2011-06-21T00:03:35 *** antimatroid has quit IRC (Ping timeout: 250 seconds) 2011-06-21T00:15:45 *** Zaphus has quit IRC (Quit: Page closed) 2011-06-21T00:20:23 *** pvarga has joined #aichallenge 2011-06-21T00:21:23 *** pvarga has quit IRC (Client Quit) 2011-06-21T00:21:58 *** pvarga has joined #aichallenge 2011-06-21T00:23:43 *** pvarga has quit IRC (Client Quit) 2011-06-21T00:24:24 *** pvarga has joined #aichallenge 2011-06-21T00:30:29 *** pvarga has quit IRC (Quit: pvarga) 2011-06-21T01:06:45 *** Zaphus has joined #aichallenge 2011-06-21T01:10:04 *** okay is now known as okayzed 2011-06-21T01:10:59 *** slkjcna_ has quit IRC (Read error: Operation timed out) 2011-06-21T01:11:03 *** slkjcna has joined #aichallenge 2011-06-21T01:22:44 *** mleise has joined #aichallenge 2011-06-21T01:30:14 how was the meeting going? 2011-06-21T01:30:15 *** amstan has quit IRC (Ping timeout: 250 seconds) 2011-06-21T01:41:03 *** _flag_ <_flag_!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (Quit: Lost terminal) 2011-06-21T02:08:48 *** stocha has joined #aichallenge 2011-06-21T02:08:58 how does trueskill do pairings really ? 2011-06-21T02:10:31 what do you mean by pairings ? 2011-06-21T02:11:01 Zaphus: how is the choosing of which bot to put in the same game fighting one another done 2011-06-21T02:11:37 Trueskill doesnt do that, trueskill just generates ratings. However if you want bots of similar strength in a battle, use ones of similar rating 2011-06-21T02:12:42 of course that slows down your ability to 'climb' the ladder, mixing it up a bit gets better results (in my opinion) 2011-06-21T02:14:50 plus you'd hate to play the same bots over and over again, which is what would tend to happen 2011-06-21T02:16:14 *** janzert has quit IRC (Read error: Connection reset by peer) 2011-06-21T02:17:25 stocha: is the reason you keep questioning Trueskill because it's giving rankings that dont match with the data (the game results) ? if so that may indicate it's not implemented correctly. now that the java bots are playing and everyone's getting lots of games, it seems to have settled pretty nicely from what I see 2011-06-21T02:18:13 *** janzert has joined #aichallenge 2011-06-21T02:19:30 *** Palmik has joined #aichallenge 2011-06-21T02:21:09 stocha: the one thing I'm not a fan of with this implementation is the increase in sigma (decreasing the probability that your rating is correct, effectively) that happens when you don't play games - but I kind of understand the logic behind it 2011-06-21T02:23:30 Zaphus: whats the logic behind ? 2011-06-21T02:26:50 stocha: it's "we haven't seen you for a while, so maybe your rating isnt quite correct" - and by increasing sigma it gives you the chance to move more after the result of the game than you otherwise would. helps take into account the new/changed bots may have affected your rating even if you havent played them yet 2011-06-21T02:27:01 for humans it's more about the fact that we aren't consistent 2011-06-21T02:27:03 Zaphus: I don't have enougth data to judge what bot are stronger. Though i'd still want to see those matrix for each bot. I'm too lazy to do it without a ready gui :o However, i watched the ranking a lot, and it feels that it moves a lot. So if you trusted the values, depending on when you look at it, you would draw out different conclusions 2011-06-21T02:27:39 I'd love the matrix - I have enough data in the SQL to get it now (no idea how to make a matrix from it, but I'll give someone the commands) 2011-06-21T02:27:55 Zaphus: yeah, but bot are consistent. And new submission should not share anything with old submission. So really this feature is not desirable. 2011-06-21T02:28:57 but your bot isn't consistant. I can see that you sometimes beat davidju, and sometimes you lose - thats inconsistant 2011-06-21T02:29:11 davidjliu I mean 2011-06-21T02:29:23 Maybe the variations i observe are due partly to this feature (especially since sometime bots don't get a game in a day or so). But they i also think the pairing make for hard to draw conclusions in the first place. 2011-06-21T02:30:12 look today, all bots are getting multiple games an hour - and ratings are settling 2011-06-21T02:30:17 Zaphus: the bot is consistent, in the sense that it does always play the same. There is a lot of randomness in games. But for a given set of opponent, you'll end up with a probility value that one is first for example. 