2010-11-20T00:01:05 janzert: really? i thought waterloo and the main server were in est 2010-11-20T00:01:26 i mean, they were edt at the start of the tournament, which would explain why it seems like cst now... 2010-11-20T00:02:13 nope it did the switch at the daylight savings time like it should 2010-11-20T00:02:18 ah 2010-11-20T00:02:22 i dunno then 2010-11-20T00:02:22 messed up the averages when it did :) 2010-11-20T00:02:39 yeah, I don't know where it's actually located but I do know it has been CST all along 2010-11-20T00:03:09 you should see what happens with monitoring graphs when daylight savings occurs 2010-11-20T00:03:13 it's hilarious 2010-11-20T00:03:15 hehe 2010-11-20T00:03:17 yeah 2010-11-20T00:08:41 i've been out of the loop for a while... i love the new map generation 2010-11-20T00:18:17 Top 10 players: bocsimacko(3414), bix0r4ever(3212), wagstaff(3111), protocolocon(3105), Error323(3104), medrimonia(3096), davidjliu(3081), dmj111(3075), smloh1(3062), george(3053) 2010-11-20T00:20:09 *** delt0r___ has joined #aichallenge 2010-11-20T00:21:50 *** delt0r has quit IRC (Ping timeout: 276 seconds) 2010-11-20T00:23:54 *** cfaftw has quit IRC (Quit: Leaving) 2010-11-20T00:37:36 "To be safe download from sites like pirate bay or kick arse torrents. And read the comments people have posted about them." ahahaha 2010-11-20T00:47:53 what would be the easiest way to get a text copy of a .png file? 2010-11-20T00:48:10 it's a black and white image file for a huge maze, and i have the urge to solve it 2010-11-20T00:52:02 *** mega1 has quit IRC (Read error: Operation timed out) 2010-11-20T01:01:05 *** Appleman1234 has quit IRC (Ping timeout: 255 seconds) 2010-11-20T01:11:02 *** mega1 has joined #aichallenge 2010-11-20T01:18:18 Top 10 players: bocsimacko(3355), bix0r4ever(3216), wagstaff(3113), Error323(3113), protocolocon(3102), smloh1(3100), davidjliu(3081), dmj111(3072), medrimonia(3072), george(3054) 2010-11-20T01:31:14 *** bhasker has joined #aichallenge 2010-11-20T01:32:41 evening 2010-11-20T01:34:05 evening 2010-11-20T01:38:13 why would c++ cout characters for me as i like, but the moment i put them in an output file, they're completely messed up jumble? 2010-11-20T01:56:51 are you interleaving to the file 2010-11-20T01:57:02 as in are two processes writing tot he same file 2010-11-20T01:57:14 no, i have two different files 2010-11-20T01:57:23 if you could paste some code 2010-11-20T01:57:26 i've given up, i'll do that once ai is done 2010-11-20T01:57:27 i can help you 2010-11-20T01:57:31 k, hold up 2010-11-20T01:57:34 though if this is for logging 2010-11-20T01:57:54 i would just log everything to stdout 2010-11-20T01:57:59 sorry stderr 2010-11-20T01:58:04 and redirect to file 2010-11-20T01:58:08 to capture the logs:) 2010-11-20T01:58:25 http://codepad.org/LfXLmB3y 2010-11-20T01:58:35 that seems to almost do it, but it's still not working quite right :\ 2010-11-20T01:58:58 i'm trying to strip uesless rows and columns from a file 2010-11-20T01:59:05 but that should just be respitting it out 2010-11-20T01:59:21 *** smloh1 has joined #aichallenge 2010-11-20T01:59:22 the code looks fine to me 2010-11-20T01:59:26 let me try and run it 2010-11-20T01:59:42 how is it messed up? no end-of-lines? 2010-11-20T02:00:42 hang on, i'm doing a paste of input and output now 2010-11-20T02:00:48 it's a fairly large maze 2010-11-20T02:00:49 that looks just fine 2010-11-20T02:00:52 i don't see any issues 2010-11-20T02:01:01 the file got recreated just fine 2010-11-20T02:02:10 http://codepad.org/e7oicJR6 2010-11-20T02:02:19 that's the first 6 lines of my input 2010-11-20T02:02:50 http://codepad.org/eWQRENdj 2010-11-20T02:03:06 that's the first 6 lines of my output, the mazee stuff is wrong and there is crap at the end of lines 2010-11-20T02:04:49 i think its newlines 2010-11-20T02:04:54 are you on windows? 2010-11-20T02:05:05 i am 2010-11-20T02:05:12 i'm using notepad++ though 2010-11-20T02:05:15 that usually does me well 2010-11-20T02:06:16 i have a feeling its reading beyond end of line 2010-11-20T02:06:33 rather writing 2010-11-20T02:06:41 what you code in won't save you from the horrors of \r\n 2010-11-20T02:07:12 yea 2010-11-20T02:07:26 though c++ on windows should deal with \r\n correctly 2010-11-20T02:07:35 bah, i'll worry about it in a weeks time 2010-11-20T02:07:37 the problem might be your input file is \n delimited 2010-11-20T02:07:49 so getline is reading more than it should 2010-11-20T02:07:51 http://upload.wikimedia.org/wikipedia/commons/c/cf/MAZE.png 2010-11-20T02:08:01 i was going to try and get all paths through and and highlight the shortest 2010-11-20T02:08:02 and you are getting junk as a result 2010-11-20T02:08:03 hi 2010-11-20T02:08:25 do the following 2010-11-20T02:08:31 do you have emacs installed? 2010-11-20T02:08:35 nope :P 2010-11-20T02:08:41 or does notepad++ have a mode to show end of lines 2010-11-20T02:08:55 if so then make sure every line is terminated by a \r\n 2010-11-20T02:09:04 which in ctrl chars should be like ^M^D 2010-11-20T02:09:08 if its just ^D 2010-11-20T02:09:16 could he just open it and save it again? 2010-11-20T02:09:19 then you need to replace all line terminators with a ^M^D 2010-11-20T02:09:23 might work 2010-11-20T02:09:38 sometimes editors try to retain files original terminators 2010-11-20T02:09:44 or maybe cut and paste it into notepad and save it 2010-11-20T02:09:57 already tried resaving 2010-11-20T02:10:00 try to tell the editor to save the file in DOS mode 2010-11-20T02:10:07 if it supports that 2010-11-20T02:10:24 view|show end of line 2010-11-20T02:10:30 no, if it read past the \n it would read the whole file, not random jun 2010-11-20T02:10:31 junk 2010-11-20T02:10:50 it's reading fine, it's writing wrong 2010-11-20T02:11:09 i don't see why it should be writing wrong though 2010-11-20T02:11:47 but zannick what you say is right 2010-11-20T02:11:54 when i tell it to show line ends i just get LF in a black box there 2010-11-20T02:11:55 it will just read the file into a single line 2010-11-20T02:12:43 that is not right i think, on windows it should be a LF followed by a NL 2010-11-20T02:12:46 antimatroid1: that's easy.. 2010-11-20T02:12:49 antimatroid1: flood fill 2010-11-20T02:12:54 what if you output maze[i].c_str() instead of just maze[i]? 2010-11-20T02:13:09 Zannick that should not be required though with a c++ string 2010-11-20T02:13:10 and cout 2010-11-20T02:13:13 i know 2010-11-20T02:13:16 antimatroid1: implement a tron bot using flood fill, that should give you shortest 2010-11-20T02:13:38 i'm guessing it's outputting part of the vector container rather than just the string 2010-11-20T02:13:41 amstan: i was going to just turn it into a graph 2010-11-20T02:13:44 wait.. 2010-11-20T02:13:46 nvm 2010-11-20T02:13:49 will it? hmm 2010-11-20T02:14:03 that would be a bug in the string output code , what compiler are you using antimatroid? 2010-11-20T02:14:43 migi32 i think 2010-11-20T02:14:58 i'm pretty sure i would have been lazy and just got it with codeblocks 2010-11-20T02:15:54 amstan: it would be interesting to see two tron bots play in a connected maze 2010-11-20T02:16:06 antimatroid1: did that, wasn't that fun 2010-11-20T02:16:17 bots were too stupid? 2010-11-20T02:16:26 it's not that exciting 2010-11-20T02:16:42 oh yeah, i wouldn't want to do it more than a few times 2010-11-20T02:17:11 i miss having voronoi like diagrams for evaluation 2010-11-20T02:18:17 Top 10 players: bocsimacko(3381), bix0r4ever(3207), smloh1(3138), wagstaff(3106), protocolocon(3099), davidjliu(3084), dmj111(3076), Error323(3076), medrimonia(3073), BaronTrozo(3064) 2010-11-20T02:20:33 "MMMM[...]MMM" followed by "e0 a8 8d e4 b4 80 e4 b4 80 e2 80 82 e2 80 82 e2 80 82 e2 80 82 e4 [...]" 2010-11-20T02:28:29 *** bhasker has quit IRC (Quit: bhasker) 2010-11-20T02:53:57 isn't this typical, i fix a typo bug, and the new bot loses to the old one... 2010-11-20T02:55:08 *** Palmik has joined #aichallenge 2010-11-20T02:56:10 now those, were features. :) 2010-11-20T02:56:44 no 2010-11-20T02:57:31 ha! 2010-11-20T02:58:27 *** tapwater has quit IRC (Quit: tapwater) 2010-11-20T03:07:33 smloh1: story of my life 2010-11-20T03:07:57 i kind of hope the next game isn't as messy 2010-11-20T03:08:06 the difficulty compared to tron is good though 2010-11-20T03:12:31 i like the intractability of the problem 2010-11-20T03:13:01 i saw a study once that said cats are very cool! 2010-11-20T03:13:24 the next game shall be "break the captcha" 2010-11-20T03:13:27 >.> 2010-11-20T03:13:44 Zannick: but that's not a game 2010-11-20T03:13:47 or better yet, tennis 2010-11-20T03:13:54 you have to program a real life robot 2010-11-20T03:13:58 i saw a study once that said chaptcha is boring 2010-11-20T03:14:16 oh man.. it's working, lol 2010-11-20T03:14:19 recaptcha's pretty cool 2010-11-20T03:14:30 s/i think that/i saw a study once that said/g 2010-11-20T03:14:46 i spent a summer working for them 2010-11-20T03:14:51 heh, stop trying out ideas from xkcd 2010-11-20T03:15:05 it'll get you in trouble :) 2010-11-20T03:15:11 lol 2010-11-20T03:15:41 Zannick: apparently chaptcha is what got jeff to google 2010-11-20T03:16:05 amstan: oh? that's pretty neat 2010-11-20T03:16:35 amstan: Besides the people I had listed and Jeff's second account you don't know of any other organizer's offhand? 2010-11-20T03:16:41 i don't think it got me here 2010-11-20T03:16:49 * to google 2010-11-20T03:16:58 janzert: hold on 2010-11-20T03:17:11 i think i may have registered a few times 2010-11-20T03:17:12 I see looking at /home on the main server there are 4 more accounts but they may be unrelated to the contests 2010-11-20T03:17:28 janzert: yeah, those are not related 2010-11-20T03:17:32 ok 2010-11-20T03:17:44 there are some starter accounts, i think? 2010-11-20T03:18:02 ya 2010-11-20T03:18:04 StarterPack_Rage and StarterPack_Dual, but i don't know the ids 2010-11-20T03:18:18 Zannick: if those win.. i don't care, lol 2010-11-20T03:18:19 Top 10 players: bocsimacko(3412), Raschi(3245), bix0r4ever(3226), _iouri_(3119), wagstaff(3118), Error323(3112), medrimonia(3101), protocolocon(3098), dmj111(3090), davidjliu(3089) 2010-11-20T03:18:21 ok, I can look up the id's 2010-11-20T03:18:24 Zannick: then nobody winds 2010-11-20T03:18:28 amstan: hahahaha 2010-11-20T03:18:36 but yeah it may be more interesting to leave them active 2010-11-20T03:18:43 so people can see where they place 2010-11-20T03:18:53 i doubt it, because i wouldn't be able to find them in the rankings as is 2010-11-20T03:18:58 janzert: 5775 amstan2 2010-11-20T03:19:32 thanks 2010-11-20T03:19:42 janzert: ya.. that's all for me 2010-11-20T03:21:42 janzert: SELECT * FROM `users` WHERE `email` LIKE '%camero%' 2010-11-20T03:21:53 janzert: he has 4 accounts 2010-11-20T03:21:59 heh, ok 2010-11-20T03:22:04 oh man... he's a sneaky guy.. but we won't let him win 2010-11-20T03:22:32 the starterpacks are 3930 and 3928 btw if anyone wants to look at them 2010-11-20T03:37:53 YEAAAAAAAAAAAAH still in the top 10 2010-11-20T03:38:01 @rainbow wh0000000t 2010-11-20T03:38:01 Error323: wh0t 2010-11-20T03:43:19 cool, I didn't know they were playing. (always felt they should be) 2010-11-20T04:07:02 *** Mistmanov has joined #aichallenge 2010-11-20T04:09:47 *** Appleman1234 has joined #aichallenge 2010-11-20T04:13:58 search by username now live 2010-11-20T04:15:38 oh thats god 2010-11-20T04:15:40 good* 2010-11-20T04:18:19 Top 10 players: bocsimacko(3429), jimrogerz(3209), Raschi(3204), bix0r4ever(3200), wagstaff(3131), Error323(3121), _iouri_(3107), felixcoto(3099), medrimonia(3097), dmj111(3097) 2010-11-20T04:18:21 beats googling "site:ai-contest.com +username" 2010-11-20T04:18:27 \o/ \o/ 2010-11-20T04:18:39 medrimonia: you have one hell of an anti-error323 bot there :/ 2010-11-20T04:18:42 bastard :P 2010-11-20T04:18:45 janzert: <3 2010-11-20T04:19:03 http://www.ai-contest.com/profile.php?user_id=3945 2010-11-20T04:19:08 my page is fairly boring and has been a while aha 2010-11-20T04:19:24 o_O 2010-11-20T04:20:02 the errors were from it not liking my old bot being spread over multiple folders, then i haven't got the rewrite to state to upload :P 2010-11-20T04:20:09 in fact there's stil a lot of work to do :\ 2010-11-20T04:20:26 Error323: shall we reupload our old bots and pretend nothing has happened -_- 2010-11-20T04:20:46 Mistmanov: :/ 2010-11-20T04:20:57 me at 40 and you at 80 was better than me at 158 and you at 8 :p 2010-11-20T04:21:16 well... no :P 2010-11-20T04:21:28 :( 2010-11-20T04:21:36 but the fact that ur dropped that hard is quite harsh 2010-11-20T04:21:40 how did that happen? 2010-11-20T04:21:58 I seriously just lost from #800... 2010-11-20T04:22:16 I think I made some calculation errors in me expansion calculator 2010-11-20T04:22:26 that made it way to conservative 2010-11-20T04:22:29 contestbot @rainbow 158!! who do I kill to get that ranking? 2010-11-20T04:22:30 Mjothvitnir: Run as fast as you can and don't look back. 2010-11-20T04:22:34 janzert, amstan I'm wondering how important it is to win the first 3 games instead of losing the first 3. Will you always climb back up or can you be trapped in local minima/maxima 2010-11-20T04:22:38 oops 2010-11-20T04:22:43 @rainbow 158!! who do I kill to get that ranking? 2010-11-20T04:22:43 Mjothvitnir: !! who do I kill to get that ranking? 2010-11-20T04:22:49 mistmanov: why is there mixtmaov, mistmanov2 and mistmanovx users for you? :P 2010-11-20T04:23:05 mixtmaov? shouldn't be me :p 2010-11-20T04:23:12 oh, yeah :P 2010-11-20T04:23:16 just went into them 2010-11-20T04:23:37 Error323: Say on resubmit you win 3 lose 3 then win the rest or lose 3 then win all the rest 2010-11-20T04:23:52 Error323: in the limit you should end up in the same place 2010-11-20T04:24:02 ah, and mistmanov2 was something about entering a non-existing email adress -_- 2010-11-20T04:24:18 well in theory janzert yes... but I wonder what happens in practice :/ 2010-11-20T04:24:30 I wonder what would happen to Mistmanov if he won the first 3 2010-11-20T04:24:30 (got a whole bunch of different aliases coupled to different domains for my university email... something I pick a combination that doesn't exist) 2010-11-20T04:25:10 it changes who you play against if you lose early yeah 2010-11-20T04:25:19 I'd say his recent loss to the 812 rank was much more important than the first loss to th 92 rank 2010-11-20T04:25:31 well not just say, I'm quite sure 2010-11-20T04:26:05 if you lose against a 800, you don't desrve a spot in the top 100 I think =) 2010-11-20T04:26:27 well 2010-11-20T04:26:32 it happens 2010-11-20T04:26:34 hehe 2010-11-20T04:26:40 give me a bit and I can give you what his ranking would be without either of those games 2010-11-20T04:26:59 although you're right that my bot performs better against better opponents 2010-11-20T04:27:10 it doesn't handle players who expand recklessly very well 2010-11-20T04:27:25 (at least, on the tcp server it didn't get utterly destroyed..) 