2011-09-17T00:00:31 *** dvladim has joined #aichallenge 2011-09-17T00:26:50 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (Ping timeout: 260 seconds) 2011-09-17T00:30:08 *** eashoka has quit IRC (Ping timeout: 252 seconds) 2011-09-17T00:38:39 *** csr20 has joined #aichallenge 2011-09-17T00:40:29 *** csr20 has left #aichallenge 2011-09-17T00:42:15 aichallenge: McLeopold epsilon * r9b725f4 / website/ranking.php : fix function name mismatch - http://git.io/R90PAw 2011-09-17T00:43:15 *** ChrisH has quit IRC (Quit: Page closed) 2011-09-17T00:58:52 aichallenge: McLeopold epsilon * rc81b0bb / website/sql.php : fix missing value in insert statement - http://git.io/yiY9CQ 2011-09-17T01:01:00 @later tell janzert I forgot to commit the sql to match the api_game_result, oops, workers are posting now 2011-09-17T01:01:00 McLeopold: Yep. 2011-09-17T01:20:24 I need a math guy 2011-09-17T01:20:32 jmcarthur: ? 2011-09-17T01:20:43 delt0r_: ? 2011-09-17T01:39:38 not sure if i count as a math guy, but shoot 2011-09-17T01:39:43 McLeopold: ^^ 2011-09-17T01:40:06 wow it's later than i thought 2011-09-17T01:40:09 my time is limited 2011-09-17T01:40:20 *** amstan has quit IRC (Ping timeout: 260 seconds) 2011-09-17T01:45:58 *** dvladim has quit IRC (Ping timeout: 260 seconds) 2011-09-17T01:57:03 jmcarthur: I need a way to prevent trueskill from favoring players 2011-09-17T02:00:55 what do you mean by 'favoring' ? Oo 2011-09-17T02:01:46 well, if you have a seed player that can play two other players, trueskill match quality will favor the player with the lower sigma 2011-09-17T02:01:57 if there mu's are the same 2011-09-17T02:02:10 after they play, they get updated and the sigmas drop more 2011-09-17T02:02:39 so a player that gets a few games in will have low sigma and suddenly be picked more for matches than others 2011-09-17T02:02:56 just look at the current rankings to see the issue 2011-09-17T02:11:10 I've seen the effect, but I had no idea it was because of equal mu's 2011-09-17T02:14:12 Morning 2011-09-17T02:15:49 McLeopold: I am a math guy and i read the TrueSkill paper... but i know of no easy way to fix such things. Basically its the model 2011-09-17T02:15:53 more or less 2011-09-17T02:16:39 Personally i think there is a need for a better model --or at least better inference of "true" parameters from games played 2011-09-17T02:17:09 I have pondered this and came up with only a ML idea... that could work --perhaps 2011-09-17T02:22:50 delt0r_: I figure we need to make sure each user has about the same number of games in a 24 hour period, so there needs to be some sort of skew to the match quality to pick opponents that may not be the better match 2011-09-17T02:23:21 it is a hard problem 2011-09-17T02:23:31 yes 2011-09-17T02:23:36 esp when we consider new submissions 2011-09-17T02:23:43 yep 2011-09-17T02:23:48 thus the 24 hour period 2011-09-17T02:24:07 new submissions would get some preference at the start 2011-09-17T02:24:10 i would probably try to have something that works IIF there are no new submissions 2011-09-17T02:25:11 Of course the even harder problem is getting it so others don't complain :D 2011-09-17T02:25:32 I usually don't care about that, but right now it isn't good 2011-09-17T02:25:52 But if we have the compute power for a lot of games --it should not be to hard right? 2011-09-17T02:26:02 * delt0r_ looks at how it works now 2011-09-17T02:26:40 the more games at once, the smaller the problem becomes 2011-09-17T02:26:56 because I restrict matches to those not in a game 2011-09-17T02:27:08 good idea 2011-09-17T02:27:30 so if everyone is playing at once, then roughly everyone gets one game at the same rate 2011-09-17T02:27:37 also what works best for true skill? 2 player, or n player games 2011-09-17T02:27:42 but then player rotation would be poor 2011-09-17T02:27:58 it depends 2011-09-17T02:28:44 generally more players without ties cause bigger updates 2011-09-17T02:29:00 mmm 2011-09-17T02:29:17 what is the word on going live? 2011-09-17T02:29:28 next week 2011-09-17T02:29:39 but I got to fix this first 2011-09-17T02:30:05 right --is the php trueskill thing working 2011-09-17T02:30:22 so, if you have a choice of playing 2 players with match qualities 50% and 51%, right now I always pick 51% 2011-09-17T02:30:42 delt0r_: yes, if draw prob is 1.75 or higher it makes a world of difference 2011-09-17T02:30:58 10 players with only 3 ranks is the worst 2011-09-17T02:31:11 oh good. that would work with the jskills one too 2011-09-17T02:31:19 yeah, same issue 2011-09-17T02:31:21 should be higher than 1% anyway IMO 2011-09-17T02:31:31 convergence is much faster with a higher draw prob 2011-09-17T02:31:44 i think it will match the data better too 2011-09-17T02:32:02 well, because the scores are based on minute differences, I figured it would be close to zero 2011-09-17T02:32:54 the higher the draw prob, the bigger the update for a win, and I was thinking the game is to noisy for that 2011-09-17T02:33:53 I think it would be noise regardless... since it only considers rank--and esp at the start when most player scores are random anyway 2011-09-17T02:34:08 noisy 2011-09-17T02:34:25 yea, so back to my example, pick the player with 50% or 51%? 2011-09-17T02:34:30 do we keep track of who played who? 2011-09-17T02:34:40 yes 2011-09-17T02:34:45 there is some no repeat logic 2011-09-17T02:34:51 cool 2011-09-17T02:35:15 well i am i guy who does a lot of stochastic simulations 2011-09-17T02:35:29 yeah, so you can help me with this matchmaking then :) 2011-09-17T02:35:34 so i would pick the 50% with some p or 51% otherwise 2011-09-17T02:35:41 right 2011-09-17T02:35:50 if the average bot has 24 games a day 2011-09-17T02:36:09 and the 50% guy has only had 20, and the 51% guy has had 22? 