R/Simulator.Simple.R

In frbl/OnlineSuperLearner: Online SuperLearner package

#' SimpleSimulator
#'
#' Very basic simulator to generate data from a simple binary model
#'
#' @docType class
#' @importFrom R6 R6Class
#' @importFrom stats glm
#' @field model the most recent / best model fitted.
#'
#' @section Methods:
#' \describe{
#'   \item{\code{new()}}{Creates a new simulator}
#'
#'   \item{\code{fit(nr.of.observations)}}{Generate a number of observations from the model (multple observations can be retrieved at once by using the \code{nr.of.observations} param}
#'}
Simulator.Simple <- R6Class("Simulator.Simple",
  private =
    list(
        model = NULL
        ),
  public =
    list(
        initialize = function() {
          nobs <- 1000
          # We have x1 and x2 to create a variable y
          # some continuous variables
          x1 <- rnorm(nobs)
          x2 <- rnorm(nobs, mean=10)

          # linear combination with a bias
          z <- 1 + 2*x1 + 1.2*x2

          # Pass through the inverse logit function
          pr <- plogis(z)

          # Generate binary outcomes using this
          y <- rbinom(nobs, 1, pr)

          noise <- rnorm(nobs, mean = 0, sd = 1)
          df <- data.frame(y = y, x1 = x1, x2 = x2, noise = noise)


          # Store the model, so we can simulate from it later
          private$model <- glm( y ~ x1 + x2 + noise, data = df, family = "binomial")
        },

        getObservation = function(nr.of.observations = 1) {
          x1 <- rnorm(nr.of.observations)
          x2 <- rnorm(nr.of.observations)
          noise <- rnorm(nr.of.observations, mean = 0, sd = 1)
          df <- data.table(x1 = x1, x2 = x2, noise=noise)
          y <- plogis(predict(private$model, df))
          cbind(df, y)
        }
    )
)