regret metrics
Akshaysharma29 opened this issue · comments
Hi Team, Thanks for sharing this wonderful library. can you guide me on how to use regret metrics while using MAB?
Hey @Akshaysharma29
In the context of multi-armed bandits, regret is typically defined as the difference between the cumulative reward given the optimal decision at each time step and the cumulative reward observed for the selected policy. In practice, we typically do not know the optimal decisions though (otherwise we would just use those). In our implementation we therefore focus on finding policies that maximize the cumulative reward.
Make sense @bkleyn. Thank you 👍