policy_mab_softmax.R

#' @export
SoftmaxPolicy <- R6::R6Class(
  portable = FALSE,
  class = FALSE,
  inherit = Policy,
  public = list(
    tau = NULL,
    class_name = "SoftmaxPolicy",
    initialize = function(tau = 0.1) {
      super$initialize()
      self$tau <- tau
    },
    set_parameters = function(context_params) {
      self$theta_to_arms <- list('n' = 0, 'mean' = 0)
    },
    get_action = function(t, context) {
      exp_est_tau <- exp(unlist(self$theta$mean)/tau)
      p <- exp_est_tau/sum(exp_est_tau)
      action$choice <- categorical_draw(p)
      return(action)
    },
    set_reward = function(t, context, action, reward) {
      arm <- action$choice
      reward <- reward$reward
      inc(self$theta$n[[arm]]) <- 1
      inc(self$theta$mean[[arm]]) <- (reward - self$theta$mean[[arm]]) / self$theta$n[[arm]]
      self$theta
    },
    categorical_draw = function(probs) {
      arms <- length(probs)
      cumulative_probability <- 0.0
      z <- runif(1)
      for (i in 1:arms) {
        inc(cumulative_probability) <- probs[i]
        if ( cumulative_probability > z ) return(i)
      }
      return(arms)
    }

  )
)

#' Policy: Softmax
#'
#' \code{SoftmaxPolicy} is very similar to \link{Exp3Policy}, but selects an arm based on the probability from
#' the Boltmann distribution. It makes use of a temperature parameter tau,
#' which specifies how many arms we can explore. When tau is high, all arms are explored equally,
#' when tau is low, arms offering higher rewards will be chosen.
#'
#' @name SoftmaxPolicy
#'
#' @section Usage:
#' \preformatted{
#' policy <- SoftmaxPolicy(tau = 0.1)
#' }
#'
#' @section Arguments:
#'
#' \describe{
#'   \item{\code{tau = 0.1}}{
#'   double, temperature parameter tau specifies how many arms we can explore.
#'   When tau is high, all arms are explored equally, when tau is low, arms offering higher
#'   rewards will be chosen.
#'   }
#' }
#'
#' @section Methods:
#'
#' \describe{
#'   \item{\code{new(epsilon = 0.1)}}{ Generates a new \code{SoftmaxPolicy} object. Arguments are defined in
#'   the Argument section above.}
#' }
#'
#' \describe{
#'   \item{\code{set_parameters()}}{each policy needs to assign the parameters it wants to keep track of
#'   to list \code{self$theta_to_arms} that has to be defined in \code{set_parameters()}'s body.
#'   The parameters defined here can later be accessed by arm index in the following way:
#'   \code{theta[[index_of_arm]]$parameter_name}
#'   }
#' }
#'
#' \describe{
#'   \item{\code{get_action(context)}}{
#'     here, a policy decides which arm to choose, based on the current values
#'     of its parameters and, potentially, the current context.
#'    }
#'   }
#'
#'  \describe{
#'   \item{\code{set_reward(reward, context)}}{
#'     in \code{set_reward(reward, context)}, a policy updates its parameter values
#'     based on the reward received, and, potentially, the current context.
#'    }
#'   }
#'
#' @references
#'
#' Kuleshov, V., & Precup, D. (2014). Algorithms for multi-armed bandit problems.
#' arXiv preprint arXiv:1402.6028.
#'
#' Cesa-Bianchi, N., Gentile, C., Lugosi, G., & Neu, G. (2017). Boltzmann exploration done right.
#' In Advances in Neural Information Processing Systems (pp. 6284-6293).
#'
#' @seealso
#'
#' Core contextual classes: \code{\link{Bandit}}, \code{\link{Policy}}, \code{\link{Simulator}},
#' \code{\link{Agent}}, \code{\link{History}}, \code{\link{Plot}}
#'
#' Bandit subclass examples: \code{\link{BasicBernoulliBandit}}, \code{\link{ContextualLogitBandit}},
#' \code{\link{OfflineReplayEvaluatorBandit}}
#'
#' Policy subclass examples: \code{\link{EpsilonGreedyPolicy}}, \code{\link{ContextualLinTSPolicy}}
#'
#' @examples
#'
#' horizon            <- 100L
#' simulations        <- 100L
#' weights          <- c(0.9, 0.1, 0.1)
#'
#' policy             <- SoftmaxPolicy$new(tau = 0.1)
#' bandit             <- BasicBernoulliBandit$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")
NULL