R/effectTreatmentN.R

In vtreat: A Statistically Sound 'data.frame' Processor/Conditioner

# apply a numeric impact model
# replace level with .wmean(x|category) - .wmean(x)
.catNum <- function(col,args,doCollar) {
  col <- .preProcCat(col,args$levRestriction)
  unhandledNovel <- !(col %in% names(args$scores))
  keys <- col
  pred <- numeric(length(col))
  if(length(args$scores)>0) {
    keys[unhandledNovel] <- names(args$scores)[[1]]   # just to prevent bad lookups
    pred <- as.numeric(args$scores[keys]) 
  }
  # mean delta impact averaged over all possibilities, should be zero 
  pred[unhandledNovel] <- 0.0 
  pred
}

as_rquery.vtreat_can_num <- function(tstep, 
                                       ...,
                                       var_restriction) {
  if(!requireNamespace("rquery", quietly = TRUE)) {
    stop("vtreat::as_rquery.vtreat_can_num treatmentplan requires the rquery package")
  }
  wrapr::stop_if_dot_args(substitute(list(...)), "vtreat::as_rquery.vtreat_can_num")
  if((!is.null(var_restriction)) && (!(tstep$newvars %in% var_restriction))) {
    return(NULL)
  }
  args <- tstep$args
  rquery_code_categorical(colname = tstep$origvar, 
                          resname = tstep$newvars,
                          coding_levels = names(args$scores),
                          effect_values = args$scores,
                          levRestriction = args$levRestriction,
                          default_value = 0.0)
}


# build a numeric impact model
# see: https://win-vector.com/2012/07/23/modeling-trick-impact-coding-of-categorical-variables-with-many-levels/
.mkCatNum <- function(origVarName,vcolin,rescol,smFactor,levRestriction,weights) {
  vcol <- .preProcCat(vcolin,levRestriction)
  extraModelDegrees <- max(0,length(unique(vcolin))-1)
  baseMean <- .wmean(rescol,weights)
  num <- tapply(rescol*weights,vcol,sum)
  den <- tapply(weights,vcol,sum)
  scores <- as.list((num+smFactor*baseMean)/(den+smFactor)-baseMean)
  scores <- scores[names(scores)!='zap'] # don't let zap code
  newVarName <- vtreat_make_names(paste(origVarName,'catN',sep='_'))
  treatment <- list(origvar=origVarName,
                    newvars=newVarName,
                    f=.catNum,
                    args=list(scores=scores,
                              levRestriction=levRestriction),
                    treatmentName='Scalable Impact Code',
                    treatmentCode='catN',
                    needsSplit=TRUE,
                    extraModelDegrees=extraModelDegrees)
  pred <- treatment$f(vcolin,treatment$args)
  if(!.has.range.cn(pred)) {
    return(NULL)
  }
  class(treatment) <- c('vtreat_can_num', 'vtreatment')
  treatment$scales <- linScore(newVarName,pred,rescol,weights)
  if(treatment$scales$a <= 0) {
    return(NULL) # fitting a noise effect
  }
  treatment
}