[Bioc-devel] Overloading subset operator for an S4 object with more than two dimensions

On May 14, 2015, at 12:35 GMT+2, Christian Arnold <christian.arnold at embl.de> wrote:

Hi there,

I am about to develop a Bioconductor package that implements a custom S4 object, and I am currently thinking about a few issues, including the following:

Say we have an S4 object that stores a lot of information in different slots. Assume that it does make sense to extract information out of this object in four different "dimensions" (conceptually similar to a four-dimensional object), so one would like to use the subset "[" operator for this, but extending beyond the "typical" one or two dimensions to 4:

setClass("A", representation=representation(a="numeric",b="numeric",c="numeric",d="numeric"))
a = new("A", a=1:5,b=1:5,c=1:5,d=1:5)

Now it would be nice to do stuff like a[1,2,3:4,5], which should simply return the selected elements in slots a, b, c, and d, respectively. So a[1,2,3:4,5] would return:

An object of class "A"
Slot "a":
[1] 1

Slot "b":
[1] 2

Slot "c":
[1] 3 4

Slot "d":
[1] 5

This is how far I've come:

setMethod("[", c("A", "ANY", "ANY","ANY"),
         function(x, i, j, ..., drop=TRUE)
         {
           dots <- list(...)
           if (length(dots) > 2) {
             stop("Too many arguments, must be four dimensional")
           }

           # Parse the extra two dimensions that we need from the ... argument
           k = ifelse(length(dots) > 0 , dots[[1]], c(1:5))
           l = ifelse(length(dots) == 2, dots[[2]], c(1:5))

           initialize(x, a=x at a[i],b=x at b[j],c=x at c[k],d=x at d[l])
         })

This works for stuff like a[1,2,3, 4], but fails with a general error if one of the indices is a vector such as a[1:2,2,3, 4] or a[1,2,3,4:5].


So, in summary, my questions are:
1. Is there a reasonable way of achieving the 4-dimensional subsetting that works as a user would expect it to work?
2. Does it make more sense to write a custom function instead to achieve this, such as subsetObject() without overloading "[" explicitly? What are the Bioconductor recommendations here?

I'd appreciate any help, suggestions, etc!

Thanks,
Christian

Hi there,

I am about to develop a Bioconductor package that implements a custom S4
object, and I am currently thinking about a few issues, including the
following:

Say we have an S4 object that stores a lot of information in different
slots. Assume that it does make sense to extract information out of this
object in four different "dimensions" (conceptually similar to a
four-dimensional object), so one would like to use the subset "[" operator
for this, but extending beyond the "typical" one or two dimensions to 4:

setClass("A",
representation=representation(a="numeric",b="numeric",c="numeric",d="numeric"))
a = new("A", a=1:5,b=1:5,c=1:5,d=1:5)

Now it would be nice to do stuff like a[1,2,3:4,5], which should simply
return the selected elements in slots a, b, c, and d, respectively. So
a[1,2,3:4,5] would return:

An object of class "A"
Slot "a":
[1] 1

Slot "b":
[1] 2

Slot "c":
[1] 3 4

Slot "d":
[1] 5

This is how far I've come:

setMethod("[", c("A", "ANY", "ANY","ANY"),
          function(x, i, j, ..., drop=TRUE)
          {
            dots <- list(...)
            if (length(dots) > 2) {
              stop("Too many arguments, must be four dimensional")
            }

            # Parse the extra two dimensions that we need from the ...
argument
            k = ifelse(length(dots) > 0 , dots[[1]], c(1:5))
            l = ifelse(length(dots) == 2, dots[[2]], c(1:5))

            initialize(x, a=x at a[i],b=x at b[j],c=x at c[k],d=x at d[l])
          })

This works for stuff like a[1,2,3, 4], but fails with a general error if
one of the indices is a vector such as a[1:2,2,3, 4] or a[1,2,3,4:5].


So, in summary, my questions are:
1. Is there a reasonable way of achieving the 4-dimensional subsetting
that works as a user would expect it to work?
2. Does it make more sense to write a custom function instead to achieve
this, such as subsetObject() without overloading "[" explicitly? What are
the Bioconductor recommendations here?

I'd appreciate any help, suggestions, etc!

Thanks,
Christian

I agree with Wolfgang that the semantics of [ are being violated here. 
It would though help if you could be a little less vague about your 
intent. What is this data structure going to store, how should it behave?