2011-06-21T02:30:40 Zaphus: but with humans, you woudln't have that feature. Because they tend to change over time. Try new things. Adapt, etc. 2011-06-21T02:31:10 stocha: correct, if the odds are 60:40 that you will beat Zaphus, and then you beat me three times in a row, your rating will increase and it will think you are now (for example) 80:20 - and then I'll win a few and it will move back down. 2011-06-21T02:31:18 The pairing is very important, because of the multplayer aspect. 2011-06-21T02:32:18 Zaphus: the trouble is, i may have 60:40 on one on one, and then 30:30 if we add another well choosen bot. And then 70 if we add another well choosen bot 2011-06-21T02:32:23 *** antimatroid has joined #aichallenge 2011-06-21T02:32:57 My retreating bots, seems especially good at disturbing my stronger bots, in multiplayer games. 2011-06-21T02:33:12 rock paper scissors 2011-06-21T02:33:27 so while i know how they do when i put only them in games, the ranking from the beta server, is all random. 2011-06-21T02:33:42 here's the question, look at stoha2 - it's had more or less the same rating for the last 3 hours, and has played something like 30 games in that time 2011-06-21T02:34:05 where's the randomness in there ? 2011-06-21T02:34:23 Zaphus: at this particular moment, the rating do are consistent with my data. 2011-06-21T02:34:41 matches mine too 2011-06-21T02:35:45 mostly thats from the last two fixes that had nothing to do with ratings - it was to fix it up so everyone got a game, and to get Java to work :-) 2011-06-21T02:36:37 but those changes had a big effect on the rating because now everyone gets to play - crashes messed up the ratings (sort of, it's actually accurate to say those bots were erratic) 2011-06-21T02:38:40 stocha: ignore whats happened with the rating up until today, see how it fares over the next couple of days. I *think* it's a pretty good indicator of the strength of the bots at the moment from what I see 2011-06-21T02:38:55 still the fact that sto_greed_002 is 8th leagues under davidjliu seems strange. Unless david changed his code of course 2011-06-21T02:39:31 but being there is only one submission of both, i find it hard to believe. from what i have seen in the past. Maybe my conclusion where wrong though. 2011-06-21T02:40:12 sir_macelon is the strongest, wich seems consistent with my own feeling. 2011-06-21T02:40:58 stocha: Davidjliu has played you 7 times today, and finished ahead of you in 5 of those - so I'd say he does deserve to be higher than you 2011-06-21T02:41:37 well it's not me, it's only a ref_bot that is in fact public now. :p 2011-06-21T02:41:45 but you get what I mean 2011-06-21T02:41:51 Although i have some idea to test, i didn't put them yet. 2011-06-21T02:42:06 5:2 ratio means he should be ranked higher than that bot, all else being equal 2011-06-21T02:42:35 there was far more than 7 games played involving those two bot. 2011-06-21T02:43:05 if it only use 7 out of many, then yes the conclusion over the 7 are right, but what about all the data it has gathered ? 2011-06-21T02:43:06 yeah, ignore before today 2011-06-21T02:43:18 obviously with 7 game you can only have poor rating 2011-06-21T02:43:41 well no, you've got 50+ games - only 7 involved him today though 2011-06-21T02:43:48 Zaphus: why does it ignore before today 2011-06-21T02:44:02 zaphus: no, I'm telling you to ignore before today :-) 2011-06-21T02:44:12 stocha: I mean you, not me :-) 2011-06-21T02:44:42 Zaphus: why should i do that ? 2011-06-21T02:45:15 with 7 game, you can hardly draw out any conclusion. Unless you somewhat have microdata (like a human can look at a bot behavior and draw conclusion from that. eventually) 2011-06-21T02:45:25 because the other things were tainting the ranking. However, Trueskill (like Elo) is an on-the-fly rating system, it keeps adjusting after each game - you don't go back over the entire history each time a game is played 2011-06-21T02:45:53 What's the use of that, with bot that play with consistency ? 2011-06-21T02:45:55 stocha: you have way more than 7 games data, just 7 involving him. but you've played lots of people who've played him and Trueskill uses that too 2011-06-21T02:46:03 A ranking that use all the data is far better in there. 2011-06-21T02:46:40 this conversation will last forever :-) offer something better, I can't keep arguing ! 