And as a follow-up, I implemented a barebones as.data.frame.MyClass(...).
It works when dealing with non-subsetted data frames, but fails upon a
subset(...) call:
as.data.frame.MyClass <- function(x, ...) as.data.frame.vector(x, ...)
This works for a single column, e.g.:
str(data.frame(MyClass(1:10)))
'data.frame': 10 obs. of 1 variable:
$ MyClass.1.10.:Class 'MyClass' int [1:10] 1 2 3 4 5 6 7 8 9 10
But not during a subset:
str(subset(data.frame(x = MyClass(1:10)), x %% 2 == 0))
'data.frame': 5 obs. of 1 variable:
$ x: int 2 4 6 8 10
-Murat
On Wed, Nov 12, 2014 at 10:02 PM, Murat Tasan <mmuurr at gmail.com> wrote:
Hi all --- I've stumbled upon some pretty annoying behavior, and I'm
curious how others may have gotten around it.
When using subset(...) on a data frame that contains a custom S3
field, the class is dropped in the result:
MyClass <- function(x) structure(x, class = "MyClass")
df <- data.frame(x = 1:10, y = 10:1)
df$x <- MyClass(df$x)
str(df)
'data.frame': 10 obs. of 2 variables:
$ x:Class 'MyClass' int [1:10] 1 2 3 4 5 6 7 8 9 10
$ y: int 10 9 8 7 6 5 4 3 2 1
str(subset(df, x %% 2 == 0))
'data.frame': 5 obs. of 2 variables:
$ x: int 2 4 6 8 10
$ y: int 9 7 5 3 1
And so, any generic functions hooked to MyClass suddenly don't work on
the subset results, but do work on the original data frame.
I think I could write a custom as.data.frame.MyClass for all such
classes, but this is annoying, indeed (and I don't know for sure if
that's a robust solution)
Wrapping in I(...) doesn't work, either:
df <- data.frame(x = 1:10, y = 10:1)
df$x <- I(MyClass(df$x))
str(df)
'data.frame': 10 obs. of 2 variables:
$ x:Classes 'AsIs', 'MyClass' int [1:10] 1 2 3 4 5 6 7 8 9 10
$ y: int 10 9 8 7 6 5 4 3 2 1
str(subset(df, x %% 2 == 0))
'data.frame': 5 obs. of 2 variables:
$ x:Class 'AsIs' int [1:5] 2 4 6 8 10
$ y: int 9 7 5 3 1
(note that while 'AsIs' is kept, 'MyClass' has been removed in $x)
Cheers!
-Murat