2011-09-17T02:36:15 the idea is that in the limit all possible games get played... ie after 1 million or so years 2011-09-17T02:36:23 ah 2011-09-17T02:36:30 k let me work it out 2011-09-17T02:36:46 its exponetail 2011-09-17T02:36:53 dam my bad english 2011-09-17T02:37:02 what is your language? austrian? 2011-09-17T02:37:31 no --english 2011-09-17T02:37:34 :D 2011-09-17T02:37:54 can order beer in german and food --that is about it 2011-09-17T02:38:04 oh, I was guessing from the .as uri :) 2011-09-17T02:38:39 once we get going, most bots will have 20-30 other bots with a match quality > 90% 2011-09-17T02:39:19 the top 20 all have low sigma, the higher sigma bots dont appear until around 20-30 down on the list 2011-09-17T02:39:25 even though the mu is closer 2011-09-17T02:39:55 so I need to skew the match quality to be higher for bots that haven't had many games 2011-09-17T02:39:56 So if we have n games per day with m players --that gives nm slots per day 2011-09-17T02:40:13 it would be natural to increase sigma, since that represents uncertainty 2011-09-17T02:40:38 oh... you want it not to be memory less... curse my math 2011-09-17T02:40:49 *** dvladim has joined #aichallenge 2011-09-17T02:41:01 we like ideal universes 2011-09-17T02:41:37 McLeopold: doesn't the trueskill think estimate the sigma? 2011-09-17T02:42:03 no, mu and sigma are stored in the db for every submission 2011-09-17T02:42:09 s/think/thing 2011-09-17T02:42:29 ok so now i am confused 2011-09-17T02:42:32 http://pastebin.com/N68fAUbM 2011-09-17T02:42:36 * delt0r_ reads the paper again 2011-09-17T02:43:35 you may need to increase the width to read that table correctly 2011-09-17T02:44:53 delt0r_: or view it raw 2011-09-17T02:45:01 yea... big ascii table person myself... xl is for ms slaves ;) 2011-09-17T02:45:16 google docs? 2011-09-17T02:45:32 but the mu and sigma scores are the current ranks etc correct? 2011-09-17T02:45:38 yes 2011-09-17T02:45:43 yep--i now use google docs a lot 2011-09-17T02:46:04 so on that table, see how the low game_count submissions are down the list? 2011-09-17T02:46:30 there is a mu_diff of 0.16, which is the closest mu way down there 2011-09-17T02:47:52 i c 2011-09-17T02:47:53 I'm thinking scale the sigma by the number of games in the last 24 hours somehow 2011-09-17T02:48:28 I really don't follow... should not the trueskills thing take that into account properly... 2011-09-17T02:48:47 this is a table of match quality 2011-09-17T02:49:12 if everyone is always available to play, it will set you up with the same people over and over 2011-09-17T02:49:32 yea... even on the icc that is a problem 2011-09-17T02:49:42 icc? 2011-09-17T02:49:51 internet chess club 2011-09-17T02:49:53 oh 2011-09-17T02:50:03 aka where elo got famous 2011-09-17T02:50:21 but they use glicko now? 2011-09-17T02:50:54 which is still similar --and you still have the match like in "skill" keep matching the same people 2011-09-17T02:51:07 yes 2011-09-17T02:51:31 I don't care so much about playing the same people, it's just ignoring some people too much 2011-09-17T02:51:32 last i heard it was still a problem if you wanted variety 2011-09-17T02:51:54 what about like last year with the que system? 2011-09-17T02:52:23 oh.. then you get very unlike in skill 2011-09-17T02:52:25 well, I wanted each match to calc live 2011-09-17T02:52:40 so you still need to sometimes pick folks that have had many games 2011-09-17T02:52:45 so new submissions find their rank really fast 2011-09-17T02:53:04 I just don't think that is generally possible 2011-09-17T02:53:15 yeah, so your idea about probabilites might work somehow 2011-09-17T02:53:39 a hack could just be a weighting 2011-09-17T02:54:01 the 51% is always picked if games played are the same 2011-09-17T02:54:02 1 / recent games 2011-09-17T02:54:12 mm yea 2011-09-17T02:54:42 the think is, I'm doing a sql statement with an order by and a limit 1 2011-09-17T02:55:24 hmm... 2011-09-17T02:55:47 I could limit the field to the top 10% in match quality, then choose from those the ones with the least games? 2011-09-17T02:55:50 not sure tbh --i would probably just use some selected to get a range of possibles then use some biased random selection 2011-09-17T02:56:07 yea something like that... 2011-09-17T02:56:10 we just said the same thing :D 2011-09-17T02:57:15 getting a percentage is a pita in sql :( 2011-09-17T02:57:43 I would selected a range... and deal with different sized lists 2011-09-17T02:58:26 I don't follow you 2011-09-17T02:58:42 what would be best is something from trueskill that would tell use what games would have the most "information" 2011-09-17T02:58:58 that is match quality, they are the same thing 2011-09-17T02:59:05 so we have that 2011-09-17T02:59:06 selected players with mu> and mu< 2011-09-17T02:59:16 in advanced? 2011-09-17T02:59:20 yes 2011-09-17T02:59:34 well, each match created calculates it all right there 2011-09-17T02:59:38 I really should read the paper again... it seems i missed some details 2011-09-17T03:00:02 the higher the match quality, the larger the updates will be, and the faster convergence to the correct mu is 2011-09-17T03:00:04 before you play it? perhaps that may be a good way to choose... if it is a fast calc 2011-09-17T03:00:36 it is, it's just we can use quality alone, because once you get a low sigma, you just start snowballing into more games 2011-09-17T03:00:56 I think the top x% is the way to go 2011-09-17T03:01:05 just printed the wrong true skill paper 2011-09-17T03:01:12 print!? 