I would like to define my own behavior (method) for the multiplication of a data.frame with an object of a new S3 class. But I can't figure out how to get methods dispatch to find my method. Is there a way?
First, I define S3 objects 'a' (oldClass "A") and 'df' (oldClass "data.frame"):
a <- 4
oldClass(a) <- "A"
df <- data.frame(x=1:2,y=3:4)
I then use trace(Ops.data.frame, edit=TRUE) to add print("Ops.data.frame") at the first line. This way, I know when Ops.data.frame gets called. Here's a demonstration:
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
I could define an S3 method for class "A".
Ops.A <- function(e1, e2) {
print("Ops.A")
oldClass(e1) <- oldClass(e1)[oldClass(e1) != "A"]
oldClass(e2) <- oldClass(e2)[oldClass(e2) != "A"]
callGeneric(e1, e2)
}
This gets called for a*a but not for a*df:
# This successfully calls Ops.A
a*a
# [1] "Ops.A"
# [1] 16
# But this throws an error
a*df
# Error in a * df : non-numeric argument to binary operator
# In addition: Warning message:
# Incompatible methods ("Ops.A", "Ops.data.frame") for "*"
So that won't work.
remove(Ops.A)
What about using an S4 method instead? This requires defining the S4 class "A", but usually an S3 object with oldClass "A" will still be found by S4 dispatch.
setClass("A", list("numeric")) # Required to define a method for "A"
setGeneric("ATypicalMethod", function(e1, e2) {print("ATypicalMethod - default")})
setMethod("ATypicalMethod", signature=c("A","A"), function(e1, e2) {print("ATypicalMethod - A,A")})
ATypicalMethod(a,a)
# [1] "ATypicalMethod - A,A"
However, that doesn't work for Ops.
setMethod("Ops", signature=c("A","data.frame"), function(e1, e2) {
print("Ops(A,data.frame)")
callGeneric(e1@.Data, e2)
})
# Nope - when the scalar is an S3 object, we never find Ops(A,data.frame):
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
This behavior of a*df is explained by Martin Morgan (http://stackoverflow.com/a/12101238/3203184) and ?Methods, who say that if the S3 generic is called directly, then the S4 methods will never be found; this appears to be happening in a*df because both a and df are S3 objects.
And it doesn't help any to also call setOldClass; the problem is not that the S3 object can't be recognized by S4 methods dispatch, but that it isn't looked for when two S3 objects are passed to methods like *. In these cases the S3 generic is invoked directly, and no amount of S4 labeling will lead to S4 dispatch.
setOldClass("A", S4Class="A")
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
So now I'm at a loss. I've found no way to get * to discover an S4 method for my S3 object, and no way to write an S3 method that supersedes the data.frame method.
If I were willing to make the scalar an S4 object, I could get the dispatch I want:
a <- new("A", 4)
a*df
# [1] "Ops(A,data.frame)"
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
But I'm really hoping to keep 'a' as an S3 object. Is there a way to both (1) let 'a' be S3 and (2) define my own method for Ops('A', 'data.frame')?
I would like to define my own behavior (method) for the multiplication of a data.frame with an object of a new S3 class. But I can't figure out how to get methods dispatch to find my method. Is there a way?
First, I define S3 objects 'a' (oldClass "A") and 'df' (oldClass "data.frame"):
a <- 4
oldClass(a) <- "A"
df <- data.frame(x=1:2,y=3:4)
I then use trace(Ops.data.frame, edit=TRUE) to add print("Ops.data.frame") at the first line. This way, I know when Ops.data.frame gets called. Here's a demonstration:
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
I could define an S3 method for class "A".
Ops.A <- function(e1, e2) {
print("Ops.A")
oldClass(e1) <- oldClass(e1)[oldClass(e1) != "A"]
oldClass(e2) <- oldClass(e2)[oldClass(e2) != "A"]
callGeneric(e1, e2)
}
This gets called for a*a but not for a*df:
# This successfully calls Ops.A
a*a
# [1] "Ops.A"
# [1] 16
# But this throws an error
a*df
# Error in a * df : non-numeric argument to binary operator
# In addition: Warning message:
# Incompatible methods ("Ops.A", "Ops.data.frame") for "*"
So that won't work.
remove(Ops.A)
What about using an S4 method instead? This requires defining the S4 class "A", but usually an S3 object with oldClass "A" will still be found by S4 dispatch.
setClass("A", list("numeric")) # Required to define a method for "A"
setGeneric("ATypicalMethod", function(e1, e2) {print("ATypicalMethod - default")})
setMethod("ATypicalMethod", signature=c("A","A"), function(e1, e2) {print("ATypicalMethod - A,A")})
ATypicalMethod(a,a)
# [1] "ATypicalMethod - A,A"
However, that doesn't work for Ops.
setMethod("Ops", signature=c("A","data.frame"), function(e1, e2) {
print("Ops(A,data.frame)")
callGeneric(e1@.Data, e2)
})
# Nope - when the scalar is an S3 object, we never find Ops(A,data.frame):
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
This behavior of a*df is explained by Martin Morgan (http://stackoverflow.com/a/12101238/3203184) and ?Methods, who say that if the S3 generic is called directly, then the S4 methods will never be found; this appears to be happening in a*df because both a and df are S3 objects.
And it doesn't help any to also call setOldClass; the problem is not that the S3 object can't be recognized by S4 methods dispatch, but that it isn't looked for when two S3 objects are passed to methods like *. In these cases the S3 generic is invoked directly, and no amount of S4 labeling will lead to S4 dispatch.
setOldClass("A", S4Class="A")
a*df
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
So now I'm at a loss. I've found no way to get * to discover an S4 method for my S3 object, and no way to write an S3 method that supersedes the data.frame method.
If I were willing to make the scalar an S4 object, I could get the dispatch I want:
a <- new("A", 4)
a*df
# [1] "Ops(A,data.frame)"
# [1] "Ops.data.frame"
# x y
# 1 4 12
# 2 8 16
But I'm really hoping to keep 'a' as an S3 object. Is there a way to both (1) let 'a' be S3 and (2) define my own method for Ops('A', 'data.frame')?
0 commentaires:
Enregistrer un commentaire