2010-11-20T04:27:38 janzert: well my point is the probability of him facing an 800 player when winning first 3 would have been reduced drastically 2010-11-20T04:27:44 but yeah 800 is pretty low... 2010-11-20T04:28:06 Mistmanov: tcp server requires atleast 10 hours of play or so 2010-11-20T04:28:13 you can't say a lot after 2 hours 2010-11-20T04:28:16 its too random 2010-11-20T04:29:16 yeah but still, I sometimes win from you on the TCP for example 2010-11-20T04:29:20 but he hasn't beat a player ranked higher than 134 and lost to 4 of them 2010-11-20T04:29:38 losing against 800 and winning against 6 :P it still dependso n the map ofcourse 2010-11-20T04:29:46 but he does win of me (don't know with what ratio though) 2010-11-20T04:30:25 well its just something to be scared about :P 2010-11-20T04:30:26 * lavalamp_ <--- place 150 on main, yay 2010-11-20T04:30:48 Error323: yeah, and I don't think he should likely be below 800 rank either 2010-11-20T04:30:49 ah it doesn't really matter. the winner of the contest shouldn't lose because his enemy is not good enough -_- 2010-11-20T04:31:13 people in the top 100 will still be playing lower ranked bots yeah? 2010-11-20T04:31:20 yes 2010-11-20T04:31:40 it is pretty funny that the "top bots" would be so fragile that they can't handle an opponent that doesn't make "the best move" 2010-11-20T04:31:45 janzert: if I keep hoovering in the top 20 and after lets say 300 games I meet an opponent from rank 800 that beats me, i would barely drop at all 2010-11-20T04:31:57 right 2010-11-20T04:32:12 Error323: bocsimacko lost to someone in the 600s the other day and stayed in first 2010-11-20T04:32:15 but you'll have faced many opponents near that rank in that time 2010-11-20T04:32:27 janzert: that's true 2010-11-20T04:33:00 my worry is with playing better bots once you win the first few games and then having to play less games against bad bots than an equally good bot that lost some of its first games 2010-11-20T04:33:12 i'm also losing quite a few matches against lavalamp for instant lol 2010-11-20T04:33:26 oh the shame ;) 2010-11-20T04:33:36 Error323: Hmmm it's good to waking up and see that I've taken my revenge 2010-11-20T04:33:48 medrimonia: I hate you =) 2010-11-20T04:34:01 but i don't really care, that's just something i don't see as not being independent of order in which one wins 2010-11-20T04:34:08 seriously your bot beats mine nearly everytime X_X 2010-11-20T04:34:11 I've lost against #111 and I'm still in top 10^^ 2010-11-20T04:34:23 ... :P 2010-11-20T04:34:33 viv la france et hollandia :P 2010-11-20T04:34:36 I'm snipping you !!!!! 2010-11-20T04:34:50 yeah 2010-11-20T04:34:58 quite often indeed. 2010-11-20T04:35:03 i should fix that :P 2010-11-20T04:35:13 The funnier is that two of the five first in france at least are not really french 2010-11-20T04:35:47 elo code and opponent picker don't play well together :) 2010-11-20T04:35:55 I just put french because I'm studying in France 2010-11-20T04:36:01 raschi has a strategy where he just reuploads until he wins the first 4 games lol 2010-11-20T04:36:10 but it seems to work to some extend ... 2010-11-20T04:36:39 * janzert thinks he should force resubmits to pair from the middle of rankings again so people can see that they climb back up :P 2010-11-20T04:37:02 http://ai-contest.com/visualizer.php?game_id=8048495 yay, my "bail out" code is working (see turn 152.. made me keep the planet in the bottom left corner for an extra turn) 2010-11-20T04:37:47 haha 2010-11-20T04:37:52 (spectacular feature, I know ;)) 2010-11-20T04:37:55 The funnier is that Iwasn't expecting my bot to be ranked even in the top 20 2010-11-20T04:38:06 yea same here 2010-11-20T04:38:15 cuz i had quite some alcohol yesterday 2010-11-20T04:38:23 And sometimes it's the opposite^^ 2010-11-20T04:38:30 but the marges are really small in the top 100 2010-11-20T04:38:38 except for one dude ofc -_- 2010-11-20T04:39:02 I'm thinking "oh that's good, I've made an huge improve" and I'm losing ranking^^ 2010-11-20T04:39:39 * lavalamp_ grins about Error323's games with me 2010-11-20T04:39:47 hmmm my new bot on tcp seems to be better than the official one =P 2010-11-20T04:39:58 medrimonia: indeed 2010-11-20T04:40:33 this game is pretty exciting: http://72.44.46.68/canvas?game_id=1103394 2010-11-20T04:40:35 but the userspace is really small 2010-11-20T04:40:53 lavalamp_: nice work 2010-11-20T04:41:04 yeah I'm pretty happy 2010-11-20T04:41:13 good victory 2010-11-20T04:41:13 I'm terribly inconsistent though 2010-11-20T04:41:27 well apparently... so am i ^_^ 2010-11-20T04:41:40 george obviously has the rule that it's ok if he loses a planet for a turn 2010-11-20T04:41:46 oh just won against smloh1 : 2010-11-20T04:43:34 but ye medrimonia your sniping is quite annoying 2010-11-20T04:43:44 I have a sniping problem 2010-11-20T04:43:46 :( 2010-11-20T04:44:00 hahaha It's because you're not conservative enough 2010-11-20T04:44:08 I need to fix it, I get sniped in every game 2010-11-20T04:44:22 I had the same problem has you before Error323 2010-11-20T04:44:25 I thought my defense code woud take care of it but no :( 2010-11-20T04:45:31 medrimonia: I'll work it out just you wait :P 2010-11-20T04:45:50 my new defensive code is working so well otherwise :( 2010-11-20T04:46:15 Error323: Now I've other issues, but... my main issue on tcp server is BAD CONNECTION!!!!!! 2010-11-20T04:46:42 lavalamp_: make sure your defense code also defends neutrals that you attack along with the enemy 2010-11-20T04:46:48 *** tim_ has joined #aichallenge 2010-11-20T04:47:20 haha beat you on tcp Error323: 2010-11-20T04:47:28 * lavalamp_ *checks* 2010-11-20T04:47:39 medrimonia: ur beating me a lot on tcp 2010-11-20T04:47:49 more than i beat you it seems >_< 2010-11-20T04:48:01 however 2010-11-20T04:48:07 Yes, I've changed some thing to expand more 2010-11-20T04:48:15 its also about being effective against others :P 2010-11-20T04:48:20 yeah it definitely already is 2010-11-20T04:48:34 I won 5 games against dmj in a stroke^^ 2010-11-20T04:48:50 yeah I saw that 2010-11-20T04:49:02 it's possible that my eval function thinks the position is so good it doesn't matter if I get sniped 2010-11-20T04:49:13 chances of that happening are really slim when ur less good, so gratz ur probably better :D 2010-11-20T04:49:50 dmj and bocsi I have never beaten yet I don't think 2010-11-20T04:49:51 i think they are (1/2)^5 ? 2010-11-20T04:50:14 Beat mega1 or die trying! 2010-11-20T04:50:18 i did, also bocsi but its raaaaaaare 2010-11-20T04:50:52 My new bot hasn't played really against him 2010-11-20T04:51:09 But his 0.70 version is really good 2010-11-20T04:51:36 You LOSE against medrimonia 2010-11-20T04:51:37 a month ago, I was winning more against him than now I think 2010-11-20T04:51:45 I love how that just screams in my face 2010-11-20T04:51:54 I have an idea for an experiment to check first game effect, but I'm not sure I can actually carry it out. Basically start to ragebots in the contest and artificially give one a win and one a loss against the current ragebot. Then you can watch what happens 2010-11-20T04:52:17 who wants to sell me logitech g25/g27 racing wheel? 2010-11-20T04:52:45 janzert: good idea, but i think you need a real bot with a clear non-transitive relation vs other bots 2010-11-20T04:52:58 or well it might work with ragebot also... not sure 2010-11-20T04:53:25 actually I think whatever I do you'll still have doubts ;) 2010-11-20T04:53:45 lol nah its a good test imo 2010-11-20T04:53:48 janzert: Will we be able to see the games we're playing after the 27th? 2010-11-20T04:53:55 i was going to say, i'd prefer it to be done with a higher ranked bot :P 2010-11-20T04:53:56 I believe so yes 2010-11-20T04:54:12 Cool 2010-11-20T04:54:15 Jeff hasn't said anything about shutting it off and I did not mention it in the proposed checklist 2010-11-20T04:54:17 Error323: did you see this game? http://tcp.zeroviz.us/canvas.cgi?game_id=14717 2010-11-20T04:54:18 indeed that is cool 2010-11-20T04:54:50 I have the feeling that I'll be thinking "gosh I've made a very bad mistake and I can't correct it. Damn!" 2010-11-20T04:54:52 lavalamp_: ouch 2010-11-20T04:55:02 hahah 2010-11-20T04:55:09 I gues it counts growth on the last turn? 2010-11-20T04:55:27 lavalamp_: no it counts ships 2010-11-20T04:55:27 it claims I won even though you have 4 more ships at the end 2010-11-20T04:55:37 two things I don't have a higher ranked bot and whatever bot is used for the first game needs to not be resubmitted soon or the first game won't be counted in the rating calculation for either of the experiment bots anyway 2010-11-20T04:55:43 no I mean it adds on growth before comparing 2010-11-20T04:56:20 mja dunno, but i'm quite sure its correct in the game itself 2010-11-20T04:56:26 state info is for start of turn, winner is end of turn 2010-11-20T04:56:30 i'm very conservative with neutral capping 2010-11-20T04:56:39 i only take them when losing or drawing 2010-11-20T04:56:46 ah yeah 2010-11-20T04:56:57 I was going to say your expansion is not rapid enough 2010-11-20T04:58:15 myes well I haven't figured out a way to properly fix that 2010-11-20T04:58:31 but i've got more important antisnipe issues in front of me 2010-11-20T04:58:39 or above that even... 2010-11-20T04:58:40 i need food 2010-11-20T04:58:43 bbl 2010-11-20T04:58:49 @rankings 2010-11-20T04:58:50 Error323: Top 10 players: bocsimacko(3440), Raschi(3304), bix0r4ever(3179), _iouri_(3168), wagstaff(3136), felixcoto(3129), Error323(3116), medrimonia(3105), dmj111(3094), protocolocon(3089) 2010-11-20T04:58:51 xD 2010-11-20T04:59:29 oh nice I *have* beaten dmj: http://tcp.zeroviz.us/canvas.cgi?game_id=14750 2010-11-20T04:59:30 It's so good to see that!!!! 2010-11-20T04:59:51 medrimonia: Ohh yea 2010-11-20T04:59:52 :D 2010-11-20T05:01:26 lavalamp_: well your definitly getting much better, I <3 the turns 148-151, very bocsimacko like xD 2010-11-20T05:01:48 like a piechart 2010-11-20T05:02:14 hehe ^^ 2010-11-20T05:03:28 *** Zoost has quit IRC (Ping timeout: 240 seconds) 2010-11-20T05:12:57 *** Tobu has quit IRC (Quit: Leaving.) 2010-11-20T05:18:18 Top 10 players: bocsimacko(3392), Raschi(3332), _iouri_(3190), bix0r4ever(3172), jimrogerz(3134), wagstaff(3127), felixcoto(3122), medrimonia(3108), Error323(3097), protocolocon(3094) 2010-11-20T05:20:29 zzzzzzzzzzzzzzzzzzzzzzzz 2010-11-20T05:22:32 *** Zoost has joined #aichallenge 2010-11-20T05:28:04 *** choas has joined #aichallenge 2010-11-20T05:29:04 *** Zoost has quit IRC (Ping timeout: 245 seconds) 2010-11-20T06:03:52 *** tim_ has quit IRC (Remote host closed the connection) 2010-11-20T06:04:48 http://ai-contest.com/visualizer.php?game_id=8055509 2010-11-20T06:04:59 men this is clearly inferior play on my side 2010-11-20T06:05:25 and women yes Mistmanov :P 2010-11-20T06:11:45 *** Apophis has joined #aichallenge 2010-11-20T06:13:38 *** jaspervdj has joined #aichallenge 2010-11-20T06:14:54 *** nann has quit IRC (Ping timeout: 250 seconds) 2010-11-20T06:17:44 didn't take into account all his resources? 2010-11-20T06:18:08 indeed 2010-11-20T06:18:19 Top 10 players: bocsimacko(3407), Raschi(3261), _iouri_(3233), bix0r4ever(3157), felixcoto(3145), wagstaff(3135), medrimonia(3113), protocolocon(3093), dmj111(3071), Error323(3067) 2010-11-20T06:18:51 *** narnach has quit IRC (Quit: Leaving.) 2010-11-20T06:31:30 *** yasith has joined #aichallenge 2010-11-20T06:35:56 *** Zoost has joined #aichallenge 2010-11-20T06:38:27 *** Migi32 has joined #aichallenge 2010-11-20T06:41:05 *** Zoost has quit IRC (Ping timeout: 272 seconds) 2010-11-20T06:41:24 what the hell 2010-11-20T06:41:27 spain is dominating 2010-11-20T06:41:48 nooooo i'm 11th 2010-11-20T06:41:58 CURSESSS 2010-11-20T06:43:34 I'm very slowly but surely working my way up again, having a good winningstreak 2010-11-20T06:43:51 started out by beating the number 3000 (seriously), then 1000, then 100-something.. 2010-11-20T06:44:01 O_O 2010-11-20T06:44:03 so be afraid error, I'm coming for you ;) 2010-11-20T06:44:23 hehe i'll be waiting... at around 20 :P 2010-11-20T06:49:07 hi everyone 2010-11-20T06:49:26 Error323, is that your minimax bot on the main server or still the heuristic bot? 2010-11-20T06:49:46 Migi32: heuristics 2010-11-20T06:49:51 i gave up on minimax 2010-11-20T06:50:01 because of time mostly 2010-11-20T06:50:08 k. Doesn't give me good hopes as I'm working on mine now 2010-11-20T06:51:24 action generation and eval function 2010-11-20T06:51:33 are the huge issues 2010-11-20T06:51:37 *** Tobu has joined #aichallenge 2010-11-20T06:51:41 and of those, mostly eval function in my experience 2010-11-20T06:51:56 eval function is less important if you can search deeper 2010-11-20T06:52:13 and I think I've figured out a way to search really deep (like maybe 50 turns) that makes some simplified but not ridiculous assumptions 2010-11-20T06:53:30 like that it's always a bad move to take a planet if you could've taken it the turn before and did nothing 2010-11-20T06:53:36 *** tim_ has joined #aichallenge 2010-11-20T06:53:57 that's not always right but I hope it won't create too many distortions 2010-11-20T06:54:16 Migi32: O_O 2010-11-20T06:54:47 going 50 deep is impossible when probing all good actionsets 2010-11-20T06:55:13 *** tim_ has quit IRC (Remote host closed the connection) 2010-11-20T06:55:20 which means you'll base ur decisions on the assumption that the enemy can also only make those actions 2010-11-20T06:55:26 which will fail :> 2010-11-20T06:55:45 Migi32: i'm giving minimax one last go 2010-11-20T06:55:47 yes. I know I won't win but it's my best shot :P 2010-11-20T06:55:55 but it's striggling to make much of a tree 2010-11-20T06:56:15 I really hope I can get my move gen fast enough to add back in the minimax :/ 2010-11-20T06:56:38 i'm lucky to get past layer 4 if i only consider like 4-5 moves per turn 2010-11-20T06:56:48 and some of the nodes wont have that many possible moves left 2010-11-20T06:57:00 also, 50 may have been exaggerated but if you look at the first turn the branching factor is really high but if you look at turns 1 to 5(ish) the branching factor is 0 2010-11-20T06:57:12 on most maps 2010-11-20T06:57:34 first turn isn't usually so bad 2010-11-20T06:57:34 no its not 2010-11-20T06:57:41 as you don't have much to do when calculating future states 2010-11-20T06:57:58 mine bogs down when I own a lot of planets 2010-11-20T06:58:03 it's once all the planets are owned and fleets are flying everywhere that i have most issues 2010-11-20T06:58:18 you get ships each timestep, which you can distribute any way you want 2010-11-20T06:58:29 Error323, yes I know, I'm not streaming planets to the frontline. 2010-11-20T06:58:42 i don't think my new bot is going to need it 2010-11-20T06:58:46 just redirection 2010-11-20T06:59:29 streaming ships*, not planets :P 2010-11-20T06:59:43 i always make that mistake, i didn't even read it as planets :P 2010-11-20T06:59:45 now that would be awesome :P 2010-11-20T06:59:47 and I'm assuming the enemy won't stream ships to the frontline also, which is probably an even worse assumption 2010-11-20T07:00:27 yeah assuming things like that is lethal 2010-11-20T07:00:41 you need highly compound actions for evaluation 2010-11-20T07:01:03 (planet, turn) is quite compound :) 2010-11-20T07:01:17 oh well, we'll see how it goes. I don't have much time left to start something new now 2010-11-20T07:01:38 no i mean an action being a set of orders 2010-11-20T07:01:47 yeah 2010-11-20T07:02:00 a single action for me still has a single destination, then a vector of fleets 2010-11-20T07:02:24 and i make a single action per node child 2010-11-20T07:02:54 an action for me (that I feed to the minimax engine) is a sequence of moves for the next (max distance) turns 2010-11-20T07:03:32 but a single destination? 