Talk:Barracuda Challenge

From Robowiki
Revision as of 03:19, 25 January 2011 by Rednaxela (talk | contribs) (→‎Current talk: Congrats)
Jump to navigation Jump to search

From old wiki

I doubt that 1000 rounds is enough for differences in score between CC, Shadow, and RaikoMX to be statistically significant. Also, looks like YALT is a wee bit buggy. --David Alves

I could do better if I ever got around to putting some code in there for detecting special cases. As I have none I can sometimes get tagged a bit. -- jim

I think 1000 rounds is enough. The current version of CC is very stable around 99.2%. Maybe the difference between Shadow and RMX is something we could call a tie. It could be tested I guess. But running Shadow thousands of rounds isn't my idea of fun. =)

As for special casing. I know RMX used to special case head-on fire, but CC doesn't and I don't think Shadow does either. CC does special case against more complex firing though, where it switches on a some sub currents in the surfing.

Noone has asked for it, but here's my howto to go for 99+%. I have focused on:

  • ironing out bugs
  • going more and more precise in my calculations
  • maintain a minimum distance
  • giving my bot more than one remaining option when the forward option brings it too close to its enemy (due to WallSmoothing)

-- PEZ

Anyone find it curious how the order of the bots in the above results table reflects the order of the bots in the RoboRumble Rankings? -- PEZ

I have long suspected that the top bots are better against all others than I am, especially the poorer aimers. This does not suprise me. I would also like a Barracuda that shoots near +1 to see how those look too. I would suspect that the results would be the same. -- jim

Try wiki.nano.DevilFISH for that. It's a little harder to avoid linear targeting as perfectly as head on targeting though. -- Pulsar

Please make a GF1 Barracuda and post some results. DevilFISH poses other problems than just firing linear. It's a close combat bot too. -- PEZ

A GF1 Barracuda wouldn't pose much of a challenge. For instance, sample.TrackFire would get a very high score against it just by sitting still. --David Alves

Would it? I would think it would fire head-on on a stationary bot. -- PEZ

Firing directly at the bot is GF0. Firing with maximum lead on the shot is GF1, maximum negative lead is GF-1. This is true regardless of the bot's current velocity. --David Alves

Yeah, maybe GF1 is a bad label. I'd like a Barracuda with the DevilFISH gun to test against. But a /DevilFISHChallenge is interesting in itself since it also tests your closer combat skills. -- PEZ

Do you mean LinearTargeting? That might be an interesting challenge. As for DevilFISH, the fact that it's a close combat bot makes me less interested. This /BarracudaChallenge is fun because as a person watching the battles, I can see how a robot could get 100%. At close range perfect scores are not possible. Why don't I make a WaveSurfingChallenge where you try and get the maximum score without firing vs. 2 stationary bots, one that fires with LinearTargeting and one that fires with HeadOnTargeting? That would be a more formal competition along the lines of the MovementChallenge and TargetingChallenge. --David Alves

Yes, maybe I mean LinearTargeting. Though I find using it on Barracuda instead of some stationary bots much more interesting. Barracuda has a great movement, designed for competetive OneOnOne battle. The results in this challenge seem to reflect the competetive quality of your own movement. Something that can not be said about the MovementChallenge. But please feel invited to create those MovementChallenge reference bots. I think it would enrich the MC. DevilFISH is too interesting to test against I think. You'll see CassiusClay's results in the /DevilFISHChallenge soon. =) -- PEZ

Hi u all, folks! Being far a little from RC, but planning to return very soon... Great idea this HOA Challenge (as well as the DevilFISHChallenge - DF was allways one of my favorits testing bots). For now, i´ll run SS in both challenges and post the results. I´m planning an entirely new WS bot, where i´ll re-do all code from zero. Probably starting with a light-weight cannon (a simple GF cannon probably). Good to see(read) u all again!! -- Axe

Just added two MusashiTrick bots: Musashi & Raiko... Jam is indeed one of the greatest RC-Masters ever! 91.325% with a non-WaveSurfer bot! Beautiful! -- Axe

CassiusClay with head on HOT got ~95%, with its own gun it gets over 99%. SilverSurfer with or without HOT gets over 99%. To me that suggests that SilverSurfer's movement is much better than CassiusClay's, but CassiusClay's gun is much better than SilverSurfer's. --David Alves

  • I have no doubts that CC´s gun is a lot better than mine... -- Axe
  • I also suspect that SS's super-heavy-and-slow gun might interfere in this kind of Challenge (it skip some turns in my computer), that could explain the good performance with a simple HOT instead... -- Axe

For a comparison i also ran the previous version (0.53) of Freya which uses MinimumRiskMovement. Freya 0.61 uses a combination of MinimumRiskMovement and WaveSurfing/ShrapnelSurf for movement. Here are the results.

Bot Author Score share
Freya 0.61 Loki 89.023%
Freya 0.53 Loki 77.182%

--Loki

My MusashiTrick is ok, as it scores comparable to the best. -- GrubbmGait

1st: davidalves.Phoenix 0.312b	102703	50000	10000	35586	7117	0	0	1000	0	0
2nd: kawigi.sbf.Barracuda 1.0	561	0	0	552	0	9	0	0	1000	0

4th? aww... --David Alves

Added the score for Dookious 1.09 (with gun data saving off) over 30 seasons. -- Voidious

Awesome score Kev! Your movement seems even stronger than your gun. -- GrubbmGait

Thanks! I think you're right, but I'm not too sure by how much my movement is better. WaveSerpent's movement does especially well against simple targeters like Barracuda. It can consistently get 99.9+ against WaveSurfingChallengeBotA, but its other movement and targeting challenge scores aren't as impressive. If it's ok with Voidious, I'll release a version of DookiSerpent (Dookious's gun and WaveSerpent's movement) to see which part of my bot I need to work on more. -- Kev

That's cool with me, of course, I'd be curious to see the results too. -- Voidious

Great, I'll put it in the rumble. It will be interesting to see how it does. -- Kev

Updated Dooki's score. It's over 5 seasons and is the first time I've run this since the full rewrite of the tank. I guess something subtle is making a big difference, it's down about 0.4%... -- Voidious

Well, it doesn't really hurt the performance in the rumble, but it is a bit nagging in the background isn't it . . . ;-) -- GrubbmGait

  • Or maybe it does hurt it in the rumble? ;) -- Voidious

I thought Seraphim would get about 95%, considrering I never had this in mind during her design. --Chase-san

Glad to see I've got my HOT "issue" worked out in 1.534. That's 3 seasons. -- Voidious

Current talk

Awesome score with Diamond 1.5.25! Is Diamond that good (probably) or does Robocode 1.7.2.2 have some infuence. --GrubbmGait 00:19, 25 January 2011 (UTC)

Thanks. =) Well, the surfing code came from Dookious and I've made improvements since then, like Waves/Precise Intersection and Gun Heat Waves, so I expected it to do better than Dooki. I don't know about Robocode 1.7.2.2 - the new decel-through-zero rules should screw up my predictor slightly, but I think there are also plenty of bug fixes. --Voidious 00:56, 25 January 2011 (UTC)

Congrats on Diamond's great score indeed! Getting such a high score with this is certainly a testament to well written wavesurfing! (Hmm.... those two improvements... I'm very familiar with those two things... :)) --Rednaxela 01:19, 25 January 2011 (UTC)