xgboost and RF models should support class balancing weights in loss function

[mcloughlin2]

The xgboost, RF and NN models all have different ways to handle imbalanced classification datasets by using class-specific weights in their loss functions; but we currently only support this for NN models, by setting the weight_transform_type parameter to 'balancing'. We should add this capability for random forests and xgboost models as well. For RF models this means setting the class_weights parameter to 'balanced' when we create the RandomForestClassifier. For xgboost models you do it by setting the scale_pos_weight parameter to sum(negative instances) / sum(positive instances).

[mcloughlin2]

Implemented; changes pushed to branch 'sparsity'.