2010-11-20T07:03:36 my fleets can arrive whenever 2010-11-20T07:07:09 *** smloh1 has quit IRC (Quit: Page closed) 2010-11-20T07:07:10 lavalamp_: oh that's also interesting 2010-11-20T07:07:16 what's the big benefit of that? 2010-11-20T07:07:25 but you lose global scope 2010-11-20T07:07:40 i don't have time for those type of experiments :/ 2010-11-20T07:07:50 global scope? what do you mean? 2010-11-20T07:08:07 Migi32: benefit is that all my moves should make a lot of sense already 2010-11-20T07:08:21 so engine is just choosing between good options 2010-11-20T07:08:21 well iirc you only take nearby planets into account for your actions? 2010-11-20T07:08:28 me? 2010-11-20T07:08:30 no 2010-11-20T07:08:38 oh than i misunderstood 2010-11-20T07:08:39 srry 2010-11-20T07:08:44 lavalamp_, with that kind of moves, how are you going to implement an evaluation function? 2010-11-20T07:09:03 I have a search order in which disntance is a factor 2010-11-20T07:09:23 eval function looks at the state of the board after I apply both player's moves 2010-11-20T07:09:40 it's also used to rate the moves as I generate them 2010-11-20T07:09:58 ok. And then? Let it fight out for 200 turns and see who has more ships? 2010-11-20T07:11:40 no I just look at the board at that point and apply heuristics 2010-11-20T07:13:06 if it gets close enough to the end I just use final ship count 2010-11-20T07:15:56 yeah if I realllllly restrict my move gen I can still do depth 2 2010-11-20T07:16:15 but no point in that until my move gen stops getting sniped 2010-11-20T07:16:24 i was able to do 3.5K branching factor with 2 ply 2010-11-20T07:16:40 but only on my local machine :P 2010-11-20T07:16:42 *** medrimonia has quit IRC (Ping timeout: 245 seconds) 2010-11-20T07:17:03 and with alphabeta 2010-11-20T07:17:19 iirc 20K nodes per turn 2010-11-20T07:17:19 the problem is that that's not nearly enough depth unless your moves are really smart 2010-11-20T07:17:32 yeah you _need_ really smart move 2010-11-20T07:17:33 i get nothing like that :\ 2010-11-20T07:17:34 moves* 2010-11-20T07:18:01 I got 10k+ nps before I started adding stuff to my move gen 2010-11-20T07:18:12 Error323, is that 3.5K "actions" for player 1 and another 3.5K responses for player 2? 2010-11-20T07:18:18 Migi32: yes 2010-11-20T07:18:19 Top 10 players: bocsimacko(3386), Raschi(3250), _iouri_(3239), bix0r4ever(3149), felixcoto(3137), wagstaff(3129), medrimonia(3100), _Astek_(3097), protocolocon(3094), davidjliu(3069) 2010-11-20T07:18:35 1290255022441283000: Average of 48.15569423885912 nps 2010-11-20T07:18:35 fuuuuuuuk you astek :P 2010-11-20T07:18:36 1290255022441292000: 97.85% of time spent generating moves 2010-11-20T07:18:57 what the hell is that number in front 2010-11-20T07:18:58 from my log just now... 2010-11-20T07:19:06 time in nanoseconds 2010-11-20T07:19:31 yeah I don't know why I leave that in, it doesn't really help 2010-11-20T07:19:39 :P 2010-11-20T07:35:18 *** mceier has joined #aichallenge 2010-11-20T07:35:23 *** rangzen has joined #aichallenge 2010-11-20T07:53:46 hmm, i wish we had picked the maxRadius randomly now, but it's way too late to change that 2010-11-20T07:53:52 *** yasith has quit IRC (Ping timeout: 240 seconds) 2010-11-20T08:06:55 *** yasith has joined #aichallenge 2010-11-20T08:18:21 Top 10 players: bocsimacko(3375), Raschi(3298), _iouri_(3223), bix0r4ever(3157), _Astek_(3147), felixcoto(3128), wagstaff(3116), Accoun(3091), protocolocon(3088), medrimonia(3074) 2010-11-20T08:21:32 *** delt0r has joined #aichallenge 2010-11-20T08:21:43 *** delt0r___ has quit IRC (Ping timeout: 240 seconds) 2010-11-20T08:56:23 *** phxtrack has quit IRC (Ping timeout: 264 seconds) 2010-11-20T08:57:40 *** bhasker has joined #aichallenge 2010-11-20T08:59:08 morning 2010-11-20T09:03:38 morning :) 2010-11-20T09:04:09 bhasker: i haven't read all the logs... you figure out your peformance problems? 2010-11-20T09:04:13 howdy 2010-11-20T09:04:19 hi Error323 2010-11-20T09:04:29 nice work lately! 2010-11-20T09:05:57 oh yea 2010-11-20T09:06:06 i replaced all object lookups with tuples 2010-11-20T09:06:15 much much faster 2010-11-20T09:06:21 takes <2ms to update state for all planets 2010-11-20T09:06:23 hey dmj111 thanks :) 2010-11-20T09:06:43 e323 is that your heuristic bot 2010-11-20T09:06:46 or your minmax 2010-11-20T09:07:00 and dmj is your bot pure heuristic? 2010-11-20T09:07:18 bhasker: I always default to tuples for less than 5 items, at least at the start. 2010-11-20T09:07:24 bhasker are you using the entire 1 second per turn? 2010-11-20T09:07:28 nope 2010-11-20T09:07:38 for heuristics i barely need any time 2010-11-20T09:07:39 bhasker: it is not longer pure heurisitic, but it isn't really minimax either. 2010-11-20T09:07:57 i use my heuristics, and a bunch of hacks to explore a game tree. 2010-11-20T09:08:01 ah k 2010-11-20T09:08:31 so, i am limited to only seeing move that I think are good, even for the opponent, which isn't the best plan. 2010-11-20T09:08:34 i am still very confused on how to implement a game tree for something like this 2010-11-20T09:08:56 me too. 2010-11-20T09:08:58 its not that difficult 2010-11-20T09:08:58 so lets say i can make 5 simultaneous moves 2010-11-20T09:09:06 do i explore each move independently 2010-11-20T09:09:13 or do the 5 together make one state? 2010-11-20T09:09:29 into recursion: move a, move b, simulate, move a, move b, simulate 2010-11-20T09:09:44 but what about when i can do move a,b,c,d 2010-11-20T09:09:46 at the same time 2010-11-20T09:09:47 out of recursion: remove move b, restore state 2010-11-20T09:10:08 its not like i am limited to one move, 2010-11-20T09:10:46 at each node i pick a bunch of moves, then each child adds one of them 2010-11-20T09:10:54 then i do the same for the other player, and keep repeating 2010-11-20T09:11:04 *** yasith has quit IRC (Ping timeout: 252 seconds) 2010-11-20T09:11:08 hmm not sure i understand 2010-11-20T09:11:21 so lets say at top of the tree you could attack upto 3 planets at the same time 2010-11-20T09:11:28 a move has a single destination but a whole bunch of fleets that have different sources and different arrival times 2010-11-20T09:11:28 how does the game tree build out 2010-11-20T09:11:43 k so lets assume you have enough ships to attack 3 at the same time 2010-11-20T09:11:52 each child would have me take just one of the planets, and i'd also add a child to "do nothing" 2010-11-20T09:11:52 how will you explore the game tree for something like that 2010-11-20T09:12:13 then i'd let them make a move for each child, then for each of those children, see what i'd want to do then 2010-11-20T09:12:56 i don't model the opponent expanding at all 2010-11-20T09:12:59 bhasker: I think I am thinking the same thing as you. In that case, you have to check attacking none, (a), (b), (c), (a,b,) ..... (a,b,c).., So you have 8 states for just three planets. And that is a simple case. 2010-11-20T09:13:09 yea 2010-11-20T09:13:13 which is bad, but its too hard to have them expand and me not try to attack what they haven't already tried to claim 2010-11-20T09:13:34 dmj111: i tried that at first, but that blows out very fast 2010-11-20T09:13:45 yea but if you don't do that 2010-11-20T09:13:50 and even picking a, b st a and b are valid together seems tricky to me 2010-11-20T09:13:52 won't your solution be always suboptimal 2010-11-20T09:14:09 like if you could really capture two at same time 2010-11-20T09:14:11 yep, that is the problem. 2010-11-20T09:14:14 no, cause the game tree will have (a, nothing, b) 2010-11-20T09:14:33 hmm but what if capturing a or b is good 2010-11-20T09:14:36 but the combination is not 2010-11-20T09:14:39 because you overextend 2010-11-20T09:14:51 (a, nothing, nothing), (b, nothing, nothing) 2010-11-20T09:15:01 hmm 2010-11-20T09:15:16 i'm not sure minimax picks that up though properly now that i think about it 2010-11-20T09:15:31 but otherwise, how do you pick a,b s o that a and b are feasible together? 2010-11-20T09:15:53 so what i am doing for 1 ply max is basically looking at all available ships 2010-11-20T09:16:07 and i build out all combinations which can be executed simultaneously 2010-11-20T09:16:13 basically i build a list of all single planet orders 2010-11-20T09:16:17 and combine them recursively 2010-11-20T09:16:25 and now i have alist of combined orders 2010-11-20T09:16:27 one option is to pick say, starting at best planet 1, can i take it? if so add move, can i take 1 and 2? if so add move, move onto 2, can i take 2? etc. 2010-11-20T09:16:29 which i then score 2010-11-20T09:16:56 antimatroid1: that is basically what I did. 2010-11-20T09:17:02 so then i can get a score for, a,b,c, (a,b),(b,c) etc 2010-11-20T09:17:18 yup thats what i am going to do now 2010-11-20T09:17:24 *** Zoost has joined #aichallenge 2010-11-20T09:17:37 yeah, i think i might do that now 2010-11-20T09:17:38 after considering each planet, if i can take it, I launch the ships, update my game state, and loop again, like it was the beginning of the turn. 2010-11-20T09:18:04 sorry, I break after attacking to update the state, and then start all over. 2010-11-20T09:18:07 dmj111: do you still have them attacking you with their ships each turn? 2010-11-20T09:18:07 dmj what i do is 2010-11-20T09:18:15 how do you do that and work out which neutrals can safely be taken 2010-11-20T09:18:15 lets say the move is 2010-11-20T09:18:22 Top 10 players: bocsimacko(3382), Raschi(3273), _iouri_(3250), _Astek_(3196), bix0r4ever(3164), felixcoto(3139), wagstaff(3128), protocolocon(3088), medrimonia(3086), george(3064) 2010-11-20T09:18:30 launch from a at time t, b at t+2,c at t+3 2010-11-20T09:18:51 i just change it so that all of them launch at t and just resolve for end state 2010-11-20T09:18:52 i can't work out how to do both of those things together, without literally generating states, having them attack, making planets safe, removing their moves, generating states, then making my moves 2010-11-20T09:18:55 which is way too much 2010-11-20T09:19:04 antimatroid1: i just caculated how long it would take me to get the neutral, how long it would take them to take it away, and see if it is a net benefit. 2010-11-20T09:20:01 do you guys think that is a good idea to assume that all ships launched at t 2010-11-20T09:20:06 and resolve? 2010-11-20T09:20:14 what do you mean? 2010-11-20T09:20:22 so lets say the complete move was 2010-11-20T09:20:30 my fleets have tadded, tsent and tarrive :) 2010-11-20T09:20:35 launch from planet A at t, planet B at t+2 and planet C at t+3 2010-11-20T09:20:42 so that they all reach at the same time 2010-11-20T09:20:48 i change the order 2010-11-20T09:20:49 although, i normally just use the tsent for neutrals, i'll just defend/attack as soon as possible 2010-11-20T09:20:55 so that all launches go from a,b,c at time t 2010-11-20T09:21:06 since i don't take into account future ships 2010-11-20T09:21:17 usually for an order to be executed i do have ships to fill the order 2010-11-20T09:21:36 so i just artificially increase distance between B and the target to be t instead of t+2 2010-11-20T09:21:43 oops t-2 2010-11-20T09:21:54 and C to be t instead of t-3 2010-11-20T09:21:57 i have done that, but wonder if it is good or not. 2010-11-20T09:22:13 if you are not taking future ships into account 2010-11-20T09:22:17 i think its equivalent 2010-11-20T09:22:28 If I have to launch fleets in the future, I actually put fleets with negative ships landing at the source planet at the launch turn, and ships landing at the dest at the right time. 2010-11-20T09:22:33 but if you are executing orders assuming you wil have some ships on a planet at time t +3 2010-11-20T09:22:35 then its hard 2010-11-20T09:22:48 oh, the other thing about doing the game tree my way, cause it doesn't really take into account "when" a move is made, it will also consider your ability to make a move based on your first move, if that makes sense at all 2010-11-20T09:22:59 *** yasith has joined #aichallenge 2010-11-20T09:23:04 so you might be able to pick b, then a with benefits from b, but not a and b to begin with 2010-11-20T09:23:05 ahh, and here is a neutral fail for me: Components.classes["@mozilla.org/login-manager;1"].getService( 2010-11-20T09:23:06 Components.interfaces.nsILoginManager); 2010-11-20T09:23:09 oops. 2010-11-20T09:23:25 http://www.ai-contest.com/visualizer.php?game_id=8078226 2010-11-20T09:23:32 turn 39 2010-11-20T09:24:08 maybe turn 27 too.. 2010-11-20T09:24:24 that was an expensive neutral for the growth 2010-11-20T09:24:33 are you just removing infeasible outgoing fleets? 2010-11-20T09:25:02 i was losing a lot of games by not taking neutrals, so I took out the safeties. 2010-11-20T09:25:07 i considered trying to remove the entire sections of the tree that become invalidated, but then i figured it might actually then continue trying to make the moves to that planet, whilst cancelling the "bad" fleets already sent 2010-11-20T09:25:43 i am not thinking about infeasible fleets, I belive. 2010-11-20T09:25:46 although, now that i think about it, the way i get moves, i'm not removing the invalid fleets from there, hmmm 2010-11-20T09:26:20 djm111: when you model both players, it's possible that you decided to send x ships from a planet at time t, but they then sent a fleet giving that planet less than x ships at time t 2010-11-20T09:26:44 working out how to handle that has probably been my biggest issue 2010-11-20T09:27:11 right now i have queues for outgoing fleets, and i just delete all the outgoing fleets that become infeasible 2010-11-20T09:27:46 ahh, yes. I don't worry about that, though I should. 2010-11-20T09:28:17 i'm beginning to wonder if it's more expensive to handle than it's worth 2010-11-20T09:28:21 In fact, I tried worrying about it, and had code in place to cover for it, but it just mucked everything up. 2010-11-20T09:28:39 it probably doesn't happen very often 2010-11-20T09:28:40 I _hope_ since I mostly attack stuff I can get to first, that it won't happen often. 2010-11-20T09:28:44 i really wonder what is mega1's secret sauce 2010-11-20T09:29:05 we'll find out in a bit over a week 2010-11-20T09:29:08 yes, it will be fun to learn what he is doing. 2010-11-20T09:29:29 i just hope its something really fancy and not like a "now why didn't i think about that" 2010-11-20T09:29:35 cause then i will feel really dumb 2010-11-20T09:30:12 i don't think it's got anything overly fancy from what he's said 2010-11-20T09:30:13 hmmm. I hear ya. But if it is something easy, maybe I'll be able to use it later for something else ;) 2010-11-20T09:30:15 but it could be a cover 2010-11-20T09:30:40 i'm pretty keen to see his nash bot too 2010-11-20T09:30:57 time to debug some neutral code, I guess.... 2010-11-20T09:31:06 have fun dmj 2010-11-20T09:31:21 i think that neutral was simply a bad idea, its cost to recover was almost 23 turns 2010-11-20T09:31:29 actually 30 including travel time 2010-11-20T09:32:02 i have reverted to finalgrowthme-finalgrowthtehm for evaluation 2010-11-20T09:32:03 yes.... i agree. i had code to think like that before, but it never took anything, so I was always losing solitaire games. 