[Bioc-devel] Delayed Assignment to S4 Slots

Good day,

I have an S4 class with some slots in my Bioconductor package. One of the slots stores the range of top variables to try during feature selection (the variables might be ranked by some score, like a t-test). The empty constructor looks like

setMethod("ResubstituteParams", "missing", function()
{
   new("ResubstituteParams", nFeatures = seq(10, 100, 10), performanceType = "balanced error")
})

But, someone might have a small omics data set with only 40 features (e.g. CyTOF). Therefore, trying the top 10, 20, ..., 100 is not a good default. A good default would wait until the S4 class is accessed within cross-validation and then, based on the dimensions of the matrix or DataFrame, pick a suitable range. I looked at delayedAssign, but x is described as "a variable name (given as a quoted string in the function call)". It doesn't seem to apply to S4 slots based on my understanding of it.

r <- ResubstituteParams()
delayedAssign("r at nFeatures", nrow(measurements))
measurements <- matrix(1:100, ncol = 10)
r at nFeatures # Still the value from empty constructor.

  [1]  10  20  30  40  50  60  70  80  90 100

--------------------------------------
Dario Strbenac
University of Sydney
Camperdown NSW 2050
Australia

Hi Dario,

It would be good if you could provide a little bit more context about 
ResubstituteParams objects and how they are typically used in practice.

As you found out, delayedAssign() won't work. How about trying something 
like this?

 ? setClassUnion("numeric_OR_NULL", c("numeric", "NULL"))



 ? setClass("ResubstituteParams",

 ??? slots=c(

 ??????? nFeatures_call="call",

 ??????? nFeatures="numeric_OR_NULL"

 ??? )

 ? )



 ? ResubstituteParams <- function(nFeatures_call)

 ? {

 ??? new("ResubstituteParams", nFeatures_call=substitute(nFeatures_call))

 ? }



 ? ## Should preferrably be an S4 generic with a method for
 ? ## ResubstituteParams
 ?objects.

 ? nFeatures <- function(x)

 ? {

 ??? if (!is.null(x at nFeatures))

 ??????? return(x at nFeatures)

 ??? eval(x at nFeatures_call)

 ? }



 ? r <- ResubstituteParams(nrow(measurements))

 ? measurements <- matrix(1:100, ncol=10)

 ? nFeatures(r)

 ? # [1] 10

Accessor functions are not just a nice way to hide the internals of your 
object but they are also a powerful way to take full control of what 
needs to happen exactly when a user tries to access a particular part of 
your object, whether that part lives in a slot, a combination of slots, 
or in no slot at all.


Cheers,
H.


On 13/10/2021 21:00, Dario Strbenac via Bioc-devel wrote:

Good day,

I have an S4 class with some slots in my Bioconductor package. One of 
the slots stores the range of top variables to try during feature 
selection (the variables might be ranked by some score, like a 
t-test). The empty constructor looks like

setMethod("ResubstituteParams", "missing", function()
{
?? new("ResubstituteParams", nFeatures = seq(10, 100, 10), 
performanceType = "balanced error")
})

But, someone might have a small omics data set with only 40 features 
(e.g. CyTOF). Therefore, trying the top 10, 20, ..., 100 is not a good 
default. A good default would wait until the S4 class is accessed 
within cross-validation and then, based on the dimensions of the 
matrix or DataFrame, pick a suitable range. I looked at delayedAssign, 
but x is described as "a variable name (given as a quoted string in 
the function call)". It doesn't seem to apply to S4 slots based on my 
understanding of it.

r <- ResubstituteParams()
delayedAssign("r at nFeatures", nrow(measurements))
measurements <- matrix(1:100, ncol = 10)
r at nFeatures # Still the value from empty constructor.

? [1]? 10? 20? 30? 40? 50? 60? 70? 80? 90 100

--------------------------------------
Dario Strbenac
University of Sydney
Camperdown NSW 2050
Australia