Timeline for Example of a (strictly) proper scoring rule on a general measurable space?
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 19, 2020 at 15:38 | comment | added | usul | @aduh, Hmm, I don't know. It must have to do with $F$ being required to be weakly increasing, implying somehow that it's not possible to modify $F$ on a set of measure zero. | |
| Oct 19, 2020 at 0:09 | comment | added | aduh | Is there an easy way to see that CRPS is indeed strictly proper on probabilities over $[0,1]$? I'm having trouble showing it. | |
| Oct 18, 2020 at 14:00 | history | edited | usul | CC BY-SA 4.0 |
correction
|
| Oct 18, 2020 at 13:59 | comment | added | usul | @aduh ah, thanks for catching that! I assume it's because the expected score becomes $-\infty$ for many (any?) forecasts, making it weakly proper as they say.... But at least on $[0,1]$ it will be unrestricted. | |
| Oct 18, 2020 at 6:37 | comment | added | aduh | It looks like they say it's strictly proper only for the subclass of $\mathcal P$ consisting of those probability measures with finite first moment. | |
| Oct 18, 2020 at 3:36 | history | answered | usul | CC BY-SA 4.0 |