2010-11-20T09:32:20 it seems to stupid moves when i estimate the final number of ships for each player 2010-11-20T09:32:22 i took it out, and it does better sometimes, but this was a obvious dumb move. 2010-11-20T09:32:24 antimatroid i use a combination of two things 2010-11-20T09:32:28 ship count at end 2010-11-20T09:32:39 and growth at end state 2010-11-20T09:32:47 and initially the scoring will weight growth more 2010-11-20T09:32:55 as time goes by it will start weighting having ships more 2010-11-20T09:33:02 clever. 2010-11-20T09:33:09 that's probably a pretty good idea 2010-11-20T09:33:09 so towards then end my eval becomes a pure ship count 2010-11-20T09:33:10 mine is unaware of the turn still. 2010-11-20T09:33:31 so as game progresses it will favour states which give me higher ship counts 2010-11-20T09:33:35 than new growth 2010-11-20T09:33:43 that is the most useful idea I have stolen from here for a while :) 2010-11-20T09:33:50 i wont expand if my final growth is more than double theirs 2010-11-20T09:34:07 otherwise i just take forever to take people out 2010-11-20T09:34:13 though i realized i am evaluating beyond 200 turns 2010-11-20T09:34:21 since i was just checking the last state for each planet 2010-11-20T09:34:23 i need to fix that 2010-11-20T09:34:37 yeah, that's not hard to fix though 2010-11-20T09:34:39 though my bot that scores orders does worse overall 2010-11-20T09:34:44 i used to have that, don't at the moment but plan to re add it 2010-11-20T09:35:11 i am just coding in my order Combining logic 2010-11-20T09:35:15 to score combined orders 2010-11-20T09:35:22 lets see how it does after that 2010-11-20T09:35:31 right now its pretty dumb 2010-11-20T09:35:37 execute order with highest score 2010-11-20T09:35:49 then keep executing lower scored orders if i have neough ships 2010-11-20T09:35:55 so its a suboptimal solution 2010-11-20T09:36:54 hmmm, i'm going to try stripping out some of the not overly importnat computationally expensive stuff and see if it still does about the same 2010-11-20T09:37:56 i changed to just having states maxdist+1 for each planet and only making moves up to maxdist 2010-11-20T09:38:11 i figure the games often changes too much for those old moves to still be valid anyway 2010-11-20T09:38:18 and it halved the number of states almost 2010-11-20T09:38:49 bhasker: you're using 2*maxD + 1 yeah? 2010-11-20T09:38:57 yes 2010-11-20T09:39:43 the only reason i was doing till maxD+1 2010-11-20T09:39:47 oops 2*maxD+1 2010-11-20T09:39:53 was to account for ships in flight 2010-11-20T09:39:55 between planets 2010-11-20T09:40:01 yeah, i did that 2010-11-20T09:40:05 when i was looking far into the future 2010-11-20T09:40:10 i actually don't do that anymore 2010-11-20T09:40:13 but i cap finishing times for new fleets at maxD now 2010-11-20T09:40:15 i have an exponential drop off 2010-11-20T09:40:20 i don't even do that 2010-11-20T09:40:26 i stop at like 5 turns or something 2010-11-20T09:40:40 i weight the fleets at next 3-4 turns at full value 2010-11-20T09:40:41 is that for adding fleets or them finishing? 2010-11-20T09:40:44 adding 2010-11-20T09:40:58 and then i just exponentially drop off the new fleets 2010-11-20T09:41:02 from growth etc 2010-11-20T09:41:11 yeah, i use a prioirity queue and pick the ones that were added soonest, but keep using any fleets at any time 2010-11-20T09:41:27 taking too much of future into account just stalls my bot completely 2010-11-20T09:41:55 in fact now i am going with a zero defense code except when its under attack and when its in the initial stages of growth 2010-11-20T09:42:14 so it will launch ships even if it loses the planet it launched from due to a counter attack 2010-11-20T09:42:42 though with a 1 ply scoring its not optimal 2010-11-20T09:43:00 basically my available ships calculation is something like 2010-11-20T09:43:17 if planetUnderAttack: calculate Available ships else: availableShips = p.NumShips 2010-11-20T09:43:41 though it doesn't work out very well in all circumstances 2010-11-20T09:43:53 i have also been thinking of having multiple availability lists 2010-11-20T09:44:03 one for having ships that are free to attack neutrals 2010-11-20T09:44:20 and one for ships that are free to defend/attack mine/enemy plantes 2010-11-20T09:44:30 the first one will be calculated more strictly than the other 2010-11-20T09:44:41 but not sure how to do it 2010-11-20T09:47:23 hmm, i think i should be able to make it way more efficient than it is now without actually losing much information, bah, back in a few hours :) 2010-11-20T09:59:09 antimatroid1: funny enough... _one_ of my current problems is that I ignore that the enemy can stop my plans.... 2010-11-20T10:02:20 *** Blkt has joined #aichallenge 2010-11-20T10:03:15 *** yasith_ has joined #aichallenge 2010-11-20T10:05:23 *** yasith has quit IRC (Ping timeout: 264 seconds) 2010-11-20T10:07:40 *** medrimonia has joined #aichallenge 2010-11-20T10:07:51 i would like to see some bot, that takes a bit longer but comes with perfect gameplay 2010-11-20T10:07:54 i wonder if thats possible 2010-11-20T10:08:14 or close to it 2010-11-20T10:09:40 good day everyone 2010-11-20T10:10:17 hi 2010-11-20T10:12:56 *** medrimonia has quit IRC (Quit: Leaving.) 2010-11-20T10:13:29 *** medrimonia has joined #aichallenge 2010-11-20T10:13:33 Hello 2010-11-20T10:13:40 hey 2010-11-20T10:13:55 medrimonia: are you on linux by any chance? :P 2010-11-20T10:14:27 oh grmbl nvm ur using python 2010-11-20T10:14:45 mega1: got results yet? 2010-11-20T10:15:08 In fact I'm on both 2010-11-20T10:15:13 Linux and windows 2010-11-20T10:15:32 @tcp 2010-11-20T10:15:33 medrimonia: tcp = (#1) http://www.benzedrine.cx/planetwars/, or (#2) http://72.44.46.68/, or (#3) 98.247.248.39:995 (zeroviz.us), or (#4) http://tcp.zeroviz.us/, or (#5) http://tcp2.zeroviz.us/ (crazy!) 2010-11-20T10:15:49 bhasker: perfect gameplay would be really really hard 2010-11-20T10:15:54 wanted to switch binaries 2010-11-20T10:15:55 i'm not even close to doing anything near it 2010-11-20T10:15:57 i don't think anyone is 2010-11-20T10:16:01 but with python thats not possible 2010-11-20T10:17:57 gosh I gotta fix my new bot 2010-11-20T10:18:04 http://72.44.46.68/canvas?game_id=1106630 2010-11-20T10:18:23 Top 10 players: bocsimacko(3412), _iouri_(3240), Raschi(3238), _Astek_(3179), bix0r4ever(3168), felixcoto(3157), wagstaff(3129), protocolocon(3101), george(3065), luksian(3054) 2010-11-20T10:18:50 amazingly stupid 2010-11-20T10:19:01 Mistmanov: ur comming back :D 2010-11-20T10:19:06 #66 now 2010-11-20T10:19:10 nice 2010-11-20T10:19:37 dmj111: I'm not sure what I'm doing does an amazing job at that 2010-11-20T10:22:02 i'd be keen for another contest almost straight away 2010-11-20T10:22:12 but i'm on holidays, i assume that'd almost kill some people? 2010-11-20T10:23:08 i am looking forward to the end of this. my other hobbies have suffered. 2010-11-20T10:23:39 but it has been fun 2010-11-20T10:23:42 and, I wish it ended when I was doing better ;-) 2010-11-20T10:23:58 but learning how bocsimacko's bot works will be interesting. 2010-11-20T10:24:19 i haven't had any other hobbies while this is on aha 2010-11-20T10:26:38 Gosh I'm starting a loosing streak! 2010-11-20T10:27:25 *** krokkrok has joined #aichallenge 2010-11-20T10:28:48 Error323: You resubmit? 2010-11-20T10:28:52 medrimonia, bhasker : have you guys used pdb very much? I should have used it sooner, but have been depending on logging instead. pdb _is_ pretty nice. 2010-11-20T10:29:03 hmm not using pdb 2010-11-20T10:29:05 (not to leave out any other python users here) 2010-11-20T10:29:06 medrimonia: yeah... but two lossess >_< 2010-11-20T10:29:11 though its doing much better on tcp 2010-11-20T10:29:21 so i figured it'll do better there heh 2010-11-20T10:29:22 i just rely on extensive logging right now 2010-11-20T10:29:32 that is what I was doing too. 2010-11-20T10:29:44 i had a frustrating bug that logging just wasn't handling. 2010-11-20T10:29:55 hmm i haven't run into something like that yet 2010-11-20T10:30:01 since my bot is pretty simple 2010-11-20T10:30:07 i am not doing minmax etc 2010-11-20T10:30:19 so its a pretty simple set of steps to identify issues 2010-11-20T10:30:31 yeah. that is true... it was in the code where I was considering enemy responses, so I was losing track of who was doing what. 2010-11-20T10:30:51 though i might use it to track my stupid invalid orders code 2010-11-20T10:31:14 *** virdo has joined #aichallenge 2010-11-20T10:31:24 but in pdb, you have access to everything, and can dig through objects, call functions, etc. 2010-11-20T10:31:28 s/code/bug 2010-11-20T10:31:35 yea i have used it a little before 2010-11-20T10:31:57 I rarely do, so that is why I am a little spun up about it. usually isn't necessary. 2010-11-20T10:34:08 guys is the number of ships on a neutral ever smaller than its growthrate? 2010-11-20T10:35:22 it has happened. 2010-11-20T10:35:27 not sure about the new generator though. 2010-11-20T10:37:05 dmj111: I don't even know what's pdb^^^ 2010-11-20T10:37:19 I'm still a rookie in programmation so... 2010-11-20T10:38:25 ahhh, I wouldn't have guessed. 2010-11-20T10:38:30 it is the python debugger. 2010-11-20T10:38:55 *** mceier has quit IRC (Quit: leaving) 2010-11-20T10:38:59 i tend to use logging instead most of the time, but sometimes it is useful to have access to stuff whie it is running. 2010-11-20T10:39:46 Error323: it can start like that 2010-11-20T10:39:55 and it can most certainly happen from someone attacking it 2010-11-20T10:40:26 mmmm ok 2010-11-20T10:40:38 http://ai-contest.com/visualizer.php?game_id=8060426 2010-11-20T10:40:40 there ya go :P 2010-11-20T10:41:16 *** krokkrok has quit IRC (Quit: Page closed) 2010-11-20T10:42:07 * antimatroid1 has never used a proper debugger either :\ 2010-11-20T10:42:23 well it can make some things go faster 2010-11-20T10:42:27 but also slower :P 2010-11-20T10:42:30 i sometimes use it 2010-11-20T10:42:44 but during this project its not really possible 2010-11-20T10:43:07 ha yeah, so much information being used 2010-11-20T10:44:32 Does someone knows a good tcp client on windows? 2010-11-20T10:44:46 what's wrong with jbotmanager? 2010-11-20T10:45:15 i'm tossing up about throwing out removal of fleets now 2010-11-20T10:49:34 fuuuuuuuuck 2010-11-20T10:49:37 tcp says NOTHING :( 2010-11-20T10:49:43 i just lost from 92 2010-11-20T10:49:53 latest version is teh suck it seems 2010-11-20T10:49:54 bah 2010-11-20T10:50:36 well 2010-11-20T10:50:42 guess its too early to say 2010-11-20T10:50:44 but meh 2010-11-20T10:51:07 *** iFire has quit IRC (Remote host closed the connection) 2010-11-20T10:51:31 *** iFire has joined #aichallenge 2010-11-20T10:55:51 Error323: your old version against b0.70: 5/0/90 (win/draw/loss) 2010-11-20T10:56:04 new version: 8/1/91 2010-11-20T10:57:44 hmmm I can't get Bot Manager to work 2010-11-20T10:57:57 when I click on start, nothing happen 2010-11-20T10:58:01 mega1: thanks! 2010-11-20T10:58:39 no prob 2010-11-20T10:58:59 reddit game adds are the worst 2010-11-20T11:00:11 mega1: got time to test one more please? 2010-11-20T11:00:15 :> 2010-11-20T11:01:05 Error323: yes, but I can only start it in an hour or so 2010-11-20T11:01:15 that's ok 2010-11-20T11:01:19 thanks 2010-11-20T11:01:28 medrimonia, does it have a console window attached (some times it takes a while to start a game and without a window it looks like it is doing aboslutly nothing) 2010-11-20T11:02:04 I haven't any console windows 2010-11-20T11:02:55 run the manager from a window and you get output 2010-11-20T11:04:27 hum thanks I think I know the error 2010-11-20T11:05:53 can't find the specified file... 2010-11-20T11:07:19 *** Utkarsh has joined #aichallenge 2010-11-20T11:09:17 *** davidjliu has joined #aichallenge 2010-11-20T11:09:32 how do you write a python bot for JbotManager on windows? 2010-11-20T11:09:40 python adress or... 2010-11-20T11:13:07 I have the following message: "Cannot run "MyBot.py": createProcess error=193" 2010-11-20T11:15:35 *** bhasker has quit IRC (Read error: Connection reset by peer) 2010-11-20T11:16:09 *** bhasker has joined #aichallenge 2010-11-20T11:17:33 *** bhasker_ has joined #aichallenge 2010-11-20T11:17:33 *** bhasker has quit IRC (Read error: Connection reset by peer) 2010-11-20T11:18:22 Top 10 players: bocsimacko(3432), _Astek_(3291), _iouri_(3253), Raschi(3248), bix0r4ever(3176), wagstaff(3162), Manwe56(3132), felixcoto(3121), protocolocon(3101), george(3064) 2010-11-20T11:20:16 hum does someone know how to add a python bot on windows to jbotmanager? 2010-11-20T11:20:43 * Error323 linux only 2010-11-20T11:20:57 I should be too 2010-11-20T11:21:14 But my best computer is only on windows now... crap 2010-11-20T11:22:14 it should just be adding "python MyBot.py" in the bot tab 2010-11-20T11:22:24 doesn't work =( 2010-11-20T11:22:28 http://72.44.46.68/canvas?game_id=1107134 omg the fail 2010-11-20T11:22:32 is python in your path? 2010-11-20T11:23:01 try "fulpath\python MyPot.py" 2010-11-20T11:23:01 hum the MyBot is in my path 2010-11-20T11:23:32 I'll try 2010-11-20T11:26:43 what do you guys do when you make moves and add them to your states? do you recalculate everything afterwards? or update the states you change, and if updating, what happens when a move has multiple fleets leaving the same source at the same or different times? 2010-11-20T11:27:24 grrr hate windows.... 2010-11-20T11:28:11 does the python starter_pack work out of the box? 2010-11-20T11:29:43 I guess I'm about to find out 2010-11-20T11:29:47 Error323: yay, the bailing out thing is again working as intended :p 2010-11-20T11:30:07 2 days ago, my bot would've just sat still when it's being attacked.. 2010-11-20T11:30:16 yeah Mistmanov nice work 2010-11-20T11:30:23 but i consider this a major flaw on my end 2010-11-20T11:30:29 ofcourse :p 2010-11-20T11:30:43 capping neutrals in stupid places. 2010-11-20T11:30:43 Mjothvitnir: I can't even play a replay... 2010-11-20T11:35:45 I can start the python bot but it gives me an error "File"..." line X in main[;] current_line = raw_input() [;] EOFError: EOF when reading a line 2010-11-20T11:36:17 I would guess it is not happy about the LF/NL/CR thing 2010-11-20T11:37:16 Hmmm still trying but... 2010-11-20T11:39:03 Did you fire up the jbotmanager from a dos box? 2010-11-20T11:39:08 Hmmm maybe I'm in python 3.1.2 too 2010-11-20T11:39:09 Yes 2010-11-20T11:39:34 does it give you an error when you try to use the python bot in the combat tab? 2010-11-20T11:41:17 looks like OpenOffice uses 2.6.1 (that is where I'm get python from) :) 2010-11-20T11:42:13 no, it's not giving me anything in the dos box when I launch a combat 2010-11-20T11:42:46 not even an error 2010-11-20T11:43:01 and nothing happens combatwise, weird 2010-11-20T11:43:32 yes^^ 2010-11-20T11:43:40 virtualbox 2010-11-20T11:43:51 it's how I get things done on windows 2010-11-20T11:44:21 when combat happens I get nothing until a winner is decided then in the window it tells me who won... wait what version of the manager? 