2011-06-21T02:46:47 Zaphus: yeah, obviously those indirect data are less reliable. 2011-06-21T02:47:04 i already offered something. 2011-06-21T02:47:10 what was it ? 2011-06-21T02:47:47 Calculate out of all data, a value from the probability of winning for each pair of bot. 2011-06-21T02:48:08 i think that's what Bayeselo would. if not i'm sure there are algorithms to do just htat. 2011-06-21T02:48:38 so you have a grid of results , but where's the ranking? 2011-06-21T02:48:53 this n vs n thing complicates things 2011-06-21T02:48:53 You'd have to ponder direct data much heavier than inderect data. Let's say 10 time more. 2011-06-21T02:49:07 the grid is the objective basis. 2011-06-21T02:49:14 then you can use any system wich rely on it. 2011-06-21T02:49:22 i think bayeselo rely on it. 2011-06-21T02:49:30 beside previous ranking used bayeselo. 2011-06-21T02:49:35 so the algorithms are already ready. 2011-06-21T02:50:11 Guess I'm biased, I'm never going to agree that Bayeselo gives a better ranking 2011-06-21T02:50:13 I have no idea how they came to the conclusion it was good for 1vs1 games, and bad for multiplayer free for all games. 2011-06-21T02:50:41 what about ai-contest publishes the sql of scores and rankings 2011-06-21T02:50:49 and everyone can try his own ranking mechanism 2011-06-21T02:50:57 i would give a shot 2011-06-21T02:50:58 i don't know bayeselo really. Except for the fact that some version of it was used for sorting out features of goplaying bot. And it worked quite well indeed 2011-06-21T02:51:00 it's there on github 2011-06-21T02:51:10 the .sql of the beta server? 2011-06-21T02:51:16 oh no, the code 2011-06-21T02:51:21 yeah, i ment the database 2011-06-21T02:51:25 of scores only 2011-06-21T02:51:32 if they gives the data, then people can't complain i guess :) 2011-06-21T02:51:40 but everyone will just try to find the algorithm that gives the ranking order that they believe is correct :-) 2011-06-21T02:51:45 If they are two lazy to work on it, too bad for them : 2011-06-21T02:52:11 Zaphus: yeah? so? at least there is something to take a look 2011-06-21T02:52:14 Zaphus: hopefully most method but the very twisted one would give about similar conclusions. 2011-06-21T02:52:42 stocha: why is it that you say you don't know Trueskill, and dont trust it, and you don't know Bayeselo and do trust it ? whats magic about Bayeselo that convinces you where the science behind Trueskill doesnt ? 2011-06-21T02:52:53 stocha: I'm assuming it's not just anti-microsoft bias :-) 2011-06-21T02:53:10 Zaphus: i have seen it used for other thing. Building go playing bots. and it worked for that. 2011-06-21T02:53:48 show me how trueskill was able to sort out go playing bot features efficiently, and built top world go bot, and i'll trust it more. Even if i don't know the internals well. 2011-06-21T02:54:08 http://research.microsoft.com/pubs/67956/NIPS2006_0688.pdf - note the word "Bayesian" in the title, does that make it better :-) 2011-06-21T02:54:30 Beside, bayeselo, i have read article about. So although i have no precise view of the working, i have feelings about how it might work :p 2011-06-21T02:55:29 "can infer rating out from team play" I see the value of it for games like warcraft. However in ant, there are no teams. 2011-06-21T02:56:10 team play can mean free for all too 2011-06-21T02:56:28 Both bayselo and trueskill claim to be better than ELO. But elo is an old broken ranking system isn't it ? 2011-06-21T02:56:42 team play means team play. For example 2vs2vs2 2011-06-21T02:56:48 free for all, means no team play. 2011-06-21T02:57:09 elo isn't broken, it's just limited 2011-06-21T02:57:21 It doesn't mean it doesn't work without team play. only that it was built for a need we don't have here. 2011-06-21T02:57:22 trueskill is used for lots of free-for-all games (no teams) 2011-06-21T02:58:16 there are others as well - in the end one needs to get chosen, and the organisers chose this one. I'm yet to see it giving the wrong results (based on the input) so I remain on the 'keep it' side 2011-06-21T03:00:02 i thought bayeslo doesn't work for multiplayer games? 2011-06-21T03:00:04 only 1v1? 