[LRNRQ] Add <xgb.train()> for counts and ranks from package <xgboost>

[nipnipj]

Algorithm

xgboost for counts and ranks

Package

xgboost

Supported types

• [x] classif
• [x] regr
• [ ] surv
• [ ] dens

I have checked that this is not already implemented in

• [x] mlr3
• [x] mlr3learners
• [x] mlr3extralearners
• [x] Other core packages (e.g. mlr3proba)

Why do I think this is a useful learner?

Why not?

Further Optional Comments

I think it would be easily implemented since it will uses xgb.train() as in regr.xgboost or classif.xgboost. I think this might be relate to mlr3 ordinal.

[RaphaelS1]

Sorry I don't understand the request, what exactly do you want implemented?

[nipnipj]

Basically, I would like to use xgboost algorithm for an ordinal response. I tried using classif.xgboost and setting the parameter objetive to objective="rank:pairwise", but it doesn't work. I do not know whether this will be possible when mlr3ordinal is complete, tho.

[RaphaelS1]

Sorry I missed your reply! Next time tag me so I see it straight away! I'm still not sure what you mean by 'doesn't work', can you please provide a reprex

[nipnipj]

@RaphaelS1 Currently it is not possible to use an ordinal variable as response with xgboost learner in mlr3. I would like know whether it is possible add this to mlr3.

[RaphaelS1]

In which case no this won't be supported until mlr3ordinal is on CRAN. But we can keep this issue open until then

[sebffischer]

Closing because of no progress regarding mlr3ordinal