2010-11-20T11:44:57 I just downloaded it an hour ago 2010-11-20T11:45:21 But I think I'll keep working with my linux computer 2010-11-20T11:45:35 Even if it's not as effective 2010-11-20T11:45:39 damn... there's something reaaaaaaaaally stupid going on with my sniping 2010-11-20T11:45:41 >_< 2010-11-20T11:46:36 http://72.44.46.68/canvas?game_id=1107201 2010-11-20T11:46:40 me too.... 2010-11-20T11:46:56 I have to check my logs 2010-11-20T11:47:03 i lost while winning from Mistmanov and now from hexist: http://ai-contest.com/visualizer.php?game_id=8091975 2010-11-20T11:47:04 *** McLeopold has joined #aichallenge 2010-11-20T11:49:46 madrimonia, put random characters in as a bot command and let me know if you get error ouput when you try to use it 2010-11-20T11:50:15 oops, bad fingers, medrimonia 2010-11-20T11:50:25 no problem 2010-11-20T11:50:55 no errors with random name 2010-11-20T11:51:31 delt0r ?? 2010-11-20T11:52:08 whois barontrozo 2010-11-20T11:52:14 cuz he's going epic 2010-11-20T11:52:14 @delt0r 2010-11-20T11:52:15 Mjothvitnir: User error -- Replace user. 2010-11-20T11:52:32 @seen delt0r 2010-11-20T11:52:32 Mjothvitnir: delt0r was last seen in #aichallenge 2 weeks, 4 days, 21 hours, 6 minutes, and 53 seconds ago: demerzel__: thats what i I tell my boss 2010-11-20T11:53:02 I wonder.. 2010-11-20T11:53:06 @seen Mistmanov 2010-11-20T11:53:06 Mistmanov: Mistmanov was last seen in #aichallenge 3 seconds ago: I wonder.. 2010-11-20T11:53:10 hmm 2010-11-20T11:53:13 no infinite loop :( 2010-11-20T11:55:35 weird, I get stderr & stdout "type" output in my window, but you don't, hmm 2010-11-20T11:56:35 *** mega1 has quit IRC (Ping timeout: 255 seconds) 2010-11-20T11:57:17 did you start it by typing "jbotmanager.jar" or "java -jar jbotmanager.jar" 2010-11-20T11:57:43 java -jar jbotmanager.jar 2010-11-20T11:58:40 dang. for me the first one returns to the commandline immediately (and therefor gives no error), but the second hangs around waiting to give/get in/output 2010-11-20T12:00:51 so to be sure, your cursor is just floating there below the java command, having done/printed nothing 2010-11-20T12:01:24 hum I try 2010-11-20T12:01:44 *** amriedle has joined #aichallenge 2010-11-20T12:01:54 yes right 2010-11-20T12:02:18 I'll try with linux that'll be easier^^^ 2010-11-20T12:02:32 I am stumped and delt0r and delt0r_ seem to not be here 2010-11-20T12:02:36 *** mceier has joined #aichallenge 2010-11-20T12:07:35 *** McLeopold has quit IRC (Ping timeout: 245 seconds) 2010-11-20T12:12:22 ooo I really hope fixing this bug doesn't make my bot suck... 2010-11-20T12:14:30 depends on if it is a bug or a feature. 2010-11-20T12:16:37 * lavalamp_ cries 2010-11-20T12:16:51 sounds like it was a feature 2010-11-20T12:17:37 *** McLeopold has joined #aichallenge 2010-11-20T12:17:55 @rankings 2010-11-20T12:17:57 McLeopold: Top 10 players: bocsimacko(3409), _Astek_(3297), Raschi(3242), _iouri_(3232), bix0r4ever(3189), wagstaff(3151), protocolocon(3098), felixcoto(3080), BaronTrozo(3062), george(3061) 2010-11-20T12:18:22 Top 10 players: bocsimacko(3409), _Astek_(3297), Raschi(3242), _iouri_(3232), bix0r4ever(3189), wagstaff(3151), protocolocon(3098), felixcoto(3080), BaronTrozo(3062), george(3061) 2010-11-20T12:20:16 amstan janzert anybody here? 2010-11-20T12:20:24 j3camero: yo? 2010-11-20T12:20:27 Might be too early in the day haha. 2010-11-20T12:20:30 Oh maybe not 2010-11-20T12:20:37 Sup amstan. 2010-11-20T12:20:42 nm 2010-11-20T12:20:43 wait.. 2010-11-20T12:20:44 I wanna learn Django today. 2010-11-20T12:20:47 when's launch day? 2010-11-20T12:20:53 j3camero: oh ya, good choice 2010-11-20T12:20:58 Launch day for what? 2010-11-20T12:21:10 j3camero: end* day 2010-11-20T12:21:15 next week, good 2010-11-20T12:21:23 Yeah 7 days left I believe. 2010-11-20T12:21:34 yikes 2010-11-20T12:21:40 @time 2010-11-20T12:21:40 Mjothvitnir: 12:21 PM, November 20, 2010 2010-11-20T12:21:48 * antimatroid1 would be down for another contest almost immediately :P 2010-11-20T12:21:53 hint hint j3camero 2010-11-20T12:23:47 Yeah man so would I 2010-11-20T12:23:58 But it takes so much work to get one off the ground. 2010-11-20T12:24:11 I am hoping that the move to open source will reduce the time to market on the next one. 2010-11-20T12:26:35 *** Appleman1234 has quit IRC (Ping timeout: 276 seconds) 2010-11-20T12:27:08 amstan: I am liking the ideas you have in your framework forum post. 2010-11-20T12:27:18 Like I said, I think I will learn Django today. 2010-11-20T12:27:25 See what that's all about. 2010-11-20T12:27:58 *** Appleman1234 has joined #aichallenge 2010-11-20T12:28:17 Appleman1234 how are you liking the contest? 2010-11-20T12:28:29 *** Gromgull has joined #aichallenge 2010-11-20T12:28:30 j3camero: I'd like to be a contributor to the framework 2010-11-20T12:28:53 Hi all - am I right in assuming that for a java contribution all classes must be in the default package? 2010-11-20T12:28:54 I'm willing to do anything that doesn't prevent me from competing. :) 2010-11-20T12:28:59 j3camero, haven't had the chance to participate as much as I would like, real life too much of a blocker 2010-11-20T12:29:20 I hear that. 2010-11-20T12:29:23 *** mega1 has joined #aichallenge 2010-11-20T12:30:10 yeah, i'm happy to contribute, but i think i'd be more of a hindrance than help with the framework 2010-11-20T12:30:17 j3camero, I also found tron a more approachable task,but that is probably because when ever I see galaga I think of netrek 2010-11-20T12:30:20 *** rangzen has quit IRC (Ping timeout: 276 seconds) 2010-11-20T12:30:29 McLeopold you can contribute as much code as you want. It won't stop you from competing. The only thing that stops you from competing is if you've got root on a contest server, or you're a forum admin (not mod) or have some other kind of priviledged access to the contest. 2010-11-20T12:30:40 j3camero: Specifically, I'd like to focus on an open-source tcp server. 2010-11-20T12:31:09 I'm envisioning something that can be setup as a server, like usual, or run as a peer-to-peer. 2010-11-20T12:31:09 McLeopold: I forget, are you the one running the "main" TCP server right now? 2010-11-20T12:31:18 no, the secondary 2010-11-20T12:31:23 Ah okay. 2010-11-20T12:31:37 it's much nicer though 2010-11-20T12:31:45 also the stability issues regarding the game server (which are probably fixed now) were another factor 2010-11-20T12:31:47 until you see the code... 2010-11-20T12:32:13 Yeah there has been an idea to set up an official TCP server next time, or at least have the code for one ready-to-go in the code repo. I think it's a good idea. If you want to become the owner of that initiative that would be great. 2010-11-20T12:32:27 alright. 2010-11-20T12:33:01 official tcp would be ideal i think 2010-11-20T12:33:17 I think the move to github will make this process even easier. Somebody like McLeopold could run his own github repo and also send us push requests. 2010-11-20T12:33:20 yea, something that anyone can download and get running right away 2010-11-20T12:33:33 it would probably be worth asking dhartmei if he wanted to help, although i have the feeling he might be lucky to have that lifted off his chest 2010-11-20T12:33:52 lucky == happy 2010-11-20T12:33:55 I really should talk to him. I'd hate to steal his thunder. 2010-11-20T12:33:57 My original aim was to make the whole contest software package like that. Anyone with an Ubuntu laptop should be able to build and launch a contest instance in minutes, in my opinion. 2010-11-20T12:34:06 I'm just standing on his shoulders 2010-11-20T12:34:22 j3camero: yes, I agree 2010-11-20T12:34:28 i'm not sure on your idea of anyone being able to put up a new contest 2010-11-20T12:34:43 you'd need LOTS of servers, and there's going to be so many contests people would be spread out everywhere 2010-11-20T12:34:49 j3camero: if I do the tcp server right, 2 or more could get a tcp server going and connect them 2010-11-20T12:34:58 distributed tcp 2010-11-20T12:34:59 and, probably, to easily put up a new game / engine. 2010-11-20T12:35:02 I tried this time around. It's way easier to set up a PW instance than it was to set up a Tron isntance, that's for sure. But it's not as easy as I would like. 2010-11-20T12:36:02 would be interesting to see if a general framework is developed 2010-11-20T12:36:18 for deploying judgement servers, testing servers, tcp servers etc 2010-11-20T12:36:28 j3camero: have you looked at anywhere for funding recently? i would expect the number of competitors to continue increasing, so the number of servers is going to have to increase 2010-11-20T12:36:33 and some prize money would be awesome 2010-11-20T12:36:59 antimatroid1: you make a good point, that we might fragment the community by accident. I was thinking it would make things easier for developers to test their changes, and create new game engines. Development would be a lot faster if it was easy to launch a new instance on a test server to see the changes live in action. 2010-11-20T12:37:09 omg i take back all my complaints about how old the ghc version we are using is. i just found out about somebody whose university uses ghc 5 2010-11-20T12:37:18 i'm so glad we aren't them 2010-11-20T12:37:23 :) 2010-11-20T12:38:00 *** fmeyer has joined #aichallenge 2010-11-20T12:38:06 j3camero: i like the idea of having old contest servers available, and possibly even multiple servers with different maps sets for each "kind of game" 2010-11-20T12:38:28 but the thought of anyone adding an extra contest seems like it'd just blow out of control till nobody knew what to "enter" anymore 2010-11-20T12:38:30 antimatroid1: Yes funding could be an issue. I am hoping that the Google people will be receptive to giving this thing serious funding. I would like to have, say, 10 servers and maybe $10,000 in prize money. That would be the ideal scenario. 2010-11-20T12:39:09 that would be awesome, $10k would be way over what I would have originally been thinking 2010-11-20T12:39:23 i don't want to have more than one contest active at any single time 2010-11-20T12:39:34 *** boegel has joined #aichallenge 2010-11-20T12:39:42 *** bhasker_ has quit IRC (Ping timeout: 250 seconds) 2010-11-20T12:39:44 jmcarthur: yeah, that's sort of how i feel 2010-11-20T12:39:48 in fact, i kind of like that this isn't a year-round thing at all 2010-11-20T12:39:53 antimatroid1: if we let people submit new contests (say, by subclassing a Game class or something), then we would have to have a ranking of contests. You would not have to wade through 5000 shitty contests to find the good ones. They would be constantly ranked by participation. 2010-11-20T12:39:57 that there are gaps in between makes it more like an *event* 2010-11-20T12:40:19 events build hype. continuous contests build project euler 2010-11-20T12:40:24 *** foucist has joined #aichallenge 2010-11-20T12:40:41 is anyone doing evolutionary computation? 2010-11-20T12:40:41 Yeah I agree. 2010-11-20T12:40:56 2 a year would be reasonable 2010-11-20T12:41:05 3 months has been very grueling 2010-11-20T12:41:13 ranking by participation doesn't seem to solve the problem in my mind 2010-11-20T12:41:25 *** fmeyer has quit IRC (Read error: Connection reset by peer) 2010-11-20T12:41:34 anyone here using the zeroviz server actively? 2010-11-20T12:41:45 for some reason, it seems to think I'm issueing invalid commands 2010-11-20T12:41:46 not at the moment sorry, i'm about to head to bed 2010-11-20T12:41:48 i'm of the opinion that the contests should be thought up and organized centrally, not introduced ad hoc 2010-11-20T12:41:49 which I'm not... 2010-11-20T12:42:02 There's a middle ground I think. Even though the old contests would stay open, the most recently launched contest would be "the main one", and it would have a submission deadline and a final tournament and such. Those would only happen 2 to 4 times a year, with spaces between them. 2010-11-20T12:42:02 Maybe some "off season" contests. Like a re-run of the tron contest to see if people can do better. 2010-11-20T12:42:10 yes 2010-11-20T12:42:14 off season sounds good 2010-11-20T12:42:24 yeah, i'd be down for a tron re run 2010-11-20T12:42:26 jmcarthur, what about programming games as a platform ? 2010-11-20T12:42:32 boegel: Have you verified that you don't send flights with 0 or negative ships? 2010-11-20T12:42:38 medrimonia: I did 2010-11-20T12:42:45 Appleman1234: i don't understand the question 2010-11-20T12:42:51 medrimonia: my debug log says this is my last order: 2010-11-20T12:43:00 j3camero: one thing that would be nice in the future is maximums on paramters to do with map sizes 2010-11-20T12:43:03 In the FAQ it says you can read from files - but it does not look like they get packaged into the jar durin gcompile? 2010-11-20T12:43:07 medrimonia: [Order {orderSource = 10, orderDestination = 9, orderShips = 3},Order {orderSource = 2, orderDestination = 9, orderShips = 5},Order {orderSource = 9, orderDestination = 9, orderShips = 5}] 2010-11-20T12:43:08 at the start of the contest 2010-11-20T12:43:14 boegel: 11 active people 2010-11-20T12:43:18 medrimonia: all planets have the required number of ships in the current state 2010-11-20T12:43:24 i do like the idea of waiting to see what types of maps will be most interesting before finalising anything there 2010-11-20T12:43:26 antimatroid1, j3camero: actually, what would be nice is a *complete* specification from the very beginning 2010-11-20T12:43:32 boegel: and I'll be using it today 2010-11-20T12:43:34 Anybody interested in creating a browser-based IDE? That's all we need before the entire cotnest happens in the browser. 2010-11-20T12:43:42 no-one has seen games being counted as invalid when it's really not on there? 2010-11-20T12:43:43 allowing variable parameters is fine though 2010-11-20T12:43:56 I'm not even getting a "You LOSE" message in that case 2010-11-20T12:43:56 as long as the range is defined at the beginning 2010-11-20T12:44:05 yeah, general rules are good, but they need to be very clear, and stuck to the generality specified 2010-11-20T12:44:06 boegel: hum strange, do you write something to stdout? 2010-11-20T12:44:12 and as long as there is a way for the bot to know them dynamically 2010-11-20T12:44:20 medrimonia: ah, hmm... my debugging might be on 2010-11-20T12:44:23 that's actually the main thing, i think 2010-11-20T12:44:29 variable parameters should be communicated to the bots 2010-11-20T12:44:40 boegel: maybe ctrl-c: When you stop the game like that it's percieved like an invalid orders 2010-11-20T12:44:42 * boegel tries without debugging 2010-11-20T12:44:57 a generic key-value mapping would be plenty flexible i think 2010-11-20T12:45:08 http://phonons.wordpress.com/2010/06/01/cells-a-massively-multi-agent-python-programming-game/ and en.wikipedia.org/wiki/Programming_game/ 2010-11-20T12:45:10 yeah, my life would have been a lot easier for this too if the map passed me a value at the start saying how many planets there are and the maximum distance of the map 2010-11-20T12:45:31 i'm thinking things like number of turns. things you can't compute from the map state alone 2010-11-20T12:46:02 jmcarthur: I think a general contest language could be defined for all types of contests. 