Unfortunately, it looks like there is a large hidden constant in there: https://www.youtube.com/watch?v=1deOVlruc-o#t=13m15s https://www.youtube.com/watch?v=1d

Unfortunately, it looks like there is a large hidden constant in there: https://www.youtube.com/watch?v=1deOVlruc-o#t=13m15s https://www.youtube.com/watch?v=1deOVlruc-o#t=13m15s > By considering a lot of cases, find O(1) candidates to the flip nearest u that improves the embedding. Also a comment underneath by one of the authors: > We have not tested it (yet), but I suspect the current version will be too slow in practice due to the ~400 cases it has to check for each flip. Also, some of the core data structures we use do not yet have a good practical implementation. I do plan on making an implementation at some point, but that could be months or years away. :-(