levels of categorical variables not preserved ?

[markusloecher]

It seems that the original ordering of the levels of categorical variables can change randomly, which might catch the unsuspecting user by surprise.

Here is a self.contained example using the categorical features of the titanic data:

library(titanic)

data(titanic_train)
titanic_train = titanic_train[,c("Survived", "Pclass", "Sex")]
for (j in 1:3) titanic_train[,j] = factor(titanic_train[,j])
str(titanic_train)

GenSynData = function(x = titanic_train){
  # Train the ARF
  arf <- adversarial_rf(x, verbose = 0)
  # Estimate distribution parameters
  psi <- forde(arf, x)
  #generate:
  x_syn = forge(psi, nrow(x)) 
  LevChanged = 0
  for (j in 1:ncol(x)){
    if (any(levels(x[,j]) != levels(x_syn[,j]))) {
      LevChanged = LevChanged + 1
      cat("new levels for ",colnames(x)[j] , levels(x_syn[,j]), "\n")
    }
  }
  cat(LevChanged, " categorical levels changed overall!\n")
}

set.seed(123)
for (k in 1:5) GenSynData()

'data.frame': 891 obs. of 3 variables:
$ Survived: Factor w/ 2 levels "0","1": 1 2 2 2 1 1 1 1 2 2 ...
$ Pclass : Factor w/ 3 levels "1","2","3": 3 1 3 1 3 3 1 3 3 2 ...
$ Sex : Factor w/ 2 levels "female","male": 2 1 1 1 2 2 2 2 1 1 ...
new levels for Survived 1 0
new levels for Pclass 3 1 2
new levels for Sex male female
3 categorical levels changed overall!
new levels for Pclass 3 2 1
new levels for Sex male female
2 categorical levels changed overall!
new levels for Survived 1 0
new levels for Pclass 3 2 1
2 categorical levels changed overall!
new levels for Pclass 2 3 1
1 categorical levels changed overall!
new levels for Survived 1 0
new levels for Pclass 1 3 2
new levels for Sex male female
3 categorical levels changed overall!

[mnwright]

I think this is fixed already on Github (but not on CRAN). Another reason for a new CRAN release!