For non-stationary bandits, the following reference introduce the awesome LM-DSEE and SW-UCB# algorithms: ["On Abruptly-Changing and Slowly-Varying Multiarmed Bandit Problems", by Lai Wei, Vaibhav Srivastava, 2018, arXiv:1802.08380].

I need to:

• Write the LM-DSEE algorithm!
• Write the SW-UCB# algorithm!
• Implement it in a very generic way, as children of BaseWrapperPolicy, so any index policy can directly be used and not just UCB!
• Test it on simple problems, and check what the authors claim in their paper,
• Compare them with DiscountedUCB, SlindingWindow-UCB or SW-klUCB or SWRestart-UCB.