R/prediction_with_pp_La.R

In mengluchu/APMtools: statistical air pollution mapping tools

#' Aggregate random forest using lasso, implementing the idea from the Hastie's book
#' @param rfmodel ranger model
#' @param trainingXY training matrix, with same response and predictor names as in the rfmodel
#' @param testingX  testing matrix, with same response presented in the rfmodel
#' @param trainingY training response vection
#' @export

prediction_with_pp_La = function(rfmodel, trainingXY, trainingY, testingX)
{
  allp = predict(rfmodel,trainingXY , predict.all = T)%>%predictions #get all the tree predictions, instead of the mean
  cvfit <- glmnet::cv.glmnet(allp,trainingY ,
                             type.measure = "mse", standardize = TRUE, alpha = 1,
                             lower.limit = 0)
  # aggregate using a regularization, here lasso, you can also do elastic net, training alpha or specify alpha between 0 and 1
  rpre= predict(rfmodel,testingX, predict.all=T)%>%predictions # get all the tree predictions
  # prediction
  pred = predict(cvfit, newx = rpre) # use the regularization (here lasso) model to predict
  # get the entire distribution
  Ls= lassoselected(cvfit)
  Ls_num= as.vector(sapply(Ls, function(x) as.numeric(substr(x, start =2, stop = nchar(x)))))

  # aggregating trees using lasso, compare with original random forest, obtained better results

  reduced_rf = rpre[,Ls_num] # 62 trees

  return(list(pred= pred, dist= reduced_rf))
}