Figure 2.

Regrets and suboptimal draw counts obtained using context-free multi-arm-bandit and contextual-bandit approaches for the IST database. Treatments were selected using random assignment, the UCB approach, and Thompson sampling. Plots include mean values of regrets and suboptimal draw counts and 25th percentile confidence intervals. (a) Regret: context-free multi-arm bandit. (b) Regret: contextual bandit. (c) Suboptimal draws: context-free multi-arm bandit. (d) Suboptimal draws: contextual bandit.