[Home]StrongestBotsRumble

Robo Home | Changes | Preferences | AllPages

Showing revision 9

The Strongest Bots Rumble

What about if we choose the top-10 PremierLeague bots (for instance) and let them duke it out in 1000 round battles? 10 particpants woule mean 45 battles to run ((10 * 9) / 2). Or 45,000 rounds! Maybe I should set up a RR@H server for this?

Ascendant beats the crap outof CC huh? And beats DT and Shadow too! Now maybe Ascendant would win that StrongBotsLeague?... -- PEZ

What about the top 20 in regular 1x1? -- Axe

Like 190,000 rounds would make it easier? =) -- PEZ

Besides. Being a top-20 bot in regular RR rankings doesn't say anything if you're a strong bot or not. Just trash the weak bots really good and you're there. I think the PL is a much better measure. -- PEZ

Sounds interesting. Not easy to select the bots that may participate though. The top 10 of the PL has the disadvantage of being decided by some close battles that really could tip either way. Just some bad luck could drop a bot out of this StrongBotsLeague? until the next release (or maybe re-release)... --Mue

If you drop out of PL top-10, then you're just not strong enough. -- PEZ

Since I never hesitate to contradict myself... What about we let bot 20 through 6 to duke it out in a qualifying eliminiation cup to decide which ones fit the StrongBotsRumble?? -- PEZ

I like this idea :-). Perhaps we should take bots 6 to 21 (since 16 bots make a fairer grid), and the 4 bots that survive the first two rounds (consisting of 1000-round battles each?) are qualified? That would result 48,000 rounds (12,000 for qualification and 36,000 rounds for the league). --Mue

36,000? -- PEZ

That's equivalent to a metric ass-ton of battles. How about you run 100 round battles instead, and only run a 1000 round battle for a particular pairing if the 100 round battle is too close to call (maybe 60-40 or less) --David Alves

Yeah! A metric ass-ton! Neat name. But that won't stop us. It's just a matter of letting the PC work some. =) I calculated it to 45,000. That's why I asked. Isn't it (n(n - 1)) / 2? How hard would it be to add this tournament setup to the RR@H client you think? Anyone want to try? I guess I can try it, but first I must go figure why Ascendent eats CC alive. -- PEZ

Well, i thought that determining 5 bots in an elimination cup with 15 participants is somewhat strange, so i decided its better to pick only 4 out of 16 (then there would be only 9 bots for the league, and thus only 36 battles needed). But thinking about that, i see that this is not a problem. One could as well take 5 out of 10 or out of 20 bots, or .... Hehe and dont worry too much about the CC-Ascendant-result, the new version has not yet proven itself in the rumble and could be buggy (and bugs can explain a lot). Its not even uploaded yet, since i still want to see what score version 0.9b got. --Mue

48,000 rounds sounds like enough. Can you line out the rumble as you see it in some detail? Then we can figure on how hard it would be to set it up. -- PEZ

* Runs off to see if DarkHallow gets an automatic bearth* -- jim

OK, i'll try. First of all the PremierLeague should be complete (each bot has fought all opponents) when the initial pool of participants is determined. Otherwise we could miss some strong bots, that are missing some wins from pairings not yet fought. Then run a qualification tournament for bots 6 to 21 from the PL to determine the participants of the StrongestBotsRumble. The first round in that tournament could look like this:

a  6 vs 21
b 13 vs 14
c  7 vs 20
d 12 vs 15
e  8 vs 19
f 11 vs 16
g  9 vs 18
h 10 vs 17

The winners advance to the next round, which could look like this:

winner a vs winner b
winner c vs winner d
winner e vs winner f
winner g vs winner h

This way the bots are seeded based on their PL ranking, so that bot 6 will have the easiest and bot 21 the hardest time to qualify. The top 5 of the PL and the 4 winners of the second round of the qualification tournament are qualified. Well and then one battle for each pairing of these 9 bots is needed. No idea though how hard this would be to set up. --Mue

Sounds great. Thanks! Now that shouldn't be too difficult. The problem will be to find a moment when there are 21 PL bots with full pairing tables. But sometimes that happens and then we should try to be ready I guess. Maybe the PRIORITY_BATTLES feature of the RR@H client could be used for the elimination round. If not maybe we can run that one by hand first and then use a more regular approach to run the final 36 battles. -- PEZ

Well, now that we have a working PremierLeague, i'd like to bring up this topic again. The problem is that i dont really know how this kind of tournaments can be held using the RR@H approach. I could also live with manually running the elimination round (and maybe even the final round), but automizing as much as possible will make it easier to repeat this tournament every now and then. So any ideas of how to set up a tounament like this? --Mue

I know Albert used his Face2Face tournament as a basis for the RR@H client. It was an eliminiation tournament all together. It would be good if we could distribute the matches. Even using Jonathan's PowerBook this would take quite a while on just one machine. We could maybe make a special servlet for it. One that hands out pairings to a client that asks for it... -- PEZ

So then we'll also need another client (in addition to the normal RR@H client)? --Mue

Maybe so. Or maybe it's just a variation of the current client. Maybe the start of a new, smarter, version? -- PEZ

Maybe this is a good time to mention that I have secretly started a total rewrite of the RR@H client. It started when I read through the code a few days ago and found it very hard to follow and noticed several places where it could use a better design. I havn't done much yet though (it can download the participants list, but thats about it). -- Strider


Robo Home | Changes | Preferences | AllPages
Edit revision 9 of this page | View other revisions | View current revision
Edited October 10, 2004 17:38 EST by Strider (diff)
Search: