Challenge 2K9/Pre Chat

From Robowiki
< Challenge 2K9
Revision as of 15:30, 25 February 2009 by Skilgannon (talk | contribs) (some changes, better represents rumble, more variety)
Jump to navigation Jump to search

Hi, I'm going to begin this challenge. Do anyone deny?

I'll have 4 sub-classes for this challenge:

  • /SmallBot Movement Challenge Y2K9: Test your nanobot or microbot movement against good micro/nanobot gun!
    • Reference bots:
      • Pattern Matching: WeeklongObsession, Smoke, Waylander
      • GF: HedgehogGF, RaikoMicro
      • Linear: Splinter, PRAL and SRAL
      • Other: Infinity and NanoDeath
  • /WaveSurfing Challenge Y2K9: test your bot against simple targeting.
    • By having three old bots, WSCBotA, WSCBotB and WSCBotC still act like it was in WSMCY2K6. Now, I'll add Mean Targeting to it! (By my test in BlackHole, most wavesurfer show its weakness with Mean Circular Gun.)
      • WSCBotD fire Mean Linear gun averaged on velocity at history depth of 10.
      • WSCBotE fire Mean Linear gun averaged on both velocity and heading at history depth of 10.
      • WSCBotF fire Mean Circular gun averaged on velocity at history depth of 10.
      • WSCBotG fire Mean Circular gun averaged on both velocity and heading change at history depth of 10.
  • /Curve Flattening Challenge Y2K9: how much can you flat the movement curve?
    • I'm thinking of Kawagi's entropy should use to rate each robot here. - not be used for comment below.
    • Reference bots:
    • #CassiusClay 2rho.01b
    • #Hydra 0.21
    • #Dookious 1.59 with AS gun off.
      • Note: Dookious 1.59 not 1.573c
  • /AntiPatternMatcherChallenge Y2K9: how much can you random?
    • Old, simple pattern matcher aren't interesting enough. How about Statistical Pattern Matcher?
    • Reference bot: modified versions of Toorkild/Waylander:
      • #advancing/lateral matching
      • #velocity/turn matching
      • #advancing/lateral MC matching
      • #velocity/turn MC matching
Main challenge

Like MCY2K7, let's vote for reference bot! Currently candidate are:

  • GrubbmGrb 1.2.4 - the old version from MCY2K07 can be used here, I've no more ideas.
  • Shadow 3.83
  • Komarious
  • Phoenix1.02
  • MaxRisk
  • Engineer
  • Some Angular Targeting bot should be best. Maybe very old Gouldingi with ABO gun off.

/Vote For Reference Bot

The reference bot for main challenge is separate over difference kind of targeting. I think it is good enough. If any one have more ideas, let me know.

The 'default' gun shall be voted, too, but later :)

  • old RaikoGun - Pretty Well - Chosen
  • DookiLighting with AS gun off - interesting
  • DrussGunDC - should work too good - will not be use if DrussGT is reference bot
  • Bee - too good, too.
  • Horizon DC Gun - ???

/Reference Gun Voting

A QUESTION: DO ANYONE WANT DYNAMIC CLUSTERING CHALLENGE THAT EVERY CHALLENGERS MUST USE DYNAMIC CLUSTERING MAINLY IN THEIR MOVEMENT?

» Nat | Talk » 15:19, 24 February 2009 (UTC)


Comments

I have a few comments:

  • Yes, NanoBot is good! I was just thinking we need this the other day.
  • WaveSurfing is not about dodging GF and AntiSurfer much. The biggest point of wavesurfing is being able to adaptively dodge ALL kinds of guns. Including head-on, linear, circular, pattern matchers, are also all very important in REALLY gauging how well a wavesurfer performs. In fact, in some ways with wavesurfing the following is a very good rule: Unless you can get 98%+ against head-on targeting WITHOUT firing, then you have some significant bugs/flaws that you could put all focus into fixing. Similarly, simple targeting is also very important to gauge. Only after you've perfected against the simple targeting, should you try to improve it against the fancier things.
  • You seem to indicate a small number of bots with how you say "best pair of" and such. Really, it's not a fair test unless you have far more challenge bots than that.
  • The main challenge could maybe with some additions/removals/updates probably as well
  • Actually, instead of "WaveSurfing Challange" I'd suggest a "Avoid the Simple Targeting" challenge, and a seperate "Avoid the Adaptive Guns" challenge. In fact, this could absorb the "main" one as well maybe. Just have one with everything from head-on to nano pattern matchers, and one with everything like high-end pattern matchers up to strong GF guns and AntiSurfer guns.

