-armed bandit (MAB) trying to maximize reward by taking repeated action on a vending machine, but with two extras:
- Reward decay: At each step, the probability of obtaining a reward from a machine decreases by 3 percent.
- Competition: You are constrained not only by time (a limited number of attempts) but also by another player attempting to achieve the same objective. We mention this constraint mostly for the sake of completeness, as it is not crucial to incorporate explicitly in our demonstrated solution.
For a good explanation of the methods for approaching the general MAB problem, the reader is referred to https...
