3 ms·
Wait, so the authors' method is only a tiny fraction faster than the parallel version of the more general Rao-Sandelius method, and uses more random bits? But t
by blt 11y ago
Wait, so the authors' method is only a tiny fraction faster than the parallel version of the more general Rao-Sandelius method, and uses more random bits? But they claim it's "very fast" and "extremely fast"? Am I missing something?
- teraflop 11y agoYeah, the experimental results are pretty poorly presented, but on the other hand the abstract says: "While this algorithm is of lesser practical interest, we believe it may be of theoretical value." The number of random bits used by both algorithms is within a few percent of the theoretically optimal value (4.6% more for Rao-Sandelius and 5.5% more for MergeShuffle) so I don't think that's really a significant concern.