Other than that, cheers! --Rednaxela 19:45, 22 February 2009 (UTC)

I've change the class above, do you have any more comment on that, Rednaxela?

Also other robocoder (look for Skilgannon and Voidious), refused? If not the rules will going on within a week I promise. » Nat | Talk » 14:21, 23 February 2009 (UTC)

Not bad overall. With the curve flattening however, I think that using Kawagi's 'entropy' to measure is a bit sketchy. I'm simply not convinced entropy does actually measure vulnrability to GF targeting well. For one thing, any any movements that react to getting hit would not really give the same entropy information in different situations. I'd say that for 'curve flattening' it would make more sense to have it as a challenge with fired at by other bots just like the rest, in particular, but a widely varied set of GF guns, from simple unsegmented, up to very strong guns like you mention in the 'main' challenge. Also after that I'd say that really "Main" and "Curve Flattening Challenge" could be merged to make "AntiGFTargeting Challenge", and that "Movement Adaption Challenge" is unnecessary because the anti-GF and anti-PM categories would have already covered that. --Rednaxela 14:51, 23 February 2009 (UTC)

I don't think that Curve Flattening challenge and AntiGF is not the same! I sometime saw that Phoenix doesn't 'surf' the wave to lowest position, instead it almost keep itself at edge of the peek :) But I've merge Movement Adaption challenge into main challenge now. » Nat | Talk » 15:19, 24 February 2009 (UTC)

A few suggestions:

  1. Don't make all opponents fire 3 power bullets. It would be more realistic if we face these bots as they actually are.
  2. If you want to use DrussGT as a reference bot, rather use 1.3.0 as it was significantly stronger in the PL than 1.3.1. However, we already have Shadow as a WS+DC bot. The movement being GT doesn't really matter, because we are all using a standard gun. DrussGT is also fairly slow these days, and running many many seasons could be quite tedious.
  3. I suggest for the micro, that they fight the micro+nano bots, because they actually face both in the micro rumble. Similarly, the mega bots should face everybody, not just top bots.
  4. I'd suggest NanoDeath as the Rambot, because then we can include it in the nano section, and everybody can face it.
  5. We don't need to test nanos against GF, there is only 1 GF nano in the whole nanorumble so it has very little effect. It is also VERY slow, and will make the competition take much longer to run.
  6. I'd be willing to whip up a simplified, speeded up version of DrussGunDC that we could use as a reference gun. However, it will probably still run slower than Raiko's gun, which worked very well last time. But cutting down DrussGunDC will take very little time, and it will run much faster if I pull a smaller cluster and use only 6 or 7 dimensions.

--Skilgannon 18:41, 24 February 2009 (UTC)

  1. Actually, almost bot fire 3 bullets in its reference mode. But I think that for TC, right? So, the middle way. Some will fire 3 power bullet and some will not, OK?
  2. I'm thinking that between you and Horizon, who has stronger gun? The stronger shall be reference bot and the weaker shall be reference gun :)
  3. OK, merging micro and nano. I can merging it without thinking because I have to do a less work.
  4. NanoDeath is fine, but I like MaxRisk more. May add it to nano challenge.
  5. When merged, the more 'micro' GF will do that :)
  6. OK, I have look the difference from MC and MCY2K6 that almost everything is difference! I don't want the same thing and I want DC gun, so... I now know 4 GF-DC bots, LukiSaber, FireBirdGun, DrussGunDC and HorizonGun. FireBird is close source so no. LukiSaber is from Dookious so I'd prefer the DokiLighting more. Most DrussGT score is from movement so DrussGun may not as strong as those gun. The old Bee gun is very strong, the old MC score show that. I've run any battle with Horizon yet but I read his code :)

Note for you, Skilgannon, Can you improve the Toorklid for the APM challenge?

Does Voidious around here? I'd like to hear from him. » Nat | Talk » 19:19, 24 February 2009 (UTC)

