Added the balance argument to
pipeline(). When set to TRUE, the pipeline
automatically calculates inverse class frequency weights to handle
imbalanced datasets natively. These observation weights are passed
directly into the loss functions of the underlying models (glmnet,
ranger, and xgboost) without duplicating text data or artificially
bloating the sparse matrix. ## Updates
evaluate_performance() minor updates. Bug
fixed