2011-09-17T03:01:41 i prefer to read from papers from paper 2011-09-17T03:01:43 *** dvladim has quit IRC (Ping timeout: 260 seconds) 2011-09-17T03:01:52 even have a laser printer at home 2011-09-17T03:02:03 http://www.moserware.com/2010/03/computing-your-skill.html 2011-09-17T03:02:05 http://dl.dropbox.com/u/1083108/Moserware/Skill/The%20Math%20Behind%20TrueSkill.pdf 2011-09-17T03:03:53 no the orginal paper 2011-09-17T03:04:03 that one is to fluffy 2011-09-17T03:04:29 :) 2011-09-17T03:04:35 but you could have a game q score that includes the number of games played. 2011-09-17T03:04:43 I don't understand the A table notation 2011-09-17T03:05:37 a table? 2011-09-17T03:06:08 page 34 of the fluffy paper 2011-09-17T03:06:12 A matrix 2011-09-17T03:06:58 if you could provide me with a spread sheet that calcs match quality for more than 2 players, that would be awesome 2011-09-17T03:07:22 I have to use an approximation of the 2 player simplified version 2011-09-17T03:08:09 antimatroid is always throwing math notation at me, and I'm just too much out of practice 2011-09-17T03:08:37 which maths now? :P 2011-09-17T03:08:58 see what I mean, he's correcting "math" to "maths" 2011-09-17T03:09:19 meh, that's a difference in language :P 2011-09-17T03:09:27 antimatroid: let's see an integral in unicode, go! 2011-09-17T03:09:51 errr? :P 2011-09-17T03:10:04 \int e^x.dx 2011-09-17T03:10:12 LaTeX? 2011-09-17T03:10:15 sex .dx :P 2011-09-17T03:10:58 antimatroid: look at the rankings right now and notice the game count, it is awful 2011-09-17T03:11:03 @rankings 2011-09-17T03:11:04 McLeopold: An error has occurred and has been logged. Please contact this bot's administrator for more information. 2011-09-17T03:11:10 @beta 2011-09-17T03:11:10 antimatroid: beta could be http://aichallengebeta.hypertriangle.com/. 2011-09-17T03:11:38 so you restarted it? 2011-09-17T03:11:44 my bots rank has varied wildly 2011-09-17T03:12:07 kinda, we reset ranks and started counting games, like what we would do for the finals 2011-09-17T03:12:19 the game counts are not even 2011-09-17T03:12:41 what's the average number of game turns? 2011-09-17T03:12:51 my bots profile seems to have games finish within 300 turns 2011-09-17T03:13:08 I lowered the max to 255, so I could get a lot of games in 2011-09-17T03:13:19 I'm testing rankings right now, not game play 2011-09-17T03:13:27 planet wars was 1 second per turn with max 200, so n players will reduce the number of games but you get more players in each game 2011-09-17T03:13:28 How about choose a large number of possible games, then rank them according to "quality" and game count.. 2011-09-17T03:13:43 also, the beta server isn't very good yeah? we don't have ec2 servers up and running yet? 2011-09-17T03:13:48 then pop the first game off the sorted list, rinse and repeat 2011-09-17T03:14:19 delt0r_: that's kinda what I do, but I add 1 player at a time until I reach the player count 2011-09-17T03:14:25 oh, you mean the variation in game counts is awful? 2011-09-17T03:14:34 what is the performance of ec2 like... we have gone off virtilized stuff--it is about 2-3 slower than without it 2011-09-17T03:14:50 delt0r_: it's the "rank by .. and count" that I need help with 2011-09-17T03:15:13 you'' have to ask janzert about that 2011-09-17T03:15:25 McLeopold: yea, but i am talking about games here not players... so you can have 10000 possible games that you then sort 2011-09-17T03:15:33 i think what janzert had for selecting players last time was good, just pick the first player in the game as the bot with the least number of games played 2011-09-17T03:15:45 although that might be an issue with people reuploading? hmmmm 2011-09-17T03:16:10 bot with the lowest average number of games per hour/day? 2011-09-17T03:16:10 delt0r_: I just need some outer joins to get that 2011-09-17T03:16:17 since it was submitted 2011-09-17T03:16:20 antimatroid: that does not work quite as well at it looks when one considers that you want to play liked skill folks that may all have played a lot 2011-09-17T03:16:31 Oh.. db.. 2011-09-17T03:16:45 yes, you select the first player in the game, then select the rest of the players weighted towards that players ranking somehow? 2011-09-17T03:16:59 yes, the seed player is the last to have played, that is working, it's the opponent matching that is off 2011-09-17T03:17:13 *** dvladim has joined #aichallenge 2011-09-17T03:17:24 Well for one project we just loaded the whole table into ram, did what we wanted .. then sent it all back.. wrapped in a transaction. 2011-09-17T03:17:49 I've got match quality in an sql statment right now 2011-09-17T03:17:55 we have a 2 day transaction! 2011-09-17T03:18:09 had 2011-09-17T03:18:54 even with tons of possible games, I'm still going to get a cluster of opponents with low sigma, and many games 2011-09-17T03:19:15 well if you generate a lot of possible games, then first sort on what will bring the player count to more even... 2011-09-17T03:19:39 I've got to balance that with the quality though. 2011-09-17T03:19:40 well will say 1000000 2011-09-17T03:19:43 with 2011-09-17T03:20:00 the orginal games are still picked with that criteria 2011-09-17T03:20:13 I *know* how to select the player with the least games. I *know* how to select the best match quality. I just need to combine the 2 intellegently. 