On Thu, May 14, 2015 at 3:35 AM, Christian Arnold 
<christian.arnold at embl.de <mailto:christian.arnold at embl.de>> wrote:

    Hi there,

    I am about to develop a Bioconductor package that implements a
    custom S4 object, and I am currently thinking about a few issues,
    including the following:

    Say we have an S4 object that stores a lot of information in
    different slots. Assume that it does make sense to extract
    information out of this object in four different "dimensions"
    (conceptually similar to a four-dimensional object), so one would
    like to use the subset "[" operator for this, but extending beyond
    the "typical" one or two dimensions to 4:

    setClass("A",
    representation=representation(a="numeric",b="numeric",c="numeric",d="numeric"))
    a = new("A", a=1:5,b=1:5,c=1:5,d=1:5)

    Now it would be nice to do stuff like a[1,2,3:4,5], which should
    simply return the selected elements in slots a, b, c, and d,
    respectively. So a[1,2,3:4,5] would return:

    An object of class "A"
    Slot "a":
    [1] 1

    Slot "b":
    [1] 2

    Slot "c":
    [1] 3 4

    Slot "d":
    [1] 5

    This is how far I've come:

    setMethod("[", c("A", "ANY", "ANY","ANY"),
              function(x, i, j, ..., drop=TRUE)
              {
                dots <- list(...)
                if (length(dots) > 2) {
                  stop("Too many arguments, must be four dimensional")
                }

                # Parse the extra two dimensions that we need from the
    ... argument
                k = ifelse(length(dots) > 0 , dots[[1]], c(1:5))
                l = ifelse(length(dots) == 2, dots[[2]], c(1:5))

                initialize(x, a=x at a[i],b=x at b[j],c=x at c[k],d=x at d[l])
              })

    This works for stuff like a[1,2,3, 4], but fails with a general
    error if one of the indices is a vector such as a[1:2,2,3, 4] or
    a[1,2,3,4:5].


    So, in summary, my questions are:
    1. Is there a reasonable way of achieving the 4-dimensional
    subsetting that works as a user would expect it to work?
    2. Does it make more sense to write a custom function instead to
    achieve this, such as subsetObject() without overloading "["
    explicitly? What are the Bioconductor recommendations here?

    I'd appreciate any help, suggestions, etc!

    Thanks,
    Christian

Thanks for your input, highly appreciated!

I can see that the semantics of "[" are violated, so I agree that
overwriting the "subset" method  is probably a better way to go.
Essentially, the object stores several, individual-specific count
matrices from RNA-Seq experiments in an potentially allele(read
group)-specific manner. So the dimensions to subset on are the read

groups, the rows and columns of the matrices, and the individuals itself.

So I guess overloading the subset method with four arguments, each
corresponding to one of the dimensions a subset is suitable for this
kind of object, is the way to go.

Thanks,
Christian


On 14.05.2015 15:57, Michael Lawrence wrote:

I agree with Wolfgang that the semantics of [ are being violated here.
It would though help if you could be a little less vague about your
intent. What is this data structure going to store, how should it behave?

On Thu, May 14, 2015 at 3:35 AM, Christian Arnold
<christian.arnold at embl.de <mailto:christian.arnold at embl.de>> wrote:

     Hi there,

     I am about to develop a Bioconductor package that implements a
     custom S4 object, and I am currently thinking about a few issues,
     including the following:

     Say we have an S4 object that stores a lot of information in
     different slots. Assume that it does make sense to extract
     information out of this object in four different "dimensions"
     (conceptually similar to a four-dimensional object), so one would
     like to use the subset "[" operator for this, but extending beyond
     the "typical" one or two dimensions to 4:

     setClass("A",
     representation=representation(a="numeric",b="numeric",c="numeric",d="numeric"))
     a = new("A", a=1:5,b=1:5,c=1:5,d=1:5)

     Now it would be nice to do stuff like a[1,2,3:4,5], which should
     simply return the selected elements in slots a, b, c, and d,
     respectively. So a[1,2,3:4,5] would return:

     An object of class "A"
     Slot "a":
     [1] 1

     Slot "b":
     [1] 2

     Slot "c":
     [1] 3 4

     Slot "d":
     [1] 5

     This is how far I've come:

     setMethod("[", c("A", "ANY", "ANY","ANY"),
               function(x, i, j, ..., drop=TRUE)
               {
                 dots <- list(...)
                 if (length(dots) > 2) {
                   stop("Too many arguments, must be four dimensional")
                 }

                 # Parse the extra two dimensions that we need from the
     ... argument
                 k = ifelse(length(dots) > 0 , dots[[1]], c(1:5))
                 l = ifelse(length(dots) == 2, dots[[2]], c(1:5))

                 initialize(x, a=x at a[i],b=x at b[j],c=x at c[k],d=x at d[l])
               })

     This works for stuff like a[1,2,3, 4], but fails with a general
     error if one of the indices is a vector such as a[1:2,2,3, 4] or
     a[1,2,3,4:5].


     So, in summary, my questions are:
     1. Is there a reasonable way of achieving the 4-dimensional
     subsetting that works as a user would expect it to work?
     2. Does it make more sense to write a custom function instead to
     achieve this, such as subsetObject() without overloading "["
     explicitly? What are the Bioconductor recommendations here?

     I'd appreciate any help, suggestions, etc!

     Thanks,
     Christian