2010-11-20T12:46:05 yeah, but even then, it's sort of annoying to have to process the whole state, then set some things up, rather than allocate memory right at the start then fill the information in 2010-11-20T12:46:11 medrimonia: same issue if I disable my debugging output ;( 2010-11-20T12:46:12 j3camero: why a browser based IDE? i would never use it... 2010-11-20T12:46:27 McLeopold: i'm afraid of going "the xml way" 2010-11-20T12:46:33 *** bhasker has joined #aichallenge 2010-11-20T12:46:35 Yes, clarity has been an issue with the Planet Wars contest. Part of the cause of that is that I did not think people would take the contest so seriously. I'm glad they have, don't get me wrong. I am still in the mode where the contest only had 2 dozen people all in the same room and we could change the rules with a quick vote and it was no big deal. It's an issue we need to fix before the next contest so people don't get so frustrated. 2010-11-20T12:46:35 NO XML! 2010-11-20T12:46:59 McLeopold: well, that's what begins to happen when you try to make generic protocols 2010-11-20T12:47:04 boegel: strange... I haven't any invalid order that was false... 2010-11-20T12:47:08 jmcarthur: simple stuff, like A-Z is game specific, # is comment, ! is variable 2010-11-20T12:47:10 what do you mean by a contest language? 2010-11-20T12:47:17 medrimonia: strange indeed 2010-11-20T12:47:24 antimatroid1: markup/protocol stuff 2010-11-20T12:47:31 for game state, orders, etc. 2010-11-20T12:47:38 * boegel tries benzedrine 2010-11-20T12:47:49 jmcarthur: or just use json 2010-11-20T12:47:56 McLeopold: yeah i was just typing that 2010-11-20T12:47:58 but it's often so different depending on the game? 2010-11-20T12:48:04 or yaml 2010-11-20T12:48:04 * boegel tries benzedrine 2010-11-20T12:48:07 json would be fine 2010-11-20T12:48:13 (woops) 2010-11-20T12:48:17 yaml is tougher to parse if you for some reason have to do it by hand 2010-11-20T12:48:39 because indention matters? 2010-11-20T12:48:44 partly 2010-11-20T12:48:53 it also just have several specific syntax rules for various types of data 2010-11-20T12:48:56 *has 2010-11-20T12:49:01 there are good libraries for both, but I think json is more popular right now 2010-11-20T12:49:18 It's hard to create a general contest language. It would inevitably want to turn into XML, haha. What about games with more than 2 players? Games with only one player? Games that happen in one big persistent world shared by all players? What about games where state is carried between games? We can't anticipate everythign we would ever want to do. 2010-11-20T12:49:28 i love reading yaml. i just don't think i would mandate it when we have no idea what languages will be used in future contests for the bots 2010-11-20T12:49:40 json would be great in general though 2010-11-20T12:49:44 extensible, simple 2010-11-20T12:49:55 s/extensible/flexible/ 2010-11-20T12:50:07 json is also very trendy, hehe 2010-11-20T12:50:18 hmmm 2010-11-20T12:50:23 key-value stores are the best way to be able to change the protocol as the contest progresses if necessary, too 2010-11-20T12:50:36 j3camero: i think practically, it would be simpler than we can imagine 2010-11-20T12:50:40 this is kind of bike shedding though 2010-11-20T12:50:50 i'm more concerned with the big picture right now 2010-11-20T12:50:51 didn't read everything you all said so far, but what is wrong with the way the contest works now? What would json add to that? 2010-11-20T12:51:13 i want to see this taken seriously 2010-11-20T12:51:21 Migi32: it was a tangent, really 2010-11-20T12:51:27 oh ok 2010-11-20T12:51:32 Migi32: the main topic is the future of these contests 2010-11-20T12:51:39 Migi32: a way to parse the game state consistantly, allowing for additions without breaking older bots 2010-11-20T12:51:48 j3camero: one player "games" are more decisions problems than games 2010-11-20T12:51:52 *** amriedle has quit IRC (Quit: leaving) 2010-11-20T12:51:55 ah ok McLeopold that makes sense 2010-11-20T12:52:04 and n>2 players gets really messy when people start thinking about coalitions in the game 2010-11-20T12:52:17 i would be down for an n-player game though 2010-11-20T12:52:21 messy is not necessarily bad 2010-11-20T12:52:22 not so much a 1 player game 2010-11-20T12:52:30 it means there's room for more interesting ideas 2010-11-20T12:52:37 n-player planet wars is going to happen... 2010-11-20T12:52:42 I think TEAM games(2vs2) would be interesting though 2010-11-20T12:52:45 If I might add, I believe the biggest issue with future contests is labor shortage. It always has been. I find that once a new contest is launched, it's easy to get people to help work on it. But up until that point, it's a very lonely task. I am hoping that there will be more help now that it's open-sourced. 2010-11-20T12:52:49 but.. that requires totally different AI 2010-11-20T12:53:05 not nearly AS different as the a FFA game would require though 2010-11-20T12:53:09 i'm curious what n vs. n team games might be like, especially if we were to leave any interplayer communication protocol left unspecified and teams were random 2010-11-20T12:53:10 medrimonia: hmm, I'm now seeing it on benzendrine too 2010-11-20T12:53:18 would require some self-organization on part of the contestants 2010-11-20T12:53:18 i'm happy to work on anything that i can do anything helpful with 2010-11-20T12:53:28 medrimonia: really strange, because I'm *not* seeing it on the official server 2010-11-20T12:53:29 i'd be happy to do say c++ start packages etc. 2010-11-20T12:53:31 as in a channel is provided with no protocol 2010-11-20T12:53:53 and map generators etc. although they may not be the prettiest things ever 2010-11-20T12:54:06 Teams would be so cool. 2010-11-20T12:54:07 another big cause of the "labor shortage" is according to me the fact that you can't both help and participate in the contest 2010-11-20T12:54:10 boegel: The only thing that I know which is different between official server and tcp is the 0-flights... 2010-11-20T12:54:17 teams would require a new ranking algorithm 2010-11-20T12:54:29 *cough* trueskill *cough* 2010-11-20T12:54:37 i like the idea of people being able to join user made groups in the contests 2010-11-20T12:54:47 Migi32: you are right. The rules around that have been drastically relaxed. Now you can contribute code and all that and still win the contest. 2010-11-20T12:54:51 say up to like 5 groups per user 2010-11-20T12:54:59 and you can look at the leaderboards for your groups 2010-11-20T12:55:37 Self-created groups (and commenting on game replays, and friending, following, etc) would be a huge driver of interest, I am sure. 2010-11-20T12:57:00 There is no good library or tool for adding commenting to the site. We have to build it ourselves. 2010-11-20T12:57:15 maybe also another simple addition: a Java webchat client on the website, there's a bit of a split between people who use the forum and those who use IRC 2010-11-20T12:57:21 we *really* need to come up with a good sandbox and loosen the restrictions on what bots can do, e.g. logging 2010-11-20T12:57:53 logging (and being able to see your bot's logs afterwards) would be amazing 2010-11-20T12:58:18 j3camero: you aren't going to have any objections to the leaderboard being live this time? 2010-11-20T12:58:24 also, tarballs of various things 2010-11-20T12:58:39 i think most competitors are in favour of it being live 2010-11-20T12:59:01 ANyone? The rules say "You may not write to files. You may however read from files within your submission directory, which will be the current directory." but at least for a java bot this is impossible as any data files is not jarred up in the compile step? 2010-11-20T12:59:34 meh, i could read from files easily with c++ 2010-11-20T12:59:40 streams <3 2010-11-20T12:59:43 it would be cool if contestants could keep little dev blogs or at least notes on their profile pages 2010-11-20T12:59:55 just allow people to alter about me 2010-11-20T13:00:10 or add a text box on strategy description 2010-11-20T13:00:39 jmcarthur, trying to centralise dev blogs would be a step too far, I think 2010-11-20T13:00:42 antimatroid1: Really? So there is no jar step during compile? 2010-11-20T13:00:43 maybe the dev blogs could be kept private until after the contest is over 2010-11-20T13:00:54 Migi32: i don't think too far, but maybe low priority 2010-11-20T13:01:05 Gromgull: what do you mean? 2010-11-20T13:01:21 i do think it would be nice to be able to click on a bot and see a dev log right there 2010-11-20T13:01:21 * antimatroid1 can't stand java and doesn't know how it works 2010-11-20T13:01:38 It could be an option. 2010-11-20T13:01:51 i like the idea of being able to download peoples code off their profile once the contest is finished (should the user have wanted to share) 2010-11-20T13:01:53 You can toss a wiki in with Django, no problem 2010-11-20T13:01:58 Setting up dev blogs we could maybe do with relatively zero effort. I was toying with the idea of building the website on top of an exiting CMS like wordpress or somesuch. 2010-11-20T13:01:59 doing it in the forums is so messy 2010-11-20T13:02:14 antimatroid1: for java everything was compiled, then only the binaries generated zipped up - it seems any other file was not there at runtime 2010-11-20T13:02:16 jmcarthur, if you could easily change your description on your profile, you could use that to link to your blog 2010-11-20T13:02:17 I think 2010-11-20T13:02:20 IMO bots should be set up however you want and you should be expected to provide shell scripts for compilation and for running 2010-11-20T13:02:25 since I cannot see the output it's hard to say :) 2010-11-20T13:02:26 so how exactly does 'google' fit into this?.. they just donating spare processing power? 2010-11-20T13:02:36 burny: name only right now :( 2010-11-20T13:02:40 it's silly 2010-11-20T13:02:45 but it worked 2010-11-20T13:02:50 lol, true 2010-11-20T13:02:51 would binaries be doable in the future? 2010-11-20T13:02:58 that seemed like a good idea 2010-11-20T13:03:01 there is a ticket for this - but I guess it wont be solved now 2010-11-20T13:03:10 http://code.google.com/p/ai-contest/issues/detail?id=75 2010-11-20T13:03:13 antimatroid1: they better be. we just need a better sandbox 2010-11-20T13:03:14 who is payng for the processing then? 2010-11-20T13:03:27 jmcarthur burny: they also give money. 2010-11-20T13:03:50 antimatroid1: in combination with my shell scripts suggestion above we could accept pretty much anything, modulo support libraries and compilers for those who want to compile on the server and stuff 2010-11-20T13:04:06 burny: its a combination of Google, UWaterloo, and me. 2010-11-20T13:04:12 j3camero: oh? first i've heard of it 2010-11-20T13:05:08 Yup they pitched a nice wad of cash. They are very nice. 2010-11-20T13:05:26 But we've spent it all because of the need for more workhorse servers haha. 2010-11-20T13:06:02 binaries would be awesome. We could statically link libraries that wouldn't otherwise be available on the server. 2010-11-20T13:06:19 I agree. Binary submissions would be ballin. 2010-11-20T13:07:14 but keep the "no threads" rule. I don't know if that was for a technical reason or not but I like it :) 2010-11-20T13:07:25 how enforcable is that? 2010-11-20T13:07:36 i wouldn't even know how to use mutliple threads properly aha 2010-11-20T13:07:48 We had a really shady sandbox this time around. I am surprised that nobody has managed to crush the contest yet. When we have a better sandbox then binary submissions will be a 100% great idea. 2010-11-20T13:08:04 It's easy to enforce no threads. 2010-11-20T13:08:08 We do it now. 2010-11-20T13:08:23 if we use a sandbox that virtualizes a single core then that threading rule is irrelevant 2010-11-20T13:08:34 This is true. 2010-11-20T13:08:45 The ideal is to have a VM-based sandbox. 2010-11-20T13:09:07 If anybody here can solve that problem, that will be probably the single biggest contribution to the next contest. It's a really hard problem. 2010-11-20T13:09:14 even without a VM, the threading rule could be make irrelevant with control groups 2010-11-20T13:09:37 control groups allow you to impose resource constraints on a hierarchy 2010-11-20T13:09:37 * antimatroid1 doesn't have the first clue about that kind of stuff :) 2010-11-20T13:09:47 *process hierarchy 2010-11-20T13:10:28 I was working on a sandbox based on VMs (nick-named "sadbox"). It's kind of cool. You just run "./sadbox some_other_command" and it runs the command inside a fresh VM. Invincible security! 2010-11-20T13:10:49 j3camero: if we took in some more money (e.g. Google, donations, etc.) we could use slower sandboxes and just allow for more time per turn, making up for the slower execution speed with more servers 2010-11-20T13:11:02 the best sandboxes appear to be fairly slow, is my point 2010-11-20T13:11:35 j3camero: If you allow binaries, then you need to make sure different binaries can compete. A problem which a tcp server can solve. 2010-11-20T13:11:44 what the hell, now there's anothe rufes band person at like 32 2010-11-20T13:11:49 what the hell is rufes band/ 2010-11-20T13:12:09 j3camero: if every bot was wrapped to communicate via tcp to a contest server, you don't even need to run them on the same vm 2010-11-20T13:12:17 Yeah amstan what the fuck is Rufes band? 2010-11-20T13:12:27 j3camero: no idea, lol 2010-11-20T13:12:39 You must have created the group? 2010-11-20T13:12:53 What did they say it was? A university? 2010-11-20T13:12:57 people just demand a name and he adds i suspect 2010-11-20T13:13:01 I should just fucking google it. 2010-11-20T13:13:05 already tried that 2010-11-20T13:13:27 i just think of date rape when i see rufes 2010-11-20T13:14:10 aha top result on google is their organization ranking page 2010-11-20T13:14:12 rufus band exists 2010-11-20T13:15:06 They seem to be mostly from Spain. 2010-11-20T13:16:23 * janzert waves 2010-11-20T13:17:17 Hey janzert. 2010-11-20T13:17:22 BaronTrozo will be up into like second soon too :\ 2010-11-20T13:17:33 People were just brainstorming all their ideas for improving the next contests. 2010-11-20T13:17:44 great 2010-11-20T13:17:48 j3camero: a lot of that was regurgitated :) 2010-11-20T13:17:56 Yes definitely. 2010-11-20T13:18:07 although I probably won't be able to contribute much until next fall again 2010-11-20T13:18:23 Top 10 players: bocsimacko(3395), Raschi(3258), _Astek_(3210), bix0r4ever(3195), _iouri_(3179), BaronTrozo(3167), wagstaff(3148), felixcoto(3138), protocolocon(3075), george(3053) 2010-11-20T13:18:23 We should probably get a Wiki page going for all these ideas so we can order them and get people working on all them. 2010-11-20T13:18:52 Janzert are you busy with stuff for the next 8 months? 2010-11-20T13:18:56 In school, or working? 2010-11-20T13:19:17 pretty sure he's well past school 2010-11-20T13:19:35 working on my Arimaa bot and working working after that 2010-11-20T13:19:42 i could be very wrong, i should stop answering other peoples questions aha 2010-11-20T13:19:55 antimatroid1: you're correct though 2010-11-20T13:20:25 Nice. 2010-11-20T13:20:49 is arimaa that chess like game? 2010-11-20T13:20:58 I like this Arimaa sport. I should take a crack at it. I used to be huge into comp chess. My bot could beat GNU Chess :-D 2010-11-20T13:21:13 what kind of strategy do you use for picking your starting location? 