2011-06-21T03:00:17 stocha: did you comment about sharing your code? 2011-06-21T03:00:26 personally i'd prefer to go back to having people not share source code again now 2011-06-21T03:00:30 get my* 2011-06-21T03:00:57 Yeah one thing that makes distrust true skill is the fact that over and over some people here said it's not working well for 1vs1 2011-06-21T03:01:15 wich would mean it's not reliable really. but maybe it was not right to say that. 2011-06-21T03:01:16 no, i never said that 2011-06-21T03:01:20 1v1v1v1 (a v b v c v d) is just a v b, a v c, a v d, b v c, b v d, c v d 2011-06-21T03:01:24 i said the scoring system doesn't work for 1v1 2011-06-21T03:01:29 i said nothing about trueskill 2011-06-21T03:01:32 if you're referring to that 2011-06-21T03:01:41 mm, maybe i confused myself out of that. 2011-06-21T03:01:48 for a 1 v 1 game, Trueskill boils down to a Elo (slight maths differences, but no more than Chess and Go use variants) 2011-06-21T03:01:49 or maybe someone else said it. 2011-06-21T03:02:17 the scoring with spawn points might work for two player though 2011-06-21T03:02:20 it didn't without it 2011-06-21T03:02:33 Okay, so let's say true skill is accurate, and base on the assumption that each player do have a fixed consistent strength (the point i wanted to make sure through reading the article :p) 2011-06-21T03:02:38 i think it would make more sense now 2011-06-21T03:02:49 Then there is still the fact that we DO NEED a PAIRING POLICY 2011-06-21T03:03:00 either there already is one. Or we need to build one. 2011-06-21T03:03:20 stocha: because you don't want your bot playing random bots from the list, you only want to play the ones around you ? 2011-06-21T03:03:47 stocha: if so, that shouldnt be hard to do - but you won't get much variety in your games 2011-06-21T03:04:26 becomes more like boxing, you have to fight all the way up the ladder to get the right to fight the top bots 2011-06-21T03:05:28 there is already a matching algorithm 2011-06-21T03:05:39 tron had random pairing and it was awful 2011-06-21T03:06:14 what do you guys mean by pairing algorithm? like who will be against each other? 2011-06-21T03:06:19 -- pick the closest 100 available submissions (limited for speed) we just dont have more than 100 yet :-) 2011-06-21T03:06:27 Rabidus: yeah pretty much 2011-06-21T03:06:51 yes, it will only choose from bots within 100 places of your bot.. we need a few hundred more test bots to see it in action :-) 2011-06-21T03:06:58 in pw it did something like the following: pick the bot who has played the least number of games then match them against someone using a poisson distribution 2011-06-21T03:07:14 i don't see a problem. Why just player x doesn't take a opponent who is rank +- ~20 ? 2011-06-21T03:07:16 Zaphus: nah, it shouldn't restrict you to only bots near your ranking, but weight it 2011-06-21T03:07:53 for 1vs1 given enougth data, pairing policy won't influence too badly on ranking. obviously. only slower to converge. 2011-06-21T03:08:06 * +- ~20 of players rank 2011-06-21T03:08:08 antimatroid: do it on probability, the odds of playing someone decreases as the gap gets larger 2011-06-21T03:08:08 *** dr- has joined #aichallenge 2011-06-21T03:08:08 but with those multiplayer settings, i'm not confident about it converging somewhere. 2011-06-21T03:08:32 Rabidus: it needs to be on position/rank - not rating (I think thats what you are saying) - the rating value is arbitrary 2011-06-21T03:08:39 from the article, it seems that trueskill does have a pairing algorithm, or am i wrong ? 2011-06-21T03:08:47 yeah position. 2011-06-21T03:08:59 then if need more diversity, try 50% chance that the game is a close game , and 50 % it's a wider game. 2011-06-21T03:09:11 stocha: for the xbox implementation, yes it does - but trueskill is just math itself and doesnt have one 2011-06-21T03:09:18 i certainly can not find any close game at all in the current data. 2011-06-21T03:09:25 lets say that we have list x where is included all bots. we go each one thru and get random opponent set who has rank near yours. 2011-06-21T03:09:26 even 10% of close game would be good to me. 2011-06-21T03:09:35 when list has gone thru, do it again 2011-06-21T03:09:57 and before that 'do it again' we update the list 2011-06-21T03:10:08 very hard to see a problem 2011-06-21T03:10:16 Gives me at least 10% of games with all opponents closely matched, and i'll be quite happy :) 2011-06-21T03:10:30 close game: http://aichallengebeta.hypertriangle.com/visualizer.php?game=29371&user=78 2011-06-21T03:11:12 stocha: second to most recent game for first place player has 1st, 2nd, 5th and 19th ranked players? 