Difference between factor and character variables running randomForest

Question

If I run a randomForest(y ~ x, data = df) model, x a factor variable with more than 53 levels I get

Error in randomForest.default(m, y, ...) : 
  Can not handle categorical predictors with more than 53 categories.

If I change x to as.character(x) and re-run I get no errors.

What's the difference behind the scenes? Isn't both types treated as categorical variables?

Answer 1

I guess each category's name is a numeric value (because randomForest() can't treat character class when it consist of character). randomForest() treat character class which consist of numeric value as numeric variables (i.e., numeric class), NOT categorical variables (i.e., factor class). If you change each category's name, the result will change.

Here is my example. If x_ is factor class, the same results return. If x_ is integer class or character class (but composed of numeric value), the outputs depend on the value. The result you got by as.character(x) is CLEARY WRONG !!

set.seed(1); cw <- data.frame(y = subset(ChickWeight, Time==18)$weight, x1 = sample(47) )
cw$x2 <- as.factor(cw$x1)
cw$x3 <- as.character(cw$x1)
cw$x4 <- 47:1
cw$x5 <- as.factor(47:1)
cw$x6 <- as.character(47:1)
cw$x7 <- c(letters, LETTERS[1:21])
cw$x8 <- as.factor(cw$x7)
                               # %Var explained # class(x_)
set.seed(1); randomForest(y ~ x1, cw) # -29.61  integer1
set.seed(1); randomForest(y ~ x2, cw) # -0.42   factor
set.seed(1); randomForest(y ~ x3, cw) # -29.61  character (numeric name1)
set.seed(1); randomForest(y ~ x4, cw) # -31.78  integer2
set.seed(1); randomForest(y ~ x5, cw) # -0.42   factor
set.seed(1); randomForest(y ~ x6, cw) # -31.78  character (numeric name2)
set.seed(1); randomForest(y ~ x7, cw) # error   character (letter name)
set.seed(1); randomForest(y ~ x8, cw) # -0.42   factor