Talk:WaveSerpent/Version History

2.0

I hope it's not too early to say... holy cow, great work! At 400 battles and 0.7% APS above Dookious. Still plenty of battles to go, but you've surely got the #2 spot. Amazing! Congrats! --Voidious 02:06, 29 September 2009 (UTC)

It looks like I've got some competition as an early birthday present! That's a huge jump for a single release! Very impressed =) --Skilgannon 05:03, 29 September 2009 (UTC)

Wow, that must be the greatest robocode comeback of all times, congratulations! :) --ABC 14:22, 29 September 2009 (UTC)

Those are some seriously impressive results there! I'm curious about how you note use of anti aliasing. Do you refer to anti aliasing between segments like I pioneered with SaphireEdge? Or do you mean anti aliasing between bins? Or what? I find it odd how you say that since unless WaveSerpent switched from DC to VCS, I don't see how anti aliasing would be applicable. --Rednaxela 14:52, 29 September 2009 (UTC)

"Movement now uses anti aliasing" and "Unlike most other surfers, WaveSerpent doesn't use bins" I think this answered your question, unless he changed something. --Nat Pavasant 15:00, 29 September 2009 (UTC)

Nope, it doesn't answer it at all Nat... You appear to have missed "New main gun with crowd targeting and anti aliasing.". And you also missed my point of asking what anti aliasing could possibly mean in the context of DC (which is what WaveSerpent uses for movement indeed). --Rednaxela 15:12, 29 September 2009 (UTC)

WaveSerpent doesn't use DC to surf, actually, Hydra does. He definitely mentions segmentation in his movement description - I think it has segmentation but not bins. And fwiw, I did get the impression he switched to VCS in the gun... --Voidious 15:32, 29 September 2009 (UTC)

I just took a look at the code of the new WaveSerpent's code, and it appears that is has switched both movement and targeting to SaphireEdge-style anti aliased VCS indeed! May I ask if this was inspired by SaphireEdge? It would be quite a coincidence otherwise :) --Rednaxela 15:35, 29 September 2009 (UTC)

The idea to use anti-aliasing did come entirely from SaphireEdge (although I never looked at your implementation). Although the gun now uses VCS instead of DynamicClustering, the movement in fact does not. It records segmented lists of guess factors where it has been hit and uses anti-aliasing between segments. --Kev

Very impressive update! Looks like I need to put WaveSerpent back into my test set... --Darkcanuck 15:15, 29 September 2009 (UTC)

Awesome comeback, nice job. --zyx 02:00, 30 September 2009 (UTC)

2.1

Not much for your APS, but it's done wonders for your PL! Only loss being Shadow (although DrussGT and YersiniaPestis are pretty close calls) --Skilgannon 13:36, 29 November 2009 (UTC)

Wave Serpent now has losses to Pris and DrussGT, but it's still interesting that adding a simple PM gun boosted its PL score. Maybe I should look into making a specifically anti-surfer pattern matcher. --Kev 06:00, 30 November 2009 (UTC)

I suspect PM appears strong against surfers when added to VG scheme for the following reason: It provides a decent guess at enemy behaviour, but being different enough that it makes the old GF-surfing data unhelpful. Then once the surfer learns the tendencies of the PM gun, the GF gun becomes stronger again. I suspect that the PM gun on it's own would be worse than the GF gun on it's own, so I think gains you could see from anti-surfing tuning of a PM gun would be minimal at best. With various experiments with both RougeDC and SaphireEdge, I had discovered that against strong surfers, there tended to be an oscillation of which guns were most favored. --Rednaxela 06:58, 30 November 2009 (UTC)

I've noticed that too, but I've also seen that in some matches, such as when WaveSerpent fights itself, it switches to the pattern matching gun fairly early on and then uses it exclusively for the rest of the battle. Perhaps one reason the pattern matching gun works well against surfers is that the other robot's curve fattener works against it, since having a flat movement profile is worthless against pattern matchers but could make the movement predictable. --Kev 08:15, 30 November 2009 (UTC)

I hate to always be the skeptic =), but I wouldn't draw too many conclusions about the PM gun's PL prowess before running some longer benchmarks against its PL rivals. I've seen nearly identical versions of Diamond fluctuate between 6 and 14 losses pretty regularly. Also, I remember your SeaSerpent PM gun was pretty strong against surfers, wasn't it? Did that have any special AntiSurfer magic? --Voidious 20:47, 30 November 2009 (UTC)