2011-06-21T03:11:14 28 – tmwitt 2011-06-21T03:11:30 no pleasing some people :-) 2011-06-21T03:11:48 find me a game, with less that 3 rank, and i'll give credit that there exist one close game. 2011-06-21T03:12:00 but even it does exist one over 1000. It's not 10% for sure :) 2011-06-21T03:12:03 what does that mean ? less than 3 rank ? 2011-06-21T03:12:19 it means for example 1 vs 2 vs 5 vs 8 2011-06-21T03:12:34 maximim three rank of jumping between players :p 2011-06-21T03:12:45 2 vs 5 isn't less than 3 rank ;) 2011-06-21T03:13:00 (and 5 vs 8 isnt either) 2011-06-21T03:13:00 isn't it less or equal ? 2011-06-21T03:13:07 yes 2011-06-21T03:13:14 okay then :p 2011-06-21T03:13:16 1 vs 2011-06-21T03:13:24 1 vs 8 isn't 2011-06-21T03:13:28 but there exist a chain 2011-06-21T03:13:39 yeah i understood that 2011-06-21T03:13:47 so i say it's a close game. Maybe not the best definition 2011-06-21T03:13:51 1 vs 3 vs 5 vs 7 2011-06-21T03:13:52 but let's get along with this one. 2011-06-21T03:14:23 http://aichallengebeta.hypertriangle.com/visualizer.php?game=29676&user=126 i can has best expansion? 2011-06-21T03:14:26 so yes, that would be awesome. the player matching doesnt force that yet. put a bug in asking for it to do them occasionally 2011-06-21T03:14:48 let's call a close game, if there exist a chain of less than either 3 rank or 0.20 true skil points 2011-06-21T03:15:17 technically thats a well-matched game, it may or may not turn out to be close 2011-06-21T03:15:29 i would like that there was at least 10% of the game that match those criterions 2011-06-21T03:15:41 so add a bug 2011-06-21T03:15:41 yeah well-matched game then :) 2011-06-21T03:15:58 i think if this is on, i'd be far more happy. 2011-06-21T03:16:19 I doubt it, you'll just find something else to pick on :-) but it's a good idea (and essential for the finals!) 2011-06-21T03:16:24 (although you might want adjust the criterion at some point when we have more data) 2011-06-21T03:17:03 actually, if we had 5000 entries, which is 50-60x what we have now, then you'd accept a wider range for closeness I imagine 2011-06-21T03:17:50 maybe once per day, 1v2v3v4v5v6v7v8v9v10, 11v12v13v14v15v16v17v18v19v20 etc :) 2011-06-21T03:18:09 how much there was submissions in last competion ? 2011-06-21T03:18:11 1000 ? 2011-06-21T03:18:13 2000 ? 2011-06-21T03:18:15 4600 I think 2011-06-21T03:18:17 :OO 2011-06-21T03:18:21 wow 2011-06-21T03:18:22 *** mcstar has joined #aichallenge 2011-06-21T03:18:25 as in users 2011-06-21T03:18:39 i think that 5000 entries is near then 2011-06-21T03:18:43 lots of starter bots in there I suspect though 2011-06-21T03:18:49 yeah ofcourse 2011-06-21T03:19:00 *** stocha_ has joined #aichallenge 2011-06-21T03:19:02 for reference, we had under 1000 bots for tron 2011-06-21T03:19:06 i had very simple bot and lasted somewhere ... 500 ? 2011-06-21T03:19:08 so who knows how many bots ants will have 2011-06-21T03:19:22 ants looks very promising 2011-06-21T03:19:28 so - somewhere between 1000 and (1000*4*4 = 16,000) :-) 2011-06-21T03:19:33 :D 2011-06-21T03:19:37 imo ants is a less interesting problem to planet wars, but it's way more suited to ai 2011-06-21T03:19:39 nice guess 2011-06-21T03:19:46 in that there are a lot more obvious problems to tackle 2011-06-21T03:19:52 with many obvious approaches to them 2011-06-21T03:19:55 i didn't think planet wars as a problem 2011-06-21T03:19:56 (i dreamt about extending lisp in ways i wouldnt think of not dreaming) 2011-06-21T03:20:21 Gotta go - have a beach volleyball game to get to - cya later 2011-06-21T03:20:27 see ya 2011-06-21T03:20:33 bb 2011-06-21T03:20:34 but ants had multiple "simple" problems in first steps (pathfinding, fighting) 2011-06-21T03:20:38 *** stocha has quit IRC (Ping timeout: 252 seconds) 2011-06-21T03:20:50 yeah, food collection is the first thing for ants 2011-06-21T03:20:51 *** Zaphus has quit IRC (Quit: Page closed) 2011-06-21T03:20:56 i think if there exist wide gaps in trueskill give ranking possible chain, we should use the rank number, otherwise, we can just give a trueskill range for defining what closely related opponent are. 2011-06-21T03:21:26 there might be a problem that there are too many submissions.. 2011-06-21T03:21:56 too early to understand that sentence stocha 2011-06-21T03:21:58 for example if 1: 93.0 2 : 78.0 3 : 77.9 4 : 77. 