Update xgboost learner to latest CRAN version

[pat-s]

• add the already existing (new) parameters of the latest CRAN version of xgboost
• add parameter dependencies

In addition, some small fixes which including #30

fixes #45

[mb706]

tree_method default should be "auto" I believe: https://xgboost.readthedocs.io/en/latest/parameter.html

Also the inequality condition could be done with CondAnyOf.

[codecov-io]

Codecov Report

Merging #31 into master will increase coverage by 0.2%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master      #31     +/-   ##
=========================================
+ Coverage   97.53%   97.74%   +0.2%     
=========================================
  Files          18       18             
  Lines         691      753     +62     
=========================================
+ Hits          674      736     +62     
  Misses         17       17

Impacted Files	Coverage Δ
R/LearnerRegrXgboost.R	98.94% <100%> (+0.5%)	⬆️
R/LearnerClassifXgboost.R	98.36% <100%> (+0.55%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e5fd62a...e4f44e2. Read the comment docs.

[be-marc]

All missing parameters are added. Some parameters have a complex dependency on the updater parameter. This might be impossible to check with paradox at the moment. However, xgboost prints the possible combinations if something is wrong.

[be-marc]

Closes #45

[pat-s]

👍

This might be impossible to check with paradox at the moment.

Did you open an issue or similar? Couldn't find anything in the repo.

In addition, once accepted, could you please reflect those change also to mlr2?

[pat-s]

Thanks! 👍

[be-marc]

If the booster parameter is set to gbtree or dart, updater must be a comma separated string which can contain the following words "grow_colmaker", "distcol", "grow_histmaker", "grow_local_histmaker", "grow_skmaker", sync", "refresh", "prune". If the booster parameter is set to gblinear, updater must be a string containing shotgun or coord_descent.

We can check this with the following code

if(x$booster == "gblinear") {
  if(!x$updater %in%  c("shotgun", "coord_descent")) {
     # Error or replace by default value?
   }
  } else if (x$booster == "gbtree" || x$booster == "dart") {
    split_paramter = strsplit(x$updater, ",")
    split_paramter = lapply(split_paramter, function(x) gsub(" ", "", x))
    if(!all(split_paramter %in% c("grow_colmaker", "distcol", "grow_histmaker",
                                                   "grow_local_histmaker", "grow_skmaker",
                                                    "sync", "refresh", "prune"))) {
     # Error or replace by default value?
   }
  }

So updater is not a really tune-able parameter. This is more about preventing the user from inserting a wrong combination. Am I right that this is not possible with add_dep since it depends on the values in both parameters? We can include it as a trafo function. How do I apply the trafo before invoking the training?

[pat-s]

AFAIR we agreed on taking the updater arg easy as it won't be used that often and people using it should be aware of what's required.
Is that correct?

@be-marc Maybe we can finish up this PR in the next days.

[pat-s]

@be-marc some tests are failing. Could you take a look?

[be-marc]

@pat-s The failing tests are resolved in #46

[pat-s]

Thanks again @be-marc!