surveillance (version 1.5-4)

estimateGLRNbHook: Hook function for in-control mean estimation

Description

Allows the user to specify his own estimation routine for the in-control mean of algo.glrpois Needs to work for GLRNbHook

Usage

estimateGLRNbHook()

Value

• A list
• mod
  resulting model of a call of glm.nb
• range
  vector of length as range containing the predicted values

encoding

latin1

Details

This hook function allows the user to customize the behaviour of the algorithm.

See Also

algo.glrnb and algo.glrpois

Examples

estimateGLRNbHook <- function() {
  #Fetch control object from parent
  control <- parent.frame()$control
  #The period
  p <- parent.frame()$disProgObj$freq
  #Current range to perform surveillance on
  range <- parent.frame()$range

  #Define training & test data set (the rest)
  train <- 1:(range[1]-1)
  test <- range
  
  #Perform an estimation based on all observations before timePoint
  #Event better - don't do this at all in the algorithm - force
  #user to do it himself - coz its a model selection problem
  data <- data.frame(y=parent.frame()$disProgObj$observed[t],t=train)
  #Build the model equation
  formula <- "y ~ 1 "
  if (control$mu0Model$trend) { formula <- paste(formula," + t",sep="") }
  for (s in 1:control$mu0Model$S) {
    formula <- paste(formula,"+cos(2*",s,"*pi/p*t)+ sin(2*",s,"*pi/p*t)",sep="")
  }
  #Fit the GLM
  m <- eval(substitute(glm.nb(form,data=data),
                      list(form=as.formula(formula))))

  #Predict mu_{0,t}
  return(as.numeric(predict(m,newdata=data.frame(t=range),type="response")))
}