2011-09-17T03:20:28 the idea is that there is some games that are good and help with games played, its just finding them that are hard 2011-09-17T03:20:45 its possible that such a game does not exist of course 2011-09-17T03:21:06 yes, I follow, but how do we decide which is more important? 2011-09-17T03:21:24 complaints weighting ? 2011-09-17T03:21:33 right, what's the weighting? 2011-09-17T03:22:22 I've got 2 math guys here. Can I get a formula? 2011-09-17T03:23:02 It was a joke... but not really... there is no closed form thing here that would work off the top of my head 2011-09-17T03:23:12 ad hoc all the way 2011-09-17T03:24:10 :'( 2011-09-17T03:24:42 Note i didn't want to pick the player with the least games... just generate a set based on normal criteria.. then sort on fixing played games.. if the set is large enough --i think that should converge 2011-09-17T03:24:54 okay, I'm going to go with a percenticle of the best match quality combos and choose the one with the least played games 2011-09-17T03:25:02 normal criteria means ignoring played games/ 2011-09-17T03:25:26 its as good as i can come up with right now... 2011-09-17T03:25:41 so, how about the spreadsheet? 2011-09-17T03:25:51 ? 2011-09-17T03:26:03 for calculating match quality for more than 2 players 2011-09-17T03:26:20 oh --that is what you mean by table A 2011-09-17T03:26:30 A matrix 2011-09-17T03:26:32 no 2011-09-17T03:26:38 page 34, fluffy paper 2011-09-17T03:26:52 titled "The 'A' Matrix" 2011-09-17T03:26:54 I though the trueskill implementations would do it for you? no 2011-09-17T03:27:04 uh.. 2011-09-17T03:27:28 speedsheet ? calculating match q 2011-09-17T03:27:39 not following .. 2011-09-17T03:27:53 * delt0r_ is trying to read the paper at the same time 2011-09-17T03:27:54 so I can look at the formulas and play with them 2011-09-17T03:28:16 http://research.microsoft.com/en-us/projects/trueskill/twoteamtrueskill.xls 2011-09-17T03:28:27 they just didn't include the match quality 2011-09-17T03:29:04 we have only one player per team, so this makes it a bit easier 2011-09-17T03:29:22 yes, without partial play 2011-09-17T03:29:43 the implementation does give me the quality 2011-09-17T03:30:53 but I'm doing it in sql 2011-09-17T03:31:57 oh fun... sql for the win (not!) 2011-09-17T03:32:05 :( 2011-09-17T03:32:25 *** greghaynes has left #aichallenge ("WeeChat 0.3.4") 2011-09-17T03:37:07 *** berak has joined #aichallenge 2011-09-17T03:40:55 berak: is a tcpserver running? 2011-09-17T03:41:48 delt0r_: do you think you could generate a 10 player game in 4 seconds with the 10000 combos method? 2011-09-17T03:42:22 yea.. but with 10 players i don't 10000 is enough 2011-09-17T03:43:10 i don't think 2011-09-17T03:43:34 ~2.5 for the root 2011-09-17T03:43:37 i would expect python to handle that fine too? no? 2011-09-17T03:44:00 it's getting the data from the database to the app that I think would be slow 2011-09-17T03:44:06 McLeopold: ready to run, but not hosted anywhere 2011-09-17T03:44:20 I know why i didn't know about game q scores... the orginal paper does not have them... its a later refinement 2011-09-17T03:44:33 game q? 2011-09-17T03:44:35 McLeopold: oh 2011-09-17T03:44:39 game quality 2011-09-17T03:44:50 ok 2011-09-17T03:46:07 it uses belief propagation --but is more approximate than i thought... 2011-09-17T03:46:26 true draws would need a *lot* of iterations 2011-09-17T03:47:40 And xbox thing is biased towards ensuring the top 3 sigma players are ranked well and not to give much of a crap about the rest :/ 2011-09-17T03:48:27 hm 2011-09-17T03:48:57 again orginal paper 2011-09-17T03:49:15 also for live games players choose opponents. 2011-09-17T03:49:32 in some games 2011-09-17T03:49:56 Oh yea... some leaderboards you don't get too... thats right 2011-09-17T03:50:01 been a while 2011-09-17T03:51:10 match quality is a measure of draw probability 2011-09-17T03:52:17 also note that if i have a 3 player game with A>B>C 2011-09-17T03:52:56 the possibility of A kicking C arse before even competing with B is not taken into account 2011-09-17T03:53:49 you mean b kick c before a? 2011-09-17T03:55:11 either or 2011-09-17T03:55:22 winning is harder in the cases even for A 2011-09-17T03:55:31 yet this is not really taken into account 2011-09-17T03:55:50 perhaps extensions do... 2011-09-17T03:55:57 * delt0r_ looks up more papers 2011-09-17T03:56:03 I've got a percentile sql solution, I'm going to try it out 2011-09-17T03:56:34 no, trueskill and match quality only take initial skills and a ranking 2011-09-17T03:59:33 *** dvladim has quit IRC (Ping timeout: 260 seconds) 2011-09-17T04:00:26 do you keep all game result data? (i think so but it never hurts to ask) 2011-09-17T04:00:32 yes 2011-09-17T04:04:10 so for shits and giggles I could for example come up with *awesomeness* score and see how that would rank everyone. ;) 2011-09-17T04:04:29 we need that for the splash game :) 2011-09-17T04:04:35 the game on the front page 2011-09-17T04:07:21 thinking of rank statistics only... ie there is no score but only rank 2011-09-17T04:07:45 we could store a separate mu and sigma for that in the database 2011-09-17T04:08:06 you mean survival order? 2011-09-17T04:10:32 well what is the ranking that give A>B>C with the minimum numbers of C>A sort of thing 2011-09-17T04:11:16 my thinking is that we only get ranks out of games... 