UCB policy for bounded bandits with a Chernoff-Hoeffding Bound
UCB1Policy constructs an optimistic estimate in the form of an Upper Confidence Bound to
create an estimate of the expected payoff of each action, and picks the action with the highest estimate.
If the guess is wrong, the optimistic guess quickly decreases, till another action has
the higher estimate.
policy <- UCB1Policy()
new()Generates a new UCB1Policy object.
set_parameters()each policy needs to assign the parameters it wants to keep track of
to list self$theta_to_arms that has to be defined in set_parameters()'s body.
The parameters defined here can later be accessed by arm index in the following way:
theta[[index_of_arm]]$parameter_name
get_action(context)here, a policy decides which arm to choose, based on the current values of its parameters and, potentially, the current context.
set_reward(reward, context)in set_reward(reward, context), a policy updates its parameter values
based on the reward received, and, potentially, the current context.
Lai, T. L., & Robbins, H. (1985). Asymptotically efficient adaptive allocation rules. Advances in applied mathematics, 6(1), 4-22.
Core contextual classes: Bandit, Policy, Simulator,
Agent, History, Plot
Bandit subclass examples: BasicBernoulliBandit, ContextualLogitBandit,
OfflineReplayEvaluatorBandit
Policy subclass examples: EpsilonGreedyPolicy, ContextualLinTSPolicy
if (FALSE) { horizon <- 100L simulations <- 100L weights <- c(0.9, 0.1, 0.1) policy <- UCB1Policy$new() bandit <- BasicBernoulliBandit$new(weights = weights) agent <- Agent$new(policy, bandit) history <- Simulator$new(agent, horizon, simulations, do_parallel = FALSE)$run() plot(history, type = "cumulative") plot(history, type = "arms") }