R/GOF_model_simulator.R

In bootGOF: Bootstrap Based Goodness-of-Fit Tests

##' @title R6 Class representing a generator/resample of the dependent variable
##'
##' @description R6 does not offer interfaces. Hence all methods
##'   are considered as abstract.
##' @export
GOF_model_simulator <- R6::R6Class(
  classname = "GOF_model_simulator",
  public = list(
    ##' @description Abstract function that resamples/generates
    ##'   the dependent variable
    ##' @param model fitted model
    ##' @return generates the dependent variable according to
    ##'   the \code{model}
    resample_y = function(model) {
      stop("Abstract method. Needs to be implemented")
    }))

##' @title Implements the "interface" GOF_model_simulator for
##'   for linear models
##' @description after the classical linear model was fitted the
##'   normal distribution of the
##'   of the dependent variable is fully specified and used here
##'   to generate new depenedent variables that follow \code{model}
##' @export
GOF_lm_sim_param <- R6::R6Class(
  classname = "GOF_lm_sim_param",
  ##' @description generates/resamples the dependent variables based
  ##'   on the parameteric nature defined by \code{model}
  ##' @param model see \link{GOF_model_simulator}
  ##' @return see \link{GOF_model_simulator}
  public = list(
    resample_y = function(model) {
      simulate(model)[,1]
    }))


##' Generates Rademacher distributed random variables
##'
##' @param n number of random variables to be generated
##' @return vector of values following the Rademacher distribution
rrademacher <- function(n) {
  ret <- 2 * stats::rbinom(n = n, size = 1, prob = 0.5) - 1
  return(ret)
}

##' @title Implements the "interface" GOF_model_simulator
##'   in a semi-parametric fashion
##' @description This is a model agnostic resampling class, where
##' Rademacher random variables are used to add or substract
##' the residuals from the fitted values.
##' @export
GOF_sim_wild_rademacher <- R6::R6Class(
  classname = "GOF_sim_wild_rademacher",
  public = list(
    ##' @param gof_model_info_extractor the info extractor that is used
    ##'   to derive the residuals and fitted values for resampling.
    initialize = function(gof_model_info_extractor) {
      private$model_info_extractor = gof_model_info_extractor
    },
    ##' @description a wild bootstrap using Rademacher random
    ##'   variables to resample the dependent variable
    ##' @param model see \link{GOF_model_simulator}
    ##' @return see \link{GOF_model_simulator}
    resample_y = function(model) {
      eps <- private$model_info_extractor$y_minus_yhat(model = model)
      yhat <- private$model_info_extractor$yhat(model = model)
      r <- rrademacher(n = length(eps))
      ret <- yhat + r * eps
      return(ret)
    }
  ),
  private = list(
    model_info_extractor = NULL
  )
)


##' @title Implements the "interface" GOF_model_simulator for
##'   for generalized linear models
##' @description after the GLM was fitted the distribution of the
##'   of the dependent variable is fully specified and used here
##'   to generate new depenedent variables that follow \code{model}
##' @export
GOF_glm_sim_param <- R6::R6Class(
  classname = "GOF_glm_sim_param",
  public = list(
    ##' @description see \link{GOF_model_simulator}
    ##' @param model see \link{GOF_model_simulator}
    ##' @return see \link{GOF_model_simulator}
    resample_y = function(model) {
      simulate(model)[,1]
    }))