2011-09-17T04:11:37 this is rocket science to mee 2011-09-17T04:11:50 even if A beats everyone. It is not really data on how much A beats everyone by 2011-09-17T04:12:22 that would be a whole new algorithm 2011-09-17T04:12:45 with true skill, you just pair them with better people for the next game and see if they continue to climb 2011-09-17T04:13:15 so their sigma stays fairly large until the lose 2011-09-17T04:13:18 you still have a total rank, ie there are better people and much better people 2011-09-17T04:30:13 *** nux67 has joined #aichallenge 2011-09-17T04:38:51 *** mcstar has joined #aichallenge 2011-09-17T05:07:35 *** dvladim has joined #aichallenge 2011-09-17T05:13:50 *** eashoka has joined #aichallenge 2011-09-17T05:16:29 *** arlequin has joined #aichallenge 2011-09-17T05:16:41 hi 2011-09-17T05:17:40 I have a problem signing in to aichallenge, I registrered a week ago, but cannot receive the confirmation mail 2011-09-17T05:18:02 I checkec my spam box but nothing 2011-09-17T05:18:15 Do someone know what I should dod ? 2011-09-17T05:18:22 *do 2011-09-17T05:24:43 *** mceier has joined #aichallenge 2011-09-17T05:30:38 wait 2011-09-17T05:33:30 *** onensora has joined #aichallenge 2011-09-17T05:36:46 arlequin: I think this beta version does not send emails yet, even though it says so on the page. u can just login 2011-09-17T05:38:20 at least that was how it was when I created my account. however, to log in I think you need the link given at the end of the results page you got after the registration. If you have lost that, I guess the best bet is for you to create a new account. 2011-09-17T05:38:34 even email registration doesn't stop spam 2011-09-17T05:39:12 they told me before that they send emails, at least at one time it was turned on 2011-09-17T05:39:20 but still arlequin just wait 2011-09-17T05:39:54 eashoka: thank you for your answer. Some friend did receive a mail. I guess I will create a new one. 2011-09-17T05:40:11 mcstar: doesn't work that well :p 2011-09-17T05:45:31 Ok, for the information: I registered with another mail accout, and I received the mail. So they still send mail. 2011-09-17T05:51:21 *** FireFly has joined #aichallenge 2011-09-17T06:18:22 *** dvladim has quit IRC (Ping timeout: 252 seconds) 2011-09-17T06:19:19 *** onensora has quit IRC () 2011-09-17T06:48:23 did some math of the basic ranking problem that true skill attempts to solve. 2011-09-17T06:49:13 I have come to the conclusion that true skill is a very good compromise on modeling and tractability. Dont think it can be done much better personall 2011-09-17T06:49:19 personaly 2011-09-17T06:49:23 whatever 2011-09-17T06:49:57 However imperfect matchmaking is not something i was considering 2011-09-17T06:51:07 delt0r_: give it some time, i think that "model" is a distillation of a couple decades' work 2011-09-17T06:58:40 i think rich hickey is abusing the big-O notation 2011-09-17T07:00:43 ? 2011-09-17T07:03:23 im just saying maybe you can come up with a better ranking alg. after a couple of years studying the problem 2011-09-17T07:06:26 haha, ok im watching his video of clojure data structures, and basically it shows a previous version of the clojure site, there was the error, apparently he fixed it since then 2011-09-17T07:14:08 *** mathis_ has joined #aichallenge 2011-09-17T07:24:14 *** delt0r_ has quit IRC (Ping timeout: 276 seconds) 2011-09-17T07:36:08 *** delt0r_ has joined #aichallenge 2011-09-17T08:10:42 hmm workers are running super slow today... 2011-09-17T08:11:07 Somewhat off topic: In game programming (online multiplayer games), which protocol is used on the application layer? http? 2011-09-17T08:13:01 lol 2011-09-17T08:13:06 *** dvladim has joined #aichallenge 2011-09-17T08:13:08 many 2011-09-17T08:13:16 most common? 2011-09-17T08:13:40 "online multiplayer game" is too hazy 2011-09-17T08:13:47 that can be many things 2011-09-17T08:13:54 counter-strike for example 2011-09-17T08:14:02 foRei: TCP and UDP custom 2011-09-17T08:14:07 WoW is TCP 2011-09-17T08:14:13 more FPS are UDP 2011-09-17T08:14:17 most 2011-09-17T08:14:20 tcp and udp are transport layer right? 2011-09-17T08:14:25 yea 2011-09-17T08:14:27 well 2011-09-17T08:14:29 wait 2011-09-17T08:14:30 delt0r_: he deosnt mean that 2011-09-17T08:14:33 shat 2011-09-17T08:14:40 I never remember that 7 layer crap 2011-09-17T08:14:47 osi 2011-09-17T08:14:57 big pile of BS 2011-09-17T08:14:58 to hell with it, he isnt interested about that 2011-09-17T08:15:14 foRei: not http that for sure 2011-09-17T08:15:23 they are all binary protocols 2011-09-17T08:15:32 mcstar I figured that aswell, but the other protocols seem somewhat "not fitting" aswell, like smtp etc 2011-09-17T08:15:37 all custom as i said above 2011-09-17T08:16:15 foRei: why do you care? its just a protocol... 2011-09-17T08:16:35 its not like one is far superior than the other 2011-09-17T08:16:39 I want to figure out which one to use for my game 2011-09-17T08:16:44 alright so http would be fine? 2011-09-17T08:16:47 no 2011-09-17T08:16:55 go with ioquake's 2011-09-17T08:17:10 yep 2011-09-17T08:17:13 not sure what is called, but try to use an opensource one 2011-09-17T08:17:28 only UDP is a PITA for a first time network person 2011-09-17T08:17:38 its ioquake 2011-09-17T08:17:53 art assets are not included.. 2011-09-17T08:18:16 Ok, thanks :) 2011-09-17T08:18:35 foRei: are you making an fps? 2011-09-17T08:19:07 uhm, somewhat :D 2d multiplayer snake/tron with ability to shoot 2011-09-17T08:19:21 argh, thats what i was afraid of 2011-09-17T08:19:35 i mean you dont have to be compatible with anything than 2011-09-17T08:20:00 a protocol that an fps uses is very different from what you need 2011-09-17T08:20:20 I've looked at WCF but im not sure 2011-09-17T08:21:24 why not stay posix? 