 |
| Stephen Brooks 2007-12-21 18:12:56 | Hmm! It turns out the baseline figure of 4.x% I used so far was calculated with version 4.42b, which did not have the change the particle decay algorithm that 4.43 introduced. The same baseline result retested with 4.44c gives the following (on random seeds, because PhaseRotC didn't have deterministic seeds): 3.039262,3.072225,2.999992,3.034676 So by most judgements, PhaseRotEb1a has already exceeded this baseline result. It's Christmas break now, but I think I'll modify PhaseRotC to use the same seeding as the PhaseRotE lattices and then generate a comparable number, so we can see how far ahead the optimisation actually is. |
| Stephen Brooks 2007-12-21 18:21:59 | Score after quarantine was 3.048721 but this is still non-deterministic, trying with the deterministic lattice now... |
| Stephen Brooks 2007-12-21 18:34:03 | The deterministic variant has started off on 3.004222 for seed #0 |
| Stephen Brooks 2007-12-21 19:08:05 | First four scores (seeds 0,1,2,3) on the deterministic one are: 3.004222,3.031520,3.015786,3.046772 |
| Stephen Brooks 2007-12-21 19:20:52 | Passed quarantine with score 3.030399. I'll now update the charts. So it turns out instead of trailing the PhaseRotC design by some 30%, we've actually just beaten it, in a fairly short period of time comparatively. I'm off home now, so Merry Christmas to everyone! |