8 5 : 60 2011-06-21T03:22:05 then 1 is close to 2 2011-06-21T03:22:11 but far from 3 2011-06-21T03:22:24 but 2 and 4 are close together. that sort of things 2011-06-21T03:22:50 3 is far for 5, but close to 4 2011-06-21T03:22:59 and 4 is close to 5 2011-06-21T03:23:17 so we say a game is "well matched", if there exist a chain of closeness 2011-06-21T03:23:44 i would say we should look at stability 2011-06-21T03:23:49 then we can have ever stronger "well matched" condition from time to time. Like two fartherst bot in, are close. 2011-06-21T03:23:51 how long a given order lasts 2011-06-21T03:23:51 not all games need to be between bots of close rank 2011-06-21T03:23:58 just on average pure random sucks ass 2011-06-21T03:24:07 i want 10 % of game closed ranked. 2011-06-21T03:24:16 stocha_: you know trueskill has a built in metric for how 'good' a specific pairing is? 2011-06-21T03:24:26 would it be stupid that in 8 players game if you win a game, you'll get 5 points, second gets 3, third gets 2 and fourth gets 1, all others 0 2011-06-21T03:24:27 (for my own satifaction. more might be okay too) 2011-06-21T03:24:36 you should go figure that out for the pairings and see how it is 2011-06-21T03:24:41 * antimatroid doesn't like 8+player games 2011-06-21T03:24:43 too many bots 2011-06-21T03:24:51 * Rabidus doesn't like either 2011-06-21T03:24:53 janzert: let's use that then ? i hope it's not activated now ? from what i see about current pairing 2011-06-21T03:24:54 yes 2011-06-21T03:25:02 i think 8 is too big already 2011-06-21T03:25:05 * mcstar likes chocolate though 2011-06-21T03:25:08 :p 2011-06-21T03:25:15 stocha_: it doesn't tell you how to pair, it tells you how good a specific pairing is 2011-06-21T03:25:17 janzert: what are your thoughts on number of bots? 2011-06-21T03:25:33 * janzert is fairly ambivalent about it 2011-06-21T03:25:52 janzert: okay so let's do it so, that 10% of the game are good pairings. And see how it feels then :p 2011-06-21T03:26:20 go start checking them and report back what trueskill says about them 2011-06-21T03:26:32 because it's multiplayer game, let's have equivalent number of game for 2 opponents up to 8. 2011-06-21T03:26:36 and see how it feels. 2011-06-21T03:26:42 we do need 1vs1 game added though. 2011-06-21T03:27:09 you're probably stuck there, cause antimatroid actively doesn't want them 2011-06-21T03:27:19 janzert: i don't know how to do that. I can say that about none match the criteria i gave though. 2011-06-21T03:27:26 what i dont get is: isnt this matching system self-balancing? why do you care at all? 2011-06-21T03:28:03 janzert: let's build an early poll and see how much antimatroid opinion match with most poeples feeling. What about the famous "decision making" team ? 2011-06-21T03:28:04 anyway I'm off 2011-06-21T03:28:36 that would be me, amstan and mcleopold and you haven't done much to sway me ;) 2011-06-21T03:29:26 i'm not as opposed to 2 player anymore with spawn points i don't think 2011-06-21T03:29:35 but i also haven't really thought too much about it since that happened 2011-06-21T03:29:41 anyway, stocha, wouldnt you be glad if the contest started already? 2011-06-21T03:29:55 let's at least try it during beta then. 2011-06-21T03:30:02 mcstar: no. 2011-06-21T03:30:10 it's ways too soon for starting it. 2011-06-21T03:30:16 although we could make publicity for the beta. 2011-06-21T03:30:22 idd 2011-06-21T03:30:27 *** sigh has quit IRC (Quit: This computer has gone to sleep) 2011-06-21T03:30:29 then tcp servers would get running too 2011-06-21T03:30:35 the beta is very important, because you can change the rules, and tweak things, and try stuff. 2011-06-21T03:30:58 tcp servers are important. Well one well built one is important. 2011-06-21T03:31:10 although i can't take profit out of them right now :p 2011-06-21T03:31:16 mcleopold said, i beleive, that if they start early the contest, they could make changes nevertheless 2011-06-21T03:31:19 because i have no hardware to run my bots. 2011-06-21T03:31:42 i can spare my network and my server for tcp server 2011-06-21T03:31:44 making change during the contest, i strongly dissupport :p 2011-06-21T03:31:49 but tcp server software is buggy atm. 2011-06-21T03:32:01 mcstar: changes now are unideal 2011-06-21T03:32:07 changes after launch are even more unideal 2011-06-21T03:32:13 but if necessary yes they will be made 2011-06-21T03:32:17 all should be carved in stone, then people make their bot, they are ranked, and one winner is ellected somehow. 