2011-09-17T08:22:26 Im not that much into stuff, I dont know posix 2011-09-17T08:23:15 posix=simple+portable 2011-09-17T08:23:30 do you really want to stick with windows? 2011-09-17T08:23:34 not that i really care 2011-09-17T08:24:00 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has joined #aichallenge 2011-09-17T08:24:26 Im just trying to make a multiplayer game, not too concerned about what I use for now. Eventually I'll try them all one by and to learn advance/disadvantage. Is WCF a bad starting point? 2011-09-17T08:25:21 i only that that it has windows in its name so yes, a bad starting point 2011-09-17T08:25:30 hehe 2011-09-17T08:25:38 thats something on top of .net right? 2011-09-17T08:25:51 Yeah, Im programming in C# .net 2011-09-17T08:26:33 i would begin with something on a lower level 2011-09-17T08:26:50 mutiplayer tic tac toe 2011-09-17T08:26:53 and you really shouldnt consider "fps" protocols for your game 2011-09-17T08:27:20 just imagine that you packets are structs, and you may use udp to send them as delt0r_ mentioned 2011-09-17T08:27:24 use plain TCP with nice easy to parse text --then tools like tcpdump can work well 2011-09-17T08:27:41 udp is not a good first choice 2011-09-17T08:27:45 delt0r_: no, why text? 2011-09-17T08:27:50 thats messy 2011-09-17T08:28:01 tools like tcpdump etc can work really well with em 2011-09-17T08:28:16 trust me there really is a good reason to keep it human readable 2011-09-17T08:28:18 you can have packets filters that convert back your binary 2011-09-17T08:28:26 you have a bug 2011-09-17T08:28:40 where is it... your game, your packets or conversion code 2011-09-17T08:28:46 but now you have to parse it 2011-09-17T08:29:06 its just one more thing that can go wrong and misslead you when bug hunting 2011-09-17T08:29:16 but ok, it might be better to be able to see it 2011-09-17T08:29:22 you can just read with you eyes... 2011-09-17T08:29:29 human parsable/readable 2011-09-17T08:29:29 and dont call me Miss. 2011-09-17T08:29:36 lol 2011-09-17T08:29:44 well that is my 2c 2011-09-17T08:29:51 from experiance... 2011-09-17T08:30:09 its all easy peasy till its on real life networks 2011-09-17T08:30:15 then it hits the fan 2011-09-17T08:33:15 my first read of that sentence had the opposite effect on me of what you meant 2011-09-17T08:36:15 *** mathis_ has quit IRC (Quit: Page closed) 2011-09-17T08:48:33 *** xathis has joined #aichallenge 2011-09-17T08:51:26 *** xathis has quit IRC (Read error: Connection reset by peer) 2011-09-17T08:51:58 *** xathis has joined #aichallenge 2011-09-17T09:07:28 *** Palmik has joined #aichallenge 2011-09-17T09:08:49 *** Palmik_ has joined #aichallenge 2011-09-17T09:09:07 *** Palmik has quit IRC (Disconnected by services) 2011-09-17T09:12:15 *** Palmik_ is now known as Palmik 2011-09-17T09:12:29 So when say 2 of my ant move to the same square they are removed without points correct? 2011-09-17T09:12:50 yet if they get killed, the killers get points? 2011-09-17T09:13:21 so if i see that my ants can't win a given fight, then suicide is in fact a good thing to do? 2011-09-17T09:15:16 it was discussed several times 2011-09-17T09:15:26 oyu can minimize the gain of the enemy yes 2011-09-17T09:16:20 just checking... does not seem like a winning strat to me... but non the less :interesting 2011-09-17T09:16:29 i had a suicidal bot some time ago, did pretty well for that 2011-09-17T09:17:43 *** kaemo0 has joined #aichallenge 2011-09-17T09:21:50 *** stocha has joined #aichallenge 2011-09-17T09:22:00 wow could hardly be tcp only could it. 2011-09-17T09:22:13 i'm not sure i will participate to this challenge. 2011-09-17T09:22:30 i mean, games are small and very crowded. The game doesn't appeal to me anymore. 2011-09-17T09:22:37 *** xathis has quit IRC () 2011-09-17T09:22:51 anyone care to convince me to participate ? :p 2011-09-17T09:22:54 *** Palmik has quit IRC (Remote host closed the connection) 2011-09-17T09:22:57 *** kaemo0 has quit IRC (Quit: *puff*) 2011-09-17T09:23:14 *** xathis has joined #aichallenge 2011-09-17T09:23:37 stocha: youre right, you really shouldnt participate 2011-09-17T09:23:47 *** Palmik has joined #aichallenge 2011-09-17T09:24:01 in fact, you make the channel crowded, why are you here? 2011-09-17T09:24:18 and now comes the laughing face 2011-09-17T09:24:20 :D 2011-09-17T09:24:30 stocha: hi, we missed you 2011-09-17T09:24:35 mcstar: i didn't finalize my combat solving code. And i'm not very inclined to work on it 2011-09-17T09:25:01 i still have read the logs. Once per month. not all. Very few of them. Just to see if the thing would launch :p 2011-09-17T09:25:13 apparently it's about to go live. 2011-09-17T09:25:19 very much so 2011-09-17T09:25:20 maybe. 2011-09-17T09:25:38 i still think it should be made a 1vs1. 2011-09-17T09:26:00 so even before the contest started you sort of "finished" your work on it? 2011-09-17T09:26:05 then maybe i would try to code something. But as it is, i just can't understand anything of what is going on. I find very few fun in that. 2011-09-17T09:26:18 little 2011-09-17T09:26:29 fun is uncountable 2011-09-17T09:27:02 the main point of the game is to write a good fighting alg. 2011-09-17T09:27:04 mcstar: we had a huge amount of time to build a final bot during the beta. The main unknown being the environnement. Maps, food etc. Which may still be unknown until the end really. 