tune class weights param for unbalanced classif

[ja-thomas]

We're supporting weights, so class weights might be a nice way to improve the performance on unbalanced data.

But at some point it will get annoying with more and more wrapper, so we should def. consider mlrCPO

[ja-thomas]

We can just use the scale_pos_weight parameter

[ja-thomas]

done

[p-schaefer]

Unless I am mistaken, scale_pos_weight is only for binary classification, is there an alternative for multiclass classifications? Perhaps with xgboost.DMatrix(..., weight = *weight array for individual weights*), or passing .weights to trainLearner.classif.xgboost, or perhaps there is a better alternative?