S4 Classes: Multiple types per slot

Question

Is it possible to create an S4 class, where one or more of the slots can be of multiple classes? For example. Let's say that you had a situation where data could be either a vector, or a data.frame.

exampleClass <- setClass("exampleClass",
    representation(raw=c("data.frame","numeric","character"),
    anotherSlot=c("data.frame","numeric")) 

Or, is this the type of situation where defining a sub-class / super-class becomes necessary?

PS: Searching for a useful tutorial on S4 classes produces limited results. Links to a good tutorial on S4 class creation/usage/documentation would be greatly appreciated.

Answer 1

R has 'class unions', so

setOldClass("data.frame")
setClassUnion("data.frameORvector", c("data.frame", "vector"))

The class data.frameORvector is virtual, so can't be instantiated but can be used in other slots (representation=), as a contained class (contains=), and for dispatch

A = setClass("A", 
        representation=representation(x="data.frameORvector"))


> A(x=1:3)
An object of class "A"
Slot "x":
[1] 1 2 3

> A(x=data.frame(x=1:3, y=3:1))
An object of class "A"
Slot "x":
  x y
1 1 3
2 2 2
3 3 1

Methods can be tricky to implement because all you know is that the slot contains one of the parent types of the class union.

setGeneric("hasa", function(object) standardGeneric("hasa"))
setMethod("hasa", "data.frameORvector", function(object) typeof(object@x))

> hasa(A(x=1:5))
[1] "integer"
> hasa(A(x=data.frame(y=1:5)))
[1] "list"

I actually find the documentation on ?Classes, ?Methods, ?setClass, and friends helpful. Hadley Wickham has a tutorial (the example on this page isn't that strong, it instantiates Person, whereas conceptually one would write a People to exploit R's vectorization strengths) and there is a section in this recent Bioconductor course. I don't think either goes in to detail about class unions.