Post-processing function roc_pivot does affect only $y_hat of explainer and not the predict function. When dealing with new data it may be misleading. Therefore roc_pivot should also change predict function
This is a good issue and it needs to be discussed in more depth as the changes for this to take effect are affecting DALEX. As for now, it will remain open.
Post-processing function
roc_pivot
does affect only$y_hat
of explainer and not the predict function. When dealing with new data it may be misleading. Thereforeroc_pivot
should also change predict functionThanks @piotrpiatyszek for bringing it up