I'd agree with Skilgannon's points. As far as firepower, actually, they only do that in TC challanger modes, not MC reference mode, many have those two mode seperate. I think having any stuck at fire power 3 that aren't normally, would HIGHLY distort the results so I wouldn't consider that acceptable. Also I'd say the "Curve Flattening Challenge" would need more lower-end GF guns, one being some unsegmented one, and another being something like Raiko's gun, and I'd say maybe remove one of the three you already list for "Curve Flattening Challenge" as it seems a bit redundant. I also believe that the "AntiPatternMatcherChallenge" needs more than one type of pattern matcher. Toorklid is of the "Multiple Choice Lateral/Advancing Velocity" variety, and I strongly beieve that all four of MC-Lateral/Advancing, nonMC-Lateral/Advancing, MC-Velocity/Turn, nonMC-Velocity/Turn all need representation. I'd also say that "statistical pattern matching" probably isn't so interesting, because to my knowledge almost no bots use it, and most patternmatchers are in too codesize restricted of an environment to generally be using it. I'd also say that "NanoLauLectik" isn't at all useful in the "Surfing" challenge because pateren matchers would already be covered well by the pattern matcher challenge, and nano-style ones would be covered by the nano section which Skilgannon suggests big bots should be run against too (which I agree with). I would note thought that while big bots should run against the nano section, they should be put in a seperate 'grouping' in the results page. As far as as a reference gun, I strongly suggest we stick with Raiko's gun. After all, it's small, simple to integrate into existing bot code, it's fast, a fair number of us are already familar with using it, and bot jars that were already made to be MC2K7 challengers can fight in the the 2K9 one without modification. --Rednaxela 19:31, 24 February 2009 (UTC)

Hmm, yes I hadn't considered being able to run old bots on the new challenge. It'll be a lot easier if we just use Raiko again. --Skilgannon 20:00, 24 February 2009 (UTC)


(Edit conflict) OK, I agree with the firepower now. For the 'statistical pattern matching', on the old wiki say with 2 words: PM + MC :so isn't that Toorklid? For the NanoLauLectik, you tell me on very first comment that we should separated the WS challenge to 2 challenges, one for HOT to nano PM, and one from advnaced PM to strong GF, don't you? I took care of every comment, but if you don't like it there, I shall not make it a reference bot for WS challenge. I think by mixing the Waylander with Toorklid can make all those 4 PM you want :) For the reference gun, I'm agreeing with you at last. Shall we close the vote right now? Or let see more people ideas. And, Nano-challenge? It is now SmallBot challenge. I have some crazy ideas here. If we moved the main challenge into Movement Adaption Challenge and the 'main' challenge score is the averaged of this sub-challenge would be nice but crazy. This mean you need to run over than twenty 500-rounds! » Nat | Talk » 20:07, 24 February 2009 (UTC)

Ahh, you're right about about 'statistical pattern matching', I just got mixed up, and generally I always hear it termed MCPM not SPM. And yes, it's true that combining/melding the features of Waylander and Toorklid could produce all four of those combinations. However in order to reduce risk of errors, AND (importantly I think) add more variety to the movement of reference bots, I think it would be good to choose different ones that already exist. About the surfing challenge, I kind of changed my mind, and decided that it only needs the simple bots because the other challenge categories will test other aspects of the bot well enough. As far as closing the vote right now? I'd wait a bit still perhaps. Also, here's another thought, I strongly think we should go with 35-round matches, not 500-round. After all I think many of us use the challenges on bots we have in the rumble, and in the recent past the 35-round challenge version seem more popular than the 500-round ones. --Rednaxela 20:34, 24 February 2009 (UTC)

OK, removed NanoLauLetik. I'd agree with you on rounds. I like 35 rounds much more since my bot go crazy slow after the round 100. So, the rule will be 15 session of 35 round each. Thanks. OK, now it is by the 4am in my country and I haven't sleep yet so see you tomorrow. » Nat | Talk » 20:46, 24 February 2009 (UTC)

Goodnight. Also, I'd say 15 seasons is plain not enough still, I think at least 30. For example see TargetingChallengeRM/Results (old wiki) where the number of seasons run is at least 50, and not uncommonly over 100, and also User:Rednaxela/SaphireEdge where I run TONS of tests at 110 35-round seasons, and even that I've found from experience does not five as much stabilty in the numbers as I'd like. --Rednaxela 21:06, 24 February 2009 (UTC)

People can run any amount sessions as they like, that just a minimum :) I don't want to increase it because even 35 session take about 6 hours on my heavily-loaded laptop computer right now. » Nat | Talk » 12:40, 25 February 2009 (UTC)

I've changed a few things, such as bullet power requirements, FunkyChicken --> Smoke etc. I've also removed DrussGT, as I think having more than 2 DC guns (Hydra, Shadow) is kind of pointless, seeing as there are probably only around 10 in the entire rumble. I also changed Komarious for WeeksOnEnd because otherwise we don't really have any light GF guns in the test, despite them being quite prolific in the rumble.

Just an idea, but how about using the same set of bots for the Targeting Challenge? We just put our gun on top of, say, Komarious's movement (WSGunChallenge platform) and run it against the same bots? Also, the WSChallenge bots ABCDEFG will have to be modified to move, otherwise they can't face our moving, firing bots. Unless we modified the Raiko gun to only fire if the enemy moves... --Skilgannon 14:30, 25 February 2009 (UTC)