Skip to main content
. 2021 Feb 12;17(2):e1008618. doi: 10.1371/journal.pcbi.1008618

Fig 1. Visualization of the logS, CRPS, and IS.

Fig 1

Left: The logarithmic score only depends on the predictive probability assigned to the observed event y (of which one takes the logarithm). Middle: The CRPS can be interpreted as a measure of the distance between the predictive cumulative distribution function and a vertical line at the observed value. Right: The interval score ISα is a piecewise linear function which is constant inside the respective prediction interval and has slope ±2/α outside of it.