2010-11-20T13:21:19 yes, although about the only chess likeness of it is that you can use a chess board and pieces to play it ;) 2010-11-20T13:22:00 antimatroid1: currently I think all bots just choose randomly from a few known good starting positions 2010-11-20T13:22:49 janzert: how are the best bots now? Are they even close to beating a human? How many weeks do I have to practice before I can beat the best bot? 2010-11-20T13:23:15 the best bots beat typical humas.. 2010-11-20T13:23:18 if you seriously study the game maybe around 2 months of work 2010-11-20T13:23:21 hmm, i may take a look at it after this ends 2010-11-20T13:23:29 and beat good humans in reaction time, but are only competitive in logic 2010-11-20T13:24:19 Is it something that would be any good as an AI contest? Or is it like chess where you have to do 5 solid days of work before you even have something that can play? 2010-11-20T13:24:26 burny: what do you mean by typical humans? best bots are ranked around 2100 elo or so but I don't think it is that hard for a human to get there either 2010-11-20T13:25:01 j3camero: yeah, I think the games are too cpu intensive and it takes too much work to get really competitive 2010-11-20T13:25:15 i like 2 player simtultaneous games for these 2010-11-20T13:25:16 janzart.. I mean that, a typical human would make mistakes 2010-11-20T13:25:26 and newish games like planet wars are awesome 2010-11-20T13:25:35 where it's a completely new problem for people to look at 2010-11-20T13:25:45 simultaneous* 2010-11-20T13:26:04 burny: yes, but fortunately bots make more because they don't understand the strategic consequences :) 2010-11-20T13:26:52 yeah for this contest we need something that can have reasonably quick games, have nice visuals, fairly simple to get started in, etc. 2010-11-20T13:26:58 I think that on average, the human programming a bot, would be about 200-500points above their bot 2010-11-20T13:27:27 burny: nah, my bot is stronger than me 2010-11-20T13:27:41 ps: 'average' 2010-11-20T13:28:03 maybe I don't really know 2010-11-20T13:28:25 but yeah I'm probably the weakest player of all the programmers 2010-11-20T13:28:37 I don't actually play hardly at all 2010-11-20T13:29:02 janzert: I take it you use minimax? 2010-11-20T13:29:07 well if the bot is a heuristic bot the maker will probably be better than their bot. But for minimax searches this is probably not so. 2010-11-20T13:29:10 and I'm pretty sure with a couple weeks practice I could climb past it in strength 2010-11-20T13:29:20 Error323: 7/1/92 and 5/1/94 for the "-test" 2010-11-20T13:29:29 antimatroid1: yeah, all the best bots are alpha/beta of some sort 2010-11-20T13:29:41 mega1: thanks!!! 2010-11-20T13:29:52 mega1, what are those numbers? 2010-11-20T13:29:53 hmmm, i'm going to give that a go i think 2010-11-20T13:29:53 it's not what you expected 2010-11-20T13:30:12 Migi32: wins/draws/losses 2010-11-20T13:30:17 against? 2010-11-20T13:30:17 janzert: it's all about the negascout and MTD(f) :-) 2010-11-20T13:30:23 Migi32: Error323's bots playing against bocsimacko-b0.70 2010-11-20T13:30:36 ok :) 2010-11-20T13:30:43 7 wins, not bad :) 2010-11-20T13:30:58 j3camero: I don't think MTD(f) is used at all, although I did test it 2010-11-20T13:31:08 How did it do? 2010-11-20T13:31:11 lots of pruning going on of course 2010-11-20T13:31:20 MTD(f) is totally silly if you can only go 2 plies deep, I think 2010-11-20T13:31:26 MTD(f) was quite a bit worse 2010-11-20T13:31:26 Oh yeah 2010-11-20T13:31:44 So you have to prune using some kind of heuristic? 2010-11-20T13:31:55 Rather than using strictly beta cutoffs? 2010-11-20T13:32:37 Migi32: well remember all the searchers go step wise instead of turn wise as well so 2 ply is 8 steps and I generally reach 12 steps or 3 ply in most positions 2010-11-20T13:32:56 well more on average I guess than most 2010-11-20T13:33:35 janzert, what's a step again? 2010-11-20T13:33:37 j3camero: yeah, there are several pruning heuristics in place. A lot of the from the chess world 2010-11-20T13:33:59 I see I see 2010-11-20T13:34:02 Migi32: moving 1 piece 1 square. you get 4 steps per turn 2010-11-20T13:34:13 Would you ever get a win from using something like the null move heuristic? 2010-11-20T13:34:21 how many people's bots aren't susceptible to being 'faked out'? 2010-11-20T13:34:30 janzert, are you talking about Arimaa all of a sudden? 2010-11-20T13:34:55 j3camero: yes, although I don't specifically remember how much it helped it is in use 2010-11-20T13:35:05 Actually, the best bot beats the best player 2010-11-20T13:35:06 Migi32: well not all of a sudden :) 2010-11-20T13:35:09 In chess 2010-11-20T13:35:10 *** Appleman1234 has quit IRC (Ping timeout: 250 seconds) 2010-11-20T13:35:10 oh, you are. I thought we were still talking about PlanetWars XD 2010-11-20T13:35:16 hehe 2010-11-20T13:35:29 But it's not really ai it's just brutal force 2010-11-20T13:35:33 I actually have a really good idea about how to make a bot for Arimaa 2010-11-20T13:35:39 medrimonia: yeah, in chess the bots have well surpassed the humans 2010-11-20T13:35:52 not in arimaa though 2010-11-20T13:36:08 i want to see an ai contest for game that involves shooting at each other, perhaps with a very large number of somewhat slow bullets so you can dodge them 2010-11-20T13:36:12 go is the game I'm most interested in, for writing AI 2010-11-20T13:36:15 does anyone know which map this is, and where I can get it? 2010-11-20T13:36:16 http://72.44.46.68/canvas?game_id=1108405 2010-11-20T13:36:16 I thought you were speaking about chess^^ 2010-11-20T13:36:25 Migi32: great come compete with us there after this is done :) 2010-11-20T13:36:32 it would have some things in common with tron in that you will want to trap your opponent without getting trapped 2010-11-20T13:36:34 the server marks this as an invalid move on the first turn 2010-11-20T13:36:37 janzert, that's exactly what I was planning to do 2010-11-20T13:36:43 and I have no idea why 2010-11-20T13:36:48 boegel: if you press "show game state" at the bottom of the screen, you get the mapfile 2010-11-20T13:36:54 Mistmanov: ah, great, thx 2010-11-20T13:37:38 for a game like glacon.. I think the most interesting way to play, would be if you don't get to see exactly what the opponent does 2010-11-20T13:37:41 maybe with some sort of momentum so you can't just weave around the bullets so freely 2010-11-20T13:37:53 Mistmanov: doesn't seem to work in this case, probably because there's no single turn being completed? 2010-11-20T13:38:01 it does for me? 2010-11-20T13:38:04 burny: yes! fog of war would be awesome :) 2010-11-20T13:38:05 "If a program wins the Arimaa challenge, it must be described in a paper submitted to the ICGA technical journal after winning the challenge in order to receive any prize." lame 2010-11-20T13:38:06 hmm, weird 2010-11-20T13:38:10 I can see the list of planets 2010-11-20T13:38:18 wut i'm owning on tcp 2010-11-20T13:38:18 eg. you can only see how many ships are on an enemy planet by owning a planet near it 2010-11-20T13:38:21 burny: things like that are what make me like poker IRL (not sure i want to do a poker bot though) 2010-11-20T13:38:26 c# main file is MyBot.cs correct? 2010-11-20T13:38:37 and you don't get to see how many ships are attackign a planet(just know that SOMEthign is attacking) 2010-11-20T13:38:51 jmcarthur: i want something like that too 2010-11-20T13:38:52 boegel: oh, it stopped working like right now 2010-11-20T13:38:56 Mistmanov: :D 2010-11-20T13:39:01 Mistmanov: wtf 2010-11-20T13:39:06 first time I looked at that map it worked, now it doesn't 2010-11-20T13:39:12 antimatroid1: which thing? the shooter idea? 2010-11-20T13:39:27 Mistmanov: maybe it can't be done by two people at the same time? :P 2010-11-20T13:39:33 could be =) 2010-11-20T13:39:44 boegel: map_id=maps/map1278.txt 2010-11-20T13:39:52 *** Blkt has quit IRC (Read error: Connection reset by peer) 2010-11-20T13:39:52 it's in the source 2010-11-20T13:39:55 jan, the AI for the game working like that, wouldn't be incredibally different 2010-11-20T13:39:56 yeah 2010-11-20T13:40:01 Zannick: ah, great, thx 2010-11-20T13:40:03 3d would be awesome 2010-11-20T13:40:05 antimatroid1: I'm not sure why that is still listed it hasn't been done for a few years 2010-11-20T13:40:07 but i'd settle for 2d 2010-11-20T13:40:08 excet, that you can't KNOW what the 'best' move it 2010-11-20T13:40:09 is 2010-11-20T13:40:24 3d would be too hard to win unless there was like a maze to navigate as well... and then it gets hard to view 2010-11-20T13:40:32 janzert: what hasn't been done? 2010-11-20T13:40:35 burny: hmm? arimaa? 2010-11-20T13:40:47 antimatroid1: the ICGA journal article 2010-11-20T13:41:06 although I think it would be great if it was started back up again 2010-11-20T13:41:08 no.. galcon 2010-11-20T13:41:16 a shooter with many players in a game would be even cooler 2010-11-20T13:41:21 people seem to spit out so many unecessary articles these days :| 2010-11-20T13:41:30 a FFA or something :) 2010-11-20T13:41:41 oh.. team based shooter could be neat actually 2010-11-20T13:42:00 yeah 2010-11-20T13:42:08 you know what's stupid 2010-11-20T13:42:08 harder to manage 2010-11-20T13:42:11 but cool 2010-11-20T13:42:18 that ur rating drops when winning from low ranked people 2010-11-20T13:42:29 http://www.teagames.com/games/topdog/play.php 2010-11-20T13:42:39 that is a half decent 2d shooter game 2010-11-20T13:43:44 i was thinking more top-down style 2010-11-20T13:44:17 I'd kind of like to see a game with partial information 2010-11-20T13:44:17 hmm... 2010-11-20T13:44:25 janzert: me too 2010-11-20T13:44:39 i'm half tempted to write up this idea i have in my head 2010-11-20T13:44:52 maybe with some diagrams to show things a little more clearly 2010-11-20T13:45:13 anim, as is, that game is way too trivial though 2010-11-20T13:46:28 how about stratego :P 2010-11-20T13:46:34 I really liked that game as a kdi 2010-11-20T13:46:36 kid* 2010-11-20T13:46:58 heh 2010-11-20T13:47:00 it's with impartial information yet not with too complex rules 2010-11-20T13:47:03 kind of simple though 2010-11-20T13:47:16 okay, i'm gonna write up some ideas... 2010-11-20T13:47:26 Mistmanov: the accounts Mistmanov and Mistmanov2 are ones you lost the password for and are no long in use right? 2010-11-20T13:47:27 *** Accoun has quit IRC () 2010-11-20T13:47:42 stratego? man, that's totally called guess who in australia 2010-11-20T13:47:47 If that's the case I'm just going to disable the bot playing under Mistmanov 2010-11-20T13:47:57 guess who is a different game 2010-11-20T13:47:59 no problem 2010-11-20T13:48:01 jznaert: only 2 2010-11-20T13:48:13 oh wait, don't pay attention to me 2010-11-20T13:48:16 i thought you were talking to x 2010-11-20T13:48:24 jmcarthur, simple? I wouldn't be too sure. I thought planetwars was simple 2010-11-20T13:48:24 the two of you existing confuses the hell out of me 2010-11-20T13:49:12 Yeah we have to pick a game for the next contest. 2010-11-20T13:49:15 Hold em? 2010-11-20T13:49:18 it's as hard as the #1 bot makes it 2010-11-20T13:49:40 i'm just not a huge board/card game fan for ai 2010-11-20T13:49:42 not interesting to watch 2010-11-20T13:49:50 Some games have an upper limit to how good you can really get. Rock Paper Scissors was a big mistake for that reason. 2010-11-20T13:49:53 and i think interestingness to watch is important for getting attention 2010-11-20T13:50:13 well yeah, rps is a terrible idea for an ai contest :P 2010-11-20T13:50:14 You're right jmcarthur 2010-11-20T13:50:29 There are so many factors, and it rules out all games, haha. 2010-11-20T13:50:41 I have a wiki page on the criteria for good game choices 2010-11-20T13:50:54 i agree with jmcarthur about bored games 2010-11-20T13:51:02 jmcarthur, "was" a big mistake? You actually did a contest on RPS? 2010-11-20T13:51:19 i think it was just uw based then 2010-11-20T13:51:22 http://code.google.com/p/ai-contest/wiki/GameCriteria 2010-11-20T13:51:22 is team games a posible consideration? 2010-11-20T13:51:27 Migi32: i didn't. j3camero did :P 2010-11-20T13:51:42 hehe yeah my bad 2010-11-20T13:51:47 sorry, I get your names confused all the time :) 2010-11-20T13:52:10 I personally wouldn't like teamgames for the main competition, especially if it involves actual cooperation with other coders 2010-11-20T13:52:20 Migi32: yeah there was a RPS contest 2010-11-20T13:52:21 do the test servers consider multiple orders to/from the same planet a problem? 2010-11-20T13:52:28 okay, there's more than one way to do a "team" game. one way is to have each team be made up of many bots. another is to have each team be made up of multiple instances of the same bot. another is to have all the agents of each team be controlled by a single instance of each competing bot 2010-11-20T13:52:28 the teams would be random(and change each game) 2010-11-20T13:52:31 Mistmanov: ^^ 2010-11-20T13:52:34 j3camaro: San Juan, the card game 2010-11-20T13:52:35 having to work with others in order to climb the rankings is exactly not what i'm looking for in a contest 2010-11-20T13:52:40 With team games pairing would be important. I think we have to match people of similar strength. 2010-11-20T13:53:23 i think doing a team game would have a lot of subtle issues that would be difficult to solve well 2010-11-20T13:53:24 implement trueskill for that, it is the situation it was designed for 2010-11-20T13:53:28 I think straight-up ELO scoring with random teams would be sufficient for simple team games 2010-11-20T13:53:32 Good point jmcarthur. 2010-11-20T13:53:40 i'd prefer to do n-player each for themseves 2010-11-20T13:53:43 team games also don't converge your rank anywhere near as quickly as, say, FFA 2010-11-20T13:54:07 What about a simple RTS? 2010-11-20T13:54:11 a team game with each team member being the player's bot that could only communicate when they're close to each other in the game-world would be interesting 2010-11-20T13:54:12 FFA generally doesn't have a meaningly useful 'best' AI though 2010-11-20T13:54:15 Either 1v1 or FFA 2010-11-20T13:54:18 RTS? 2010-11-20T13:54:19 j3camero: there's no such thing as a simple rts 2010-11-20T13:54:22 Or... MMO... 2010-11-20T13:54:31 i tried to come up with a simple rts for this one and gave up 2010-11-20T13:54:40 It's tough 2010-11-20T13:54:45 Balance would be a huge issue. 2010-11-20T13:54:46 galcon was the closest i could think of 2010-11-20T13:54:50 as in real time strategy? Isn't galcon an rts? 2010-11-20T13:54:51 yeah 2010-11-20T13:55:04 I'm talking more starcraft-like. 2010-11-20T13:55:04 well... an easy way to test the feasability of my idea.. 2010-11-20T13:55:19 j3camero: yeah, there are so many elements to that... 2010-11-20T13:55:25 resources, building, units, etc. 2010-11-20T13:55:28 would be to take all the galcon AI bots as-is, and toss them into a random 2vs2 tournament 2010-11-20T13:55:43 cutting out any of those things, IMO, destroys it 2010-11-20T13:56:07 if the elo of 1on1, compared to the ELO of random teams, is generally fairly consistent.. then my idea clearly works 2010-11-20T13:56:12 i like the idea of people being able to execute moves when their bot is ready to 2010-11-20T13:56:15 then having no time limits 2010-11-20T13:56:23 so you'd need RTS for that 2010-11-20T13:56:24 if random team elo comes out fairly random.. then I am wrong 2010-11-20T13:56:34 One way to deal with the balance issue is this. There is only one type of unit, but with many parameters. It has range, damage, health, etc. This unit could be made into a tank, melee, archer, whatever. It's just different parameters. We could change the unit types and evolve it over the life of the contest to achieve balance. 