unbundle issue with XGBoost models

[lanedrew]

When using the unbundle function on a bundled XGBoost model, even after the 0.1.3 update, there is an error associated with the argument as_booster which has become deprecated.

Below is an example of the output received when attempting to unbundle a model after loading the bundled model object:

Load the example mark model that previously was trained on the small example data

file_path <- system.file("extdata", "example_mark_model.rds", package = "ldmppr")
mark_model <- bundle::unbundle(readRDS(file_path))
Error in xgboost::xgb.load.raw(object, as_booster = TRUE) :
unused argument (as_booster = TRUE)
Calls: ... -> unbundle -> unbundle.bundle ->
Execution halted

From what I can tell, just removing the as_booster argument should address the issue in the unbundle function for XGBoost objects.

[juliasilge]

I don't think you should be able to unbundle an arbitrary model in this way from data stored in a package. (Unless it happened to be stored as a bundled object to start with?)

Can you share a more straightforward example of the kind of problem you are seeing? Something along the lines of first bundling:

library(xgboost)
library(bundle)

mod <- xgboost(mtcars[, -1], mtcars$mpg, nrounds = 3)
bundled <- bundle(mod)
saveRDS(bundled, "~/Desktop/xgboost_bundle.rds")

^{Created on 2025-12-23 with reprex v2.1.1}

And then unbundling:

library(xgboost)
library(bundle)
out <- readRDS("~/Desktop/xgboost_bundle.rds")
unbundle(out)
#> ##### xgb.Booster
#> # of features: 10 
#> # of rounds:  3

^{Created on 2025-12-23 with reprex v2.1.1}

That example was using bundle 0.1.3 and xgboost 3.1.2.1.