2011-06-21T03:32:48 well, its not too likely, but something can still come up after the rules are fixed.... 2011-06-21T03:33:07 we semi agreed to fix most of the rules this morning 2011-06-21T03:33:18 Rabidus: i would appreciate that there exists an official way of setting up a tcp server that approximate "real" conditions though. 2011-06-21T03:33:24 what if there is a simple startegy to the game, and you couldnt do better if you were god himslef? 2011-06-21T03:33:39 what if a 1000 bots would implement that startegy? 2011-06-21T03:33:43 mcstar: if that happened really early in the contest then that would be an example of the game breaking 2011-06-21T03:33:44 there is tcp server on github that is ready to run 2011-06-21T03:33:47 mcstar: that's what the beta is for. finding stuff. If they are found after beta, they live with them. 2011-06-21T03:33:50 if that happens well into the contest, good for them 2011-06-21T03:34:11 mcstar: then the challenge is a draw. and we wait for next one. 2011-06-21T03:34:18 although where one would draw the line is anyones guess as it's never happened and i don't expect it to 2011-06-21T03:35:28 up from now there was always a clear winner. I have no reason yet to think that ant will be different. it might. it always might. 2011-06-21T03:35:53 yeah, prev. experience doesnt really help there 2011-06-21T03:36:14 you can never be entirely certain about some of these things without actually running it and seeing what happens 2011-06-21T03:36:19 lets hope for the best :D 2011-06-21T03:36:20 at worse, there will be like 5 very strong bot, very closely matched. And then we have a problem. 2011-06-21T03:36:27 but i doubt it could be worse than that 2011-06-21T03:36:34 one problem we saw with planet wars maps in the final contest was that a lot sucked 2011-06-21T03:36:41 even with blocking etc. 2011-06-21T03:36:44 despite our efforts to eliminate "trivially crap maps" from being generated 2011-06-21T03:36:53 someone will still hack a way to get ahead. 2011-06-21T03:37:06 but once those are generated there's no turning back 2011-06-21T03:37:10 antimatroid: was a new map generated for every new game? 2011-06-21T03:37:15 i don't think crap map should be eliminated either. But that's another story. 2011-06-21T03:37:15 no 2011-06-21T03:37:28 i am in favour of that though, there are reasonable arguments for or against the idea 2011-06-21T03:37:56 stocha_: provided people know those types of maps wont show up, i don't mind 2011-06-21T03:38:04 i do mind if we go through and select which maps not to use 2011-06-21T03:38:08 but as far as beta goes, i'm happy with the look of the map. Although i'd prefer random generated ones. 2011-06-21T03:38:27 they are random 2011-06-21T03:38:37 and i'd like for beta at least to try the assymetric one. 2011-06-21T03:38:38 once i finish exams i am going to play with maps 2011-06-21T03:38:43 (I have my last exam thursday) 2011-06-21T03:38:55 i thought they tried asymmetric and decided they sucked? 2011-06-21T03:38:57 they are not random in the sense that the same map can happen in more than one game. 2011-06-21T03:39:28 i tryed assymetric. And decided they where okay to find out wich bot is strongest. 2011-06-21T03:39:30 stocha_: try running your own bot on assym. maps see whats ur conclusion 2011-06-21T03:40:04 mcstar: i already done that. I was always able to tell what bot was strongest. Even very closely related one. 2011-06-21T03:40:21 different versions of your own bot? 2011-06-21T03:40:29 different bots yes. 2011-06-21T03:40:31 or same bots. 2011-06-21T03:40:36 always consistent 2011-06-21T03:41:07 i don't know if it would converge much slower though. But it will converge. 2011-06-21T03:41:10 i thought of an awesome extension to my food collection idea last night 2011-06-21T03:41:12 yeah, well, if a bot expands very fast( i mean theres no error in that part) than it can compensate for an asymmetry i guess 2011-06-21T03:41:45 mcstar: i was able to distinguish bot very close in strength 2011-06-21T03:42:00 BUT .. bot strength can depend on map distribution. And food spawning. 2011-06-21T03:42:20 so the strength is for a given distribution of maps building process, and food distribution . 2011-06-21T03:42:42 yeah, well than wtf are we talking about? 