ntreelimit has been deprecated in favour of iteration_range

[tiagomaie]

This is a fix for issue #655

As ntreelimit as been deprecated, adding iterationrange to the
xgb_pred call with the correct modifications to fit the current use case.

Further work is needed at a later stage to extend the new functionality of iterationrange.

[juliasilge]

Do we want to make this change now and set xgboost (>= 1.5.0.1) in DESCRIPTION? Or try to handle both old/new versions somehow?

As a reminder, xgboost is only in Suggests for parsnip but how would this impact our ability to build on older versions of R? We are building back to oldrel-2 right now.

[topepo]

xgboost uses R (≥ 3.3.0) so it probably shouldn't been an issue for CI.

I"m inclined to add a version restriction in parsnip. The xgboost api does change a lot and programming around this (and other changes) would be messy.

[juliasilge]

@tiagomaie I just want to confirm with you that I changed the parameter to iteration_range based on the NEWS and docs (from what you submitted in your PR here, iterationrange).

[tiagomaie]

Thanks @juliasilge ! I should mention that although in some warnings and in the docs it says that iteration_range should be used, I think these stem from implementations in other languages, like python.
On my end (and also looking through xgboost R code base and documentation, see predict.xgb.Booster for example) it seems that the correct parameter and the one that produces the correct behavior is iterationrange.

Running parsnips test for xgboost (ie. devtools::test(filter='boost_tree_xgboost')), fail to produce the correct result if iteration_range is used instead of iterationrange in xgb_pred. From my understanding, this is because predict.xgb.Booster won't take it as a parameter and instead will just assign the ntreelimit with the value of 0.

[juliasilge]

Ah, thank you very much for clarifying @tiagomaie! Very helpful.

[github-actions]

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.