The largest value t such that at least sens of the predictions at presences are >= t:
the confusion-matrix threshold that fixes sensitivity. sens = 0.9 is the "10th percentile
training presence" (P10) when pred are training predictions, and the honest version of it
when they are held-out predictions. No interpolation: the result is always one of the values
in pred, which is what OBIS's pipeline does (rev(sort(x))[ceiling(n * 0.9)]), so a
published P10 is reproduced exactly.
Examples
sens_threshold(c(10, 20, 30, 40, 50, 60, 70, 80, 90, 100)) # 20: nine of ten are >= 20
#> [1] 20