2011-06-21T03:42:53 it's not the same :) 2011-06-21T03:43:18 because you can have both symetric an assymetric maps, using closely related map shape and food distribution 2011-06-21T03:43:21 im sorry i must have misunderstood you, i thought that regardless of asymmetries, you could distunguish your bots as far as "strength" is concerned 2011-06-21T03:43:34 yeah, that's whay i claim. 2011-06-21T03:43:46 however the shape , and speed of food spawning affect bots. 2011-06-21T03:43:52 but now you say that: strength is for a given distribution of maps building process, and food distribution 2011-06-21T03:44:14 yeah, but it doesn't depend on it being symetric/vs assymetric 2011-06-21T03:44:18 in depends on shape 2011-06-21T03:44:22 obviously for an asymmetric map, food spawing cant be symmetric 2011-06-21T03:44:41 only statistical homogenity can be achieved 2011-06-21T03:45:02 for food spawning, it was more about the speed of it. 2011-06-21T03:45:18 does one food spawn every turn ? does they spawn more than every turn ? etc. 2011-06-21T03:45:30 then i didn't try twisted spawning strategyes 2011-06-21T03:45:38 like : spawn only at starting positions 2011-06-21T03:45:55 but even food speed can make or break bots. 2011-06-21T03:46:13 bots very well suited for high density food, might do poorly with low density 2011-06-21T03:46:22 high spawning rate rather 2011-06-21T03:46:29 then you have the time management also 2011-06-21T03:46:44 high food rate, often mean high ant count. 2011-06-21T03:47:03 its players problem 2011-06-21T03:47:06 then the map shape : open maps are very different from tiny corridors labyrinth maps 2011-06-21T03:47:23 stocha_: how robust is path finding? 2011-06-21T03:47:25 that's also player problem 2011-06-21T03:47:52 although in both case assymetric/vs symetric would give same bot strength, the probability of a map being wholy oppen between labyrinth corridor will change the respective strength of the bots. 2011-06-21T03:48:01 i.e. are there corner cases when it breaks down? 2011-06-21T03:48:28 *** locutus2 has joined #aichallenge 2011-06-21T03:48:33 sorry, i mean Your path finding 2011-06-21T03:48:39 mcstar: path finding is perfect path finding. You probably want to know what one i use ? i use perfect path finding one, with no congestion control. 2011-06-21T03:49:01 although i had some bot not using any pathfinding also. 2011-06-21T03:49:19 ok, just because i dont want to implement bfs and a* or any graph walking bitch 2011-06-21T03:49:39 especially in very oponned map, pathfinding might not add to much to bot strength; (nothing indeed in fact) 2011-06-21T03:49:44 right now im thinking about conves hulls 2011-06-21T03:49:46 convex 2011-06-21T03:50:26 i went with floodfill. it works well. it's perfect in the sense it will alway find the best path. 2011-06-21T03:50:38 it's easy to debug etc. 2011-06-21T03:50:48 i didn't have any performances problem so far. 2011-06-21T03:51:16 what's convex hulls ? 2011-06-21T03:51:34 *** aerique has joined #aichallenge 2011-06-21T03:52:35 well, given a set of points in any dimensin ther exists a uniqe polyhedron which contains all those points, and has the property of least volume and surface 2011-06-21T03:53:07 a line segment or a triangle is always a convex hull :D 2011-06-21T03:53:30 *** stocha has joined #aichallenge 2011-06-21T03:54:04 so, if the ant and the target are both part of the convex hull, i can calculate it with the points of water included, and i can simply travel one side of the convex hull 2011-06-21T03:54:07 probably good for very big maps. 2011-06-21T03:54:17 but there will not be any very big maps 2011-06-21T03:54:20 and it will take me to the target in optimal turns 2011-06-21T03:55:06 with traffic control ? 2011-06-21T03:55:28 *** stocha_ has quit IRC (Ping timeout: 252 seconds) 2011-06-21T03:55:49 what do you mean? mutltiple ants travelling simultaenously? 2011-06-21T03:56:10 i havent given much thought into that 2011-06-21T03:59:49 my a* will crawl if there was LOTS of food i think 2011-06-21T04:00:01 although maybe not too bad 2011-06-21T04:00:17 i need to iterate through a set of target locations for every square i add to the search queue 2011-06-21T04:01:31 mcstar: were you writing a cl ants game engine? 2011-06-21T04:01:41 yep 2011-06-21T04:01:42 *** stocha has quit IRC (Ping timeout: 252 seconds) 2011-06-21T04:02:00 mcstar: is the code shared somewhere? :) 2011-06-21T04:02:18 no, sadly its proprietary 2011-06-21T04:02:30 you would like to see it? 2011-06-21T04:02:46