Editing revision 55 of AntiPatternMatcherChallenge/APMCIndex

Robo Home | Changes | Preferences | AllPages

Editing old revision 55. Saving this page will replace the latest revision with this text.

==== APMC (AntiPatternMatcherChallenge) Index ranking ====

Bots above the dotted line have completed the challenge and have beaten PatternMatcherBot.

You can calculate the INDEX like this: <pre>(your_score / patternbot_score) * 100.0 </pre>

----
Wow, Paul. That's an amazing result! -- [[PEZ]]

I know, - second place is soooooo strange -- [[Paul Evans]]

=) But without adding a reverse pattern matcher you must probably get used to it. -- [[PEZ]]

Now I'm not sure any longer that you need to run a pattern matcher to get first in this challenge. Though I guess [[rozu]] can improve GlowBlowAPM to a point where it's impossible to follow without taking the APM path. -- [[PEZ]]

I don't now. but I think we should add another reference bot to the AntiPatternMatcherChallenge, as iiley already proposed... -- [[rozu]]

Iiley's bots are open source, and I even just released an 'index' version of FloodMini, these could be used as additional tests to these fairly arbitrary ones.  Although I'd guess this side of the challenge would be somewhat more difficult with Aspid's gun at the center. -- [[Kawigi]]

One amazing thing!!Walls gets 53.2% in 1000 rounds........--unknown

I doubted that last comment's validity, so I ran it myself... Holy cow! Walls (without firing) got 71.51% of PatternMatcherBot's score, and won 802 of 1000 rounds.
<pre>
1st: lv.Patt...1.3	67509	9900	1980	51681	3948	0	0	198	802	0
2nd: sample.Walls	48278	40100	8020	0	0	158	0	802	198	0
</pre>
I added Dookious 0.50's score to the list, too.
-- [[Voidious]]

I think I've done something wrong here... I ran Capriite against the pattern matching bot first in a quick 200 round battle and got 300% with that formula for score, but nobody has higher than 100%. I thought I did something wrong so I ran the whole 500 rounds and got 349%... now I know I did something wrong. What is it?
<pre>
1st: DM.Capriite 3.7.2	        83366	24000	4800	45173	8632	723	36	482	20	0
2nd: lv.PatternMatcherBot 1.3	23875	1000	200	22181	493	0	0	20	480	0
</pre>
I did seem to win a lot of them, 482 out of 500, 96.4%.
-- [[Damij]]

Welcome to the wiki! Your bot should not fire during this challenge. That's what's throwing off your score. --[[wcsv]]

This was a good test of theory.  If I run an anti-pm movement, it will only really dominate against the same pm style gun.  Against Che (for the APMC2K6) LifelongObsession APMC lost 2 out of 500 battles.  Because it runs a reverse-Che-pattern-matcher.  Not all PM guns are created equal!  -- [[Simonton]]

Wow, lesson #2: don't give up on the theory right away!  :)  This is basically like throwing in a little (anti-) CrowdTargeting: I composed the reverse pattern matcher with reverse head-on-targeting, reverse linear targeting and reverse circular targeting, and now *that* seems to give pretty good results!  I was just telling someone yesterday that when a good idea comes along it seems to come from several people at the same time (read "crowd mevement").  I'll leave both results up there for now so y'all can see it clearly, someone can feel free to erase one of them in the near future if they'd like.  -- [[Simonton]]

Simonton, at this rate, your gonna be getting some really evil looks, most notabily when you release this APM + WS bot!

See, if you write a bot that moves exactly how everyone doesn't expect you to all the time, the bot should probably have a name like DevilsAdvocate. -- [[Kawigi]]

I think I know why walls does so well against EnderPMC. I had the same bug in my micro pattern matcher, I haven't looked at Ender's source, but here's my theory: walls is always on the border, and your bounds checking in the iterative is doing the following: 
<pre>
if(!new Rectangle2D.Double(18,18,getBattleFieldWidth() - 36, getBattleFieldHeight() - 36)
   .contains(enemyX + getX(), enemyY + getY())){
    break;
}
</pre>
The problem with this is that the 'contains' method does not include the border pixel, so the 18 should be replaced with 17 and the 36 with 32.  -- [[Skilgannon]]

IIRC, the 18 should be fine.  I use something like 35.9999999999 instead of 36 to solve the problem in my pattern matchers. -- [[Simonton]]

Summary:
Consciousness check: 11 + 6 =
This change is a minor edit. Send email notification that AntiPatternMatcherChallenge/APMCIndex has been changed.
(Visit Preferences to set your user name.)

View other revisions
Robo Home | AntiPatternMatcherChallenge | Changes | Preferences | AllPages