alteryx / crossvalidation Goto Github PK

The model call appears to not store the Laplace smoothing parameter from Naive Bayes models. Thus, the tool currently errors when these models are used. We need to figure out an appropriate way to deal with this issue. Some ideas include:

1. Re-factor the Alteryx Naive Bayes tool to add the Laplace parameter and use it in the code.
2. Skip NB models and warn the user that they're not being examined in the C-V tool.
3. Assume that the Laplace parameter is 0 (or some other value) and warn the user that this assumption is being made.

I think, we should let the user set a seed for reproducibility purposes. Without a seed, the results will vary every run and sometimes that is not desirable.

This refers to these lines. We can replace this safely by this function. The key advantage is that you only need to update the package map when support is added for more models.

#' Load packages based on a set of model classes. 
#' 
#' @param models list of model objects
#' @export
getPkgListForModels <- function(models, load = FALSE){
  modelClasses <- unlist(lapply(models, class))
  pkgMap = list(
    gbm = "gbm", rpart = "rpart", svm.formula = "e1071", svm = "e1071",
    naiveBayes = "e1071", svyglm = "survey", nnet.formula = "nnet",
    randomForest.formula = "randomForest", earth = "earth"
  )
  pkgsToLoad <- unique(unlist(pkgMap[modelClasses]))
  if (load){
    for(pkg in pkgsToLoad) library(pkg, character.only = TRUE)
  } 
  pkgsToLoad
}

To Do

• update checkXvars and uncomment it.
• fix error while running classification models. (expects target to be first column)