Skip to main content
. Author manuscript; available in PMC: 2022 May 20.
Published in final edited form as: Proc Mach Learn Res. 2022 Mar;151:9304–9333.

Figure 2:

Figure 2:

Sequential Ordering vs. Dynamic Ordering. Upper: We check each feature sequentially. Whenever we find a better feature, we always start from the beginning to find the next possible swap. Lower: We order the list, checking the feature that has failed the least amount of times first. We hold off checking less promising features until the end, saving substantial computational time.