3ab6be46e1d6b21d59a3c3a0b9d0f6ef-AuthorFeedback.pdf

As correctly pointed out by R1, this would be preferable, for example for6 debugging at model development time, becauseΩ can be computed without any uncertainty, and computational7 performance isnot amajor concern atthat stage. We chose the rank-based method to show that the uncertainty estimates reflect accuracy according to the log-odds23 metric thatiswidely accepted bytheresearch community asabenchmark forfeature importance estimation (e.g. This subtle change indefinition allows ustoefficiently compute feature importance one by one.40