2011-09-17T09:27:20 stocha: no, they will remain unknown 2011-09-17T09:27:51 ah, you meant end of beta or contest? 2011-09-17T09:27:53 nvm 2011-09-17T09:28:02 end of contest. 2011-09-17T09:28:08 yeah, than right 2011-09-17T09:28:37 i think this was the easy part, now comes the hard one 2011-09-17T09:28:41 did the top bot maker provide feed back about how fun it is to build a bot ? 2011-09-17T09:28:51 the bots dont really have any ai in them, unless they can fight well 2011-09-17T09:29:00 did they really make a lot of progress ? 2011-09-17T09:29:12 well, you can see it for yourself 2011-09-17T09:29:18 revolt has some nice tactics 2011-09-17T09:29:21 my bot had a lot of ai in them. But no fighting code yet. 2011-09-17T09:29:29 he contracts the regions 2011-09-17T09:29:36 currently the ranking is down. (to few games per bot) 2011-09-17T09:29:59 yes, its volatile these days, they were reset yesterday even 2011-09-17T09:30:22 stocha: you know i had a snake bot too 2011-09-17T09:30:25 have you seen it? 2011-09-17T09:30:29 *** Naktibalda has joined #aichallenge 2011-09-17T09:30:38 now that was almost AI, but otherwise it sucked 2011-09-17T09:31:03 snake bot have poor chance of being top bots :) 2011-09-17T09:31:10 not really 2011-09-17T09:31:16 you have to make a smart snake bot 2011-09-17T09:31:20 many little snakes 2011-09-17T09:31:23 also i'm color blind. So i just have no chance of understanding a multiplayer game. Unless it is very structured :) 2011-09-17T09:31:33 i forgot that 2011-09-17T09:31:46 lol, i cant imagine whats that like 2011-09-17T09:32:16 stocha: the visualizer can put little letters on the ants 2011-09-17T09:32:39 yeah but the letter are even less readable. 2011-09-17T09:32:51 basically, it's like every bot share the same color. 2011-09-17T09:32:55 project it to a wall 2011-09-17T09:33:15 or half of them one color, and the other half another one. All in all, it's hard to read. 2011-09-17T09:33:37 do these colors stay the same? 2011-09-17T09:33:46 or depends on the environmet you see them in 2011-09-17T09:33:53 even when zooming, i had more trouble interpreting the letters. 2011-09-17T09:34:21 it's more a problem of : i c'ant track which ant belongs to the same team. 2011-09-17T09:34:46 1vs1 would be less of a problem. 2011-09-17T09:35:03 easier to find easy to see colors. 2011-09-17T09:35:10 and less messy overall most probably. 2011-09-17T09:35:47 mm, i guess i'll be going. 2011-09-17T09:35:49 see you :) 2011-09-17T09:36:08 maybe i'll read what's going on up until launch. If it is next week. 2011-09-17T09:36:28 *** stocha has quit IRC (Quit: Page closed) 2011-09-17T09:36:56 http://aichallengebeta.hypertriangle.com/visualizer.php?game=168258 2011-09-17T09:36:59 noooo 2011-09-17T09:37:03 why did he go 2011-09-17T09:37:29 McLeopold: "so, if you have a choice of playing 2 players with match qualities 50% and 51%, right now I always pick 51%" <-- i thought it was supposed to be probabilistic? 2011-09-17T09:40:23 McLeopold: i had suggested a long time ago that at any given time we just have a subset of players that are available for play. each time a player plays a game, he is removed from the candidate list and replaced with whatever player hasn't played a game for the longest time. the matching algorithm may only pick players from the candidate list 2011-09-17T09:41:22 *replaced with a player not already in the candidate list who has not played a game for the longest time 2011-09-17T09:43:24 *** xathis has quit IRC () 2011-09-17T09:44:06 *** xathis has joined #aichallenge 2011-09-17T09:49:02 *** Naktibalda has quit IRC (Quit: ChatZilla 0.9.87 [Firefox 6.0.2/20110905175240]) 2011-09-17T10:00:12 *** Palmik has quit IRC (Ping timeout: 240 seconds) 2011-09-17T10:23:33 mcstar: imagines bot looks like little larvae 2011-09-17T10:23:58 no, its mcstar's bot 2011-09-17T10:24:20 yeam, looks more like worms than snakes :D 2011-09-17T10:25:00 stocha was here, but seems he lost his intereset 2011-09-17T10:32:46 *** Kingpin13 has joined #aichallenge 2011-09-17T10:33:31 *** Palmik has joined #aichallenge 2011-09-17T11:06:32 *** dvladim has quit IRC (Ping timeout: 276 seconds) 2011-09-17T11:22:46 *** amstan has joined #aichallenge 2011-09-17T11:22:47 *** ChanServ sets mode: +o amstan 2011-09-17T12:17:56 *** sigh has quit IRC (Remote host closed the connection) 2011-09-17T12:24:36 *** dvladim has joined #aichallenge 2011-09-17T12:36:52 *** eashoka has quit IRC (Ping timeout: 252 seconds) 2011-09-17T12:46:24 *** McLeopold has quit IRC (Quit: Leaving.) 2011-09-17T12:55:38 *** Accoun has quit IRC (Ping timeout: 260 seconds) 2011-09-17T13:05:42 *** Accoun has joined #aichallenge 2011-09-17T13:07:36 *** McLeopold has joined #aichallenge 2011-09-17T13:28:49 *** dvladim has quit IRC (Quit: Konversation terminated!) 2011-09-17T13:29:15 *** dvladim has joined #aichallenge 2011-09-17T13:32:29 *** luda has joined #aichallenge 2011-09-17T13:39:18 *** mceier has quit IRC (Ping timeout: 260 seconds) 2011-09-17T13:40:30 *** mceier has joined #aichallenge 2011-09-17T13:40:39 *** luda has quit IRC (Quit: Page closed) 2011-09-17T13:47:27 *** cody__ has joined #aichallenge 2011-09-17T14:23:14 *** olexs has joined #aichallenge 2011-09-17T14:24:20 hi guys. is there a tcp server running somewhere for ants where one could test a bot a bit faster than on the beta server? 