piabrota← proof index

One round of a contextual bandit

One round of a contextual bandit A sequence diagram generated by Archify. context c estimate(b, c, a) for all a mean, width per arm estimates P(arm) show arm a ~ P reward r update(b, c, a, r) log (c, a, r, P(a)) Decide Act Learn Homepage · Member · Sequence participant Homepage Member Agent · policy · Sequence participant Agent policy RewardModel · belief b · Sequence participant RewardModel belief b DecisionRule · greedy, ε, UCB, TS · Sequence participant DecisionRule greedy, ε, UCB, TS Event log · (c, a, r, P(a)) · Sequence participant Event log (c, a, r, P(a)) Legend main request return async trace message