Pryr stops on DOTSXP type

[dhuebner]

I noticed that DOTSXP is not handled in pryr::object_size

Here is an example:

library(caret)
train.flag <- createDataPartition(y=iris$Species,p=0.5,list=FALSE)
training <- iris[train.flag,]

modfit <- train(Species~.,method="rpart",data=training)
pryr::object_size(modfit$finalModel$call$formula)

type: 17
Error: Unimplemented type

See https://github.com/hadley/pryr/blob/master/src/object_size.cpp#L149

[ck37]

I am running into this also :/

Could there be an option to issue a warning and return 0 rather than stopping? Better to have a lower bound on the object size rather than no estimate.

Or could it fall back to object.size() for those elements, again with a warning?

Here is an example:

data(Boston, package = "MASS")
library(randomForest)
library(pryr)

set.seed(1)

outcome = Boston$medv
data = Boston[, -14]

# This works ok.
rf = randomForest(outcome ~ ., data = data, keep.forest = T)
object_size(rf)

# But not the SuperLearner version.
library(SuperLearner)
sl_rf = SuperLearner(outcome, data, family = "gaussian",
                  SL.library = c("SL.mean", "SL.randomForest"))
# Doesn't work due to the terms element.
object_size(sl_rf)
# object.size() works though.
object.size(sl_rf)

# Dig into the sl_rf details.
obj = sl_rf$fitLibrary$SL.randomForest_All$object

# Fails on $terms element of type "LANGSXP"
for (name in names(obj)) {
  elm = obj[[name]]
  cat("Element:", name, "Type:", pryr::sexp_type(elm), "Size:")
  try(cat(pryr::object_size(elm)))
  cat("\n")
}

# Dig into the terms attributes
terms = obj$terms

# All of these are ok.
for (attr_name in names(attributes(terms))) {
  elm = attr(obj, attr_name, exact = T)
  cat("Attribute:", attr_name, "Type:", pryr::sexp_type(elm), "Size:")
  try(cat(pryr::object_size(elm)))
  cat("\n")
}