Elaborations on mean_wQuantileLoss
calculations: why is it weighted by abs_target_sum
and why is it not calculated on a per-series basis in get_metrics_per_ts()
?
#3095
Unanswered
Serendipity31
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Because it is an approximation of CRPS, I'm interested in understanding more fully the choices that have been made regarding the calculation of
mean_wQuantileLoss
.Why is each
QuantileLoss[q]
weighted byabs_target_sum
? I have only found reference to the kind of weighting discussed in Gneiting and Ranjan (2011), but this doesn't make use of the target values in the way the calculation ofwQuantileLoss[q]
does.Is there a conceptual or theoretical reason why
mean_wQuantileLoss
not calculated as a part of get_metrics_per_ts()? The information that is required to calculatewQuantileLoss[q]
on a per-series basis appears to exist, yet this is not done. I'd like to know if there is a reason to refrain from calculating this metric.Is CRPS supposed to be bounded [0, 1]? If so, how does the approximation of CRPS with
mean_wQuantileLoss
maintain those bounds?Beta Was this translation helpful? Give feedback.
All reactions