Performance profiles #49
stephane-caron
started this conversation in
Design
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
In this benchmark, we use the shifted geometric mean rather than performance profiles. The main reason for this choice is that performance profiles only make sense for a pair of solvers, and can be misleading when more than two performance profiles are plotted at the same time (because, were we to remove the best-performing solver from the plot, the ranking between the remaining solvers may look completely different). See this paper and that one for more details.
See also
Beta Was this translation helpful? Give feedback.
All reactions