2011-09-17T14:29:37 *** berak has quit IRC (Quit: ChatZilla 0.9.84 [SeaMonkey 2.0a3/20090223135443]) 2011-09-17T14:31:04 theres none i know of 2011-09-17T14:36:17 hm, too bad 2011-09-17T14:38:06 *** onensora has joined #aichallenge 2011-09-17T14:38:30 *** dvladim has quit IRC (Ping timeout: 252 seconds) 2011-09-17T14:44:08 *** dr- has quit IRC (Ping timeout: 260 seconds) 2011-09-17T14:44:40 *** dr- has joined #aichallenge 2011-09-17T14:56:12 *** xathis has quit IRC () 2011-09-17T15:07:48 *** McLeopold has joined #aichallenge 2011-09-17T15:15:27 @later tell janzert how do I check the amazon spot price? what type of instances do we have? 2011-09-17T15:15:27 McLeopold: Ready to serve, my lord. 2011-09-17T15:20:04 poor stocha 2011-09-17T15:21:11 little pixels in the colors of the rainbow are not a good presentation when you are color blind 2011-09-17T15:24:14 *** delt0r_ has quit IRC (Ping timeout: 259 seconds) 2011-09-17T15:27:14 mleise: i dont really understand his point though 2011-09-17T15:27:52 he says that he lost interest in the game and that doesnt want to write a battle strategy 2011-09-17T15:28:11 on the contrary, i think this will be the interesting part 2011-09-17T15:28:45 i think that something similar will happen like in the previous contest 2011-09-17T15:28:56 the battle strategy will be a bit clearer 2011-09-17T15:29:10 and the implementation, and the ability to predict will win 2011-09-17T15:30:09 basically what im saying is that i dont expect some "magic" to happen(i dont think there are deep hidden simplifications) but there will be some tricks on the ai part 2011-09-17T15:32:53 anyway, when there are many players, even colors dont help 2011-09-17T15:36:17 *** delt0r_ has joined #aichallenge 2011-09-17T15:37:06 *** onensora has quit IRC (Ping timeout: 252 seconds) 2011-09-17T15:43:42 dumb me, searching for "crystal crack" on youtube 2011-09-17T15:47:51 *** Accoun has quit IRC () 2011-09-17T15:53:16 aichallenge: McLeopold epsilon * r577fd39 / website/api_game_result.php : fix cutoff - http://git.io/RhPtcA 2011-09-17T16:00:05 *** Accoun has joined #aichallenge 2011-09-17T16:10:48 *** Palmik has quit IRC (Remote host closed the connection) 2011-09-17T16:19:42 *** foRei has quit IRC (Read error: Connection reset by peer) 2011-09-17T16:20:46 *** foRei has joined #aichallenge 2011-09-17T16:23:12 *** onensora has joined #aichallenge 2011-09-17T17:49:58 *** mcstar has quit IRC (Quit: WeeChat 0.3.5) 2011-09-17T17:57:10 *** JaycobColeman has joined #aichallenge 2011-09-17T18:01:33 McLeopold: http://cloudexchange.org/charts/us-east-1.linux.m2.xlarge.html is a graph for the right instance type spot price 2011-09-17T18:02:11 and they're called "high memory extra large" by amazon 2011-09-17T18:03:45 *** mceier has quit IRC (Quit: leaving) 2011-09-17T18:05:14 *** olexs has quit IRC (Quit: Leaving.) 2011-09-17T18:25:32 *** Kingpin13 has quit IRC (Quit: quit) 2011-09-17T18:57:26 *** JaycobColeman has quit IRC (Quit: JaycobColeman) 2011-09-17T19:02:54 *** JaycobColeman has joined #aichallenge 2011-09-17T19:04:46 *** arlequin has quit IRC (Quit: Page closed) 2011-09-17T19:08:07 @later tell janzert how many worker.py instances per worker server? I'm seeing a lot of matchup requests, 5 times as many as posted games. 2011-09-17T19:08:07 McLeopold: Yep. 2011-09-17T19:09:24 *** antimatroid has quit IRC (Ping timeout: 240 seconds) 2011-09-17T19:12:52 *** cody__ has quit IRC (Ping timeout: 252 seconds) 2011-09-17T19:15:00 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (Ping timeout: 240 seconds) 2011-09-17T19:16:43 *** JaycobColeman has quit IRC (Quit: JaycobColeman) 2011-09-17T19:18:09 McLeopold: only one 2011-09-17T19:18:42 ok, I think a worker is misbehaving, 198, but don't do anything yet, I'm improving the logging until it makes sense for us 2011-09-17T19:18:59 ok 2011-09-17T19:19:14 going for dinner now so out for a while anyway 2011-09-17T19:19:17 cya 2011-09-17T19:31:54 *** antimatroid has joined #aichallenge 2011-09-17T19:33:18 aichallenge: McLeopold epsilon * re33a70d / worker/worker.py : allow game to mark submission status - http://git.io/Xpy73w 2011-09-17T19:41:30 *** amstan_ has joined #aichallenge 2011-09-17T19:41:30 *** ChanServ sets mode: +o amstan_ 2011-09-17T19:42:36 *** amstan has quit IRC (Read error: No route to host) 2011-09-17T19:55:04 *** amstan_ is now known as amstan 2011-09-17T19:56:10 aichallenge: McLeopold epsilon * r0f84e4e / worker/worker.py : only report game compile status if failure - http://git.io/lZxR1A 2011-09-17T20:00:24 @later tell janzert it was the go submission again, I've modifed worker to report failure status on games so submission can be turned off, the workers have been restarted 2011-09-17T20:00:24 McLeopold: OK 2011-09-17T20:06:55 *** amstan has quit IRC (Ping timeout: 260 seconds) 2011-09-17T20:08:01 aichallenge: Scott Hamilton epsilon * rb2ea67c / (4 files): better php_api logs - http://git.io/dLmQSQ 2011-09-17T20:13:20 *** FireFly has quit IRC (Quit: FireFly) 2011-09-17T20:32:03 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has joined #aichallenge 2011-09-17T20:56:25 *** amstan has joined #aichallenge 2011-09-17T20:56:25 *** ChanServ sets mode: +o amstan 2011-09-17T20:58:06 *** nux67 has quit IRC (Ping timeout: 252 seconds) 2011-09-17T21:16:12 *** foRei has quit IRC (Read error: Connection reset by peer) 2011-09-17T21:17:18 *** foRei has joined #aichallenge 2011-09-17T21:52:09 *** sigh has joined #aichallenge 2011-09-17T23:24:12 *** delt0r_ has quit IRC (Ping timeout: 240 seconds) 2011-09-17T23:36:52 *** delt0r_ has joined #aichallenge 2011-09-17T23:42:02 *** xathis has joined #aichallenge 2011-09-17T23:52:59 *** xathis has quit IRC () 2011-09-17T23:55:01 *** ChrisH has joined #aichallenge