[R] Add parameters constructor

[david-cortes]

ref #9810

This PR adds a utility function to return a list of XGBoost parameters, with docs about all of them copied from the .rst files with slight modifications considering that they are meant exclusively for the R package. I'm not entirely sure whether all of them are applicable though.

The idea behind this constructor function is to offer an IDE-friendly way of creating and discovering XGBoost parameters, with full in-package documentation as is typical in R packages. It should be possible to reuse these same docs later on in xgboost() as needed through @inheritParams, but for now I'm only using this for xgb.train and xgb.cv.

Since this introduces a new function with a somewhat similar name as a completely different method in the package, it also renames that method in the process in order to avoid confusions and IDE miscompletions.

[trivialfis]

Since this PR concerns the user interface, would be great if we could get some help for review from an R user's perspective @mayer79 @jameslamb

[mayer79]

Very good idea!

[trivialfis]

This is a Python example, do we want to include it here?

[david-cortes]

I'd say python examples are still relevant.

[jameslamb]

Since this PR concerns the user interface, would be great if we could get some help for review from an R user's perspective @mayer79 @jameslamb

Thanks for the @, sorry I did not get to it before this was merged.

The idea behind this constructor function is to offer an IDE-friendly way of creating and discovering XGBoost parameters, with full in-package documentation as is typical in R packages

I think it's a great idea, and would definitely be helpful for interactive development.

Some suggestions (to consider in follow-ups, given that this was already merged):

• it would be good to have explicit unit tests on this new function, not just rely on implicit coverage from other tests. For example:
  • confirming that only non-NULL items show up in the return value
  • confirming that calling this with 0 arguments returns an empty list
• I understand that existing names with . in them need to be preserved for backwards compatibility, but I'd stop adding new ones (e.g. I'd prefer xgb_params to xgb.params)
  • to reduce the risk of "R thinks this is an S3 method" confusion like the ones that led to [R] Fix method name. #9577
• I'd have asked to change the line wrapping of the roxygen comments instead of ignoring the project's preferred line length with # nolint comments
  • or at least, asked that the linter be allowed to cover the function signature, which it currently cannot because of this placement:

xgboost/R-package/R/xgb.train.R

Line 811 in 85ccb8f

# nolint end

I also would have asked for a little research supporting it being safe to change xgb.parameters to xgb.model.parameters without a deprecation cycle. For example, it looks to me like nothing on CRAN is currently using xgb.parameters (https://github.com/search?q=org%3Acran%20%22xgb.parameters%22&type=code), so I guess it at least won't break anything on CRAN. (I don't recall where the discussion about xgboost2 vs. releasing this as a new version of xgboost ended, disregard this last comment if the plan is to release a new package).

[trivialfis]

@jameslamb Thank you for the suggestion, I have converted your comment into a new tracking issue #11083