Figure 6 of the StyLit paper shows a knee point finding step on the sorted errors of a reverse NNF. It looks like this implementation only ever uses regular NNFs (i.e. target-sized 2D arrays of indices in the source). Am I misunderstanding the implementation here? It seems like this implementation uses algorithm 1 from the StyLit paper, but never does the step in the paragraph that starts with "To complete the algorithm." I would appreciate any insights, thanks!
Hi,
Figure 6 of the StyLit paper shows a knee point finding step on the sorted errors of a reverse NNF. It looks like this implementation only ever uses regular NNFs (i.e. target-sized 2D arrays of indices in the source). Am I misunderstanding the implementation here? It seems like this implementation uses algorithm 1 from the StyLit paper, but never does the step in the paragraph that starts with "To complete the algorithm." I would appreciate any insights, thanks!