dataset_subset hides dataset subclass

[jonthegeek]

I expected dataset_subset to keep dataset properties, but it hides them. Could we either sort out a way to elevate the dataset to the "top" of that object, or at least give the new dataset a subclass so methods can know to look down a level for "real" dataset properties?

library(torch)
minimal_dataset <- dataset(
  "minimal",
  initialize = function() {
    self$data <- torch_tensor(1:5)
  },
  .length = function() {
    self$data$size()[[1]]
  }
)
testing <- minimal_dataset()
class(testing)
#> [1] "minimal" "dataset" "R6"
testing_sub <- dataset_subset(testing, 1:2)
class(testing_sub)
#> [1] "dataset" "R6"

^{Created on 2022-09-15 with reprex v2.0.2}

[dfalbel]

Indeed! This is a bug, we should keep the same class as the provided dataset!

[jonthegeek]

It isn't just class, anything else the provided dataset has (in my case we have some methods) should come through. After implementing a workaround, I think the best option might be to give it a "dataset_subset" class, so we can make S3 methods for that specifically and then call that generic on the $dataset inside.