2010-11-20T13:57:38 burny: it's not a bad idea. Try to collect people's bots after this ends and you can test it out using the contest endinge and ranking scritp. 2010-11-20T13:57:47 yeah... i thought of that too 2010-11-20T13:58:00 bots aren't written to do that 2010-11-20T13:58:11 i think mcleopold is keen to do n-player games though 2010-11-20T13:58:12 Some units use a rock as a weapon, some use sissors, other go for paper cuts... 2010-11-20T13:58:13 anti: it doesn't matter.. none of them are written to do that 2010-11-20T13:58:15 I think buildings and resources could be cut out. It would just be a "micro" contest. 2010-11-20T13:58:30 *** Accoun has joined #aichallenge 2010-11-20T13:58:34 rock paper scissors units aren't interesting units though :\ 2010-11-20T13:58:39 that's essentially random 2010-11-20T13:59:03 j3camero: see i think that kills it right there 2010-11-20T13:59:05 jmcarthur: not if you have to arrange them into effective groups 2010-11-20T13:59:12 jmcarthur, I watch professional starcraft games sometimes and even battles with 1 type of unit can be very interesting 2010-11-20T13:59:22 The big advantage that RTS has is it's sexy. It would make for some awesome game playbacks and fierce competition. 2010-11-20T13:59:23 as soon as you remove something like resource gathering then the game might as well just be micro 2010-11-20T13:59:24 it's about trying to surround the enemy's army 2010-11-20T13:59:34 j3camero: Oh while you're here, I've been making the assumption that the cutoff time to start the final tournament is based on server time for the timezone 2010-11-20T13:59:37 is that right? 2010-11-20T13:59:38 units only would be all about micro 2010-11-20T13:59:45 to make the game work as teams.. all you gota do is make your teammate appear as a neutral planet that has an extra few hundred ships on it 2010-11-20T14:00:04 janzert: sure :-) 2010-11-20T14:00:04 server time is CST btw and is what displays on all the profile times and such 2010-11-20T14:00:05 burny: and what about "neutral" fleets? 2010-11-20T14:00:15 neutrals show up as is.. 2010-11-20T14:00:16 j3camero: Ok, I'll make that official then 2010-11-20T14:00:21 enemys all show up as 1 2010-11-20T14:00:25 burny: fleets 2010-11-20T14:00:29 and ally shows up as heavily populated neutrals 2010-11-20T14:00:30 burny: teammate fleets, i mean 2010-11-20T14:00:30 jmcarthur, allied fleets could be your fleets, there'd be no difference 2010-11-20T14:00:45 that completely changes the quality of your ai 2010-11-20T14:00:51 hm, ya.. teammate fleets could be awkward 2010-11-20T14:00:54 because you can't actually control them 2010-11-20T14:01:03 janzert: your plan on the forums is tight. Let's do it that way. 2010-11-20T14:01:03 you can't control your fleets either 2010-11-20T14:01:11 Migi32: but you can plan to when they land 2010-11-20T14:01:14 maybe divide the planets up 2010-11-20T14:01:14 teammate fleets, can show up as being your own fleets 2010-11-20T14:01:18 j3camero: great 2010-11-20T14:01:29 and if you do attack your ally.. you just give your fleets to your ally 2010-11-20T14:01:39 jmcarthur, no you can only control when you *send* fleets, which is the same as controlling planets only 2010-11-20T14:01:39 it clearly wouldn't be perfect.. 2010-11-20T14:01:45 janzert: do you want to be the operator in the last few hours before the deadline, or do you want me to do it? 2010-11-20T14:01:51 but it would be a good approximation, that requires very little effort 2010-11-20T14:02:14 Migi32: when there is a fleet on the way to planet it's normally perfectly fine to plan to send them off again as soon as they land... 2010-11-20T14:02:28 jmcarthur, ah you mean like that. Good point 2010-11-20T14:02:52 j3camero: I don't mind doing it and since I've had my head buried in the code for the last month plus I feel fairly confident around it 2010-11-20T14:02:56 there's also the issue of what happens when you move your own fleets to a planet owned by a teammate 2010-11-20T14:03:08 j3camero: not that I won't forget something anyway :P 2010-11-20T14:03:15 I already said jmc.. it gives your fleets to your ally 2010-11-20T14:04:19 janzert: excellent. I will also be around for backup. I guess I can focus more on forums then. After last contest I thought it would be good to nominate one single "systems guy" for the last few horus, since we were all stepping all over each other inside the server. 2010-11-20T14:04:23 ah... that seems like it would throw most AIs off then 2010-11-20T14:04:41 janzert: im sure you'll do fine. And we all make mistakes, it's cool. 2010-11-20T14:04:42 ya, it would throw them off.. but it would be fair, and easy 2010-11-20T14:04:45 j3camero: yeah, having one person do the transition is certainly best 2010-11-20T14:05:05 And I guess we'll all switch our watches to CST. 2010-11-20T14:05:06 too many moving parts to have multiple people poking at it :) 2010-11-20T14:05:09 That's probably good 2010-11-20T14:05:41 Last contest I left it open a couple minutes past the deadline to reduce frustration. This will be basically the same deal, since many people will assume it's 11:59 PM EST. 2010-11-20T14:05:59 yeah i was going to make a comment about that 2010-11-20T14:06:02 I'm going to change the front page to give the timezone now 2010-11-20T14:06:02 *** smloh1 has joined #aichallenge 2010-11-20T14:06:20 it's possibly worth suggesting on the main page that people upload at least a few hours before the deadline too 2010-11-20T14:06:44 Cool 2010-11-20T14:08:12 what is the simplest absolute way to start generating rules or heuristics for a bot to use in this? 2010-11-20T14:08:21 i don't want to waste my time building an engine.. 2010-11-20T14:08:28 *** p4p4 has joined #aichallenge 2010-11-20T14:08:31 think about simple examples 2010-11-20T14:08:37 and work out what ships you would want to send 2010-11-20T14:08:38 foucist: download a starter pack and edit it 2010-11-20T14:08:54 jmcarthur: yeah but then i have to code all sorts of stuff on top of that 2010-11-20T14:09:00 foucist: like what? 2010-11-20T14:09:04 just to start generating rules/heuristics 2010-11-20T14:09:06 foucist: it gives you a working bot from the start 2010-11-20T14:09:11 jmcarthur: i know 2010-11-20T14:09:21 i model a move with a single destination and multiple fleets, from numerious sources with different sending times 2010-11-20T14:09:26 foucist: have you seen the tutorials? 2010-11-20T14:09:35 jmcarthur: well lets say i want to generate many different rules and test them.. 2010-11-20T14:09:45 j3camero: yes i have, i've gone through all that 2010-11-20T14:09:45 front page updated 2010-11-20T14:09:54 foucist, you want to have a good bot without spending the time writing one? 2010-11-20T14:09:59 j3camero: looking for a starting point to do some proper EA with this 2010-11-20T14:10:27 Migi32: no, i just want a solid starting point to do EA/GP etc with this, but it seems that i'd have to code up an engine to handle it or something 2010-11-20T14:10:50 Ah okay so you are looking for a genome representation basically. 2010-11-20T14:10:53 i want to be able to generate different rules/heurstics for the bot 2010-11-20T14:10:57 ah, notification of who wins is a lot sooner now? 2010-11-20T14:11:04 and mutate/etc so on 2010-11-20T14:11:05 will the rankings be public during the testing? 2010-11-20T14:11:14 foucist: you're expected to do that stuff for yourself 2010-11-20T14:11:19 boeggel: not telling haha 2010-11-20T14:11:26 antimatroid1: yeah that's what i figure 2010-11-20T14:11:28 j3camero: ? :) 2010-11-20T14:11:29 i thought i'd ask though 2010-11-20T14:11:32 if the bots in the 1on1 contest are ranked with ELO of x(n), and with randomly chosen teams, their ELO is y(n). if the standard deviation of x(n) vs y(n) is less then something like.. 10-20%.. then if the AI's were properly deisgned to work in teams.. the contest would work real well 2010-11-20T14:11:37 j3camero: it would be good if they are live 2010-11-20T14:12:10 j3camero: what do you think is there an easy way to deal with generating rules/heuristics, any proto-genome-like thing i can do etc... 2010-11-20T14:12:28 foucist: you have a week :) 2010-11-20T14:12:31 antimatroid1: I like the not live idea. It means more suspense. 2010-11-20T14:12:39 antimatroid1: yep! 2010-11-20T14:12:50 j3camero: it's nowhere near as exciting as watching the leaderboard evolve 2010-11-20T14:12:51 foucist, I don't believe many people have explored this direction so far. 2010-11-20T14:12:53 foucist: let me think about that while I fry a steak for lunch. 2010-11-20T14:12:54 antimatroid1: which is why i want to figure out something super simple 2010-11-20T14:12:55 and watching the games as they're played 2010-11-20T14:13:15 j3camero: from what I can tell there is pretty overwhelming support in the community to leave the leaderboard live during the finals 2010-11-20T14:13:27 and I do think it helps keep the community active while they run 2010-11-20T14:13:40 does anyone know if the zeroviz and benzendrine server consider order sets with multiple orders to and from the same planets invalid? 2010-11-20T14:13:41 ^^ what he said 2010-11-20T14:13:59 boegel: pretty sure that's fine on both 2010-11-20T14:14:12 on the other hand I do understand the suspense aspect but I think the negatives out way it when hiding the leaderboard 2010-11-20T14:14:22 boegel: i am certain that I send those kinds of orders on both. 2010-11-20T14:14:29 antimatroid1: my experiences seem to suggest problems on the test servers 2010-11-20T14:14:46 then I have no clue why my order sets is considered wrong :( 2010-11-20T14:14:51 let me put it this way, is it more interesting to hear the result of a *place sports name here* match, or actually watch the game and see it unfold 2010-11-20T14:15:11 oh 2010-11-20T14:15:13 wait 2010-11-20T14:15:16 Yes I think I am the only one who likes taking the leaderboard down during the finals. 2010-11-20T14:15:25 same source/destination is probably not a good idea... 2010-11-20T14:15:32 yep, that is the problem :) 2010-11-20T14:15:35 seems like the official server doesn't check on that? 2010-11-20T14:16:12 boegel: I actually updated the spec last night since the official engine doesn't do anything about that 2010-11-20T14:16:21 ah, k 2010-11-20T14:16:30 janzert: so, it's OK on the official server? 2010-11-20T14:16:35 yes 2010-11-20T14:16:37 it's dumb though, I should fix it 2010-11-20T14:17:04 although it won't have *any* visible effect since they just land later on the same move 2010-11-20T14:17:25 Honestly I think it should be up to the community to decide whether to leave the leaderboard up or down. I don't think it makes sense for me to decide that. So it looks like it will probably stay up. 2010-11-20T14:17:26 janzert: well, you do waste budget you can spend otherwise in that turn... 2010-11-20T14:17:46 j3camero: :) 2010-11-20T14:17:53 right, i'm going to bed with that good news 2010-11-20T14:18:10 yeah, I meant there is no way to tell from the game state afterward that those fleets existed 2010-11-20T14:18:24 Top 10 players: bocsimacko(3378), Raschi(3195), _Astek_(3178), BaronTrozo(3168), _iouri_(3156), bix0r4ever(3146), wagstaff(3142), felixcoto(3125), protocolocon(3053), george(3043) 2010-11-20T14:18:37 how about a vote on the forums? 2010-11-20T14:18:42 boegel: i assumed you sent them to yourself because you wanted to use them there in the future? 2010-11-20T14:18:47 There is a bit of funny history here. The biggest reason I took it down in the past is because if I did not, there would be constant accusations of rigging. Once the final is under way, people get so sensitive about any change in their ranking. I think Accoun was a big factor in that :-) 2010-11-20T14:19:15 they have more reasons to question cheating if it's hidden i think 2010-11-20T14:19:33 because then they can't even screen shot all the gams they've had and see that they don't get removed 2010-11-20T14:19:42 antimatroid1: no, it's a bug in my bot 2010-11-20T14:19:56 You're right in a sense antimatroid1. But it was all emotional. Once we took it down, people just relaxed. 2010-11-20T14:20:10 lies, it was horrible! :P 2010-11-20T14:20:35 And we had some peace for a few days to generate the best rankings we could. 2010-11-20T14:20:35 i had to sit around twiddling my thumbs 2010-11-20T14:21:32 maybe you could show results without ranking? 2010-11-20T14:21:35 There was also a bit of "art" in the way we ran the tournament for Tron. I didn't want people to see that and get the wrong idea. For example, I temporarily banned all the bots except the top 50 so that they would play a round robin. It was a good idea and it worked well, but it would have casused massive panic. 2010-11-20T14:21:40 heh, well just so everyone is forewarned once the finals start I plan on inserting ear plugs and ignoring anything everyone says until the finals are finished. :P 2010-11-20T14:22:19 McLeopold: i really liked the idea of watching the rankings evolve as well 2010-11-20T14:22:39 * janzert is afk for a while 2010-11-20T14:22:41 janzert we have that luxury this time around, since the tournament system is stable enough to just run by itself. We don't have to do anything heavy-handed with it that would be mistaken for rigging if it was publicly visible. 2010-11-20T14:28:00 how about this: we let the rankings evolve and then when the top 8 best players are known, play a regular tournament with them to determine the final winners, with matches that can be watched live, 2010-11-20T14:28:33 a regular tournament with winner advances to the next round, semifinals, finals and then the eventual winner 2010-11-20T14:29:14 migi32: that's a terrible idea :P 2010-11-20T14:29:37 if anything, a top x round robin will be run i think 2010-11-20T14:30:29 i think idealy the top 10 bots should play each other quite a few times each though 2010-11-20T14:30:57 That would be sort of like flipping a coin to choose the winner 2010-11-20T14:31:02 well yea the type of tournament doesn't matter I just want to watch a series of matches that end in a nail-biting finish of the first and second best bot against each other :P 2010-11-20T14:31:12 the thing that I think would be neat about 2vs2 instead of 1vs1 for a game like glacon: is it makes it so there is no 'perfect play'.. because you don't know what 50% of your team is going to do 2010-11-20T14:31:18 Migi32: that isn't fair on who had the best bot 2010-11-20T14:31:24 That would be really cool yes. Suspenseful. Also, the web server would be crushed. 2010-11-20T14:31:38 burny: you make it sound like 1v1 is trivial 2010-11-20T14:32:42 it nearly is 2010-11-20T14:32:44 burny, have you played Galcon? Even with optimal play you can lose an FFA just by having bad enemies 2010-11-20T14:32:56 yes, ffa glacon is lame 2010-11-20T14:33:20 2v2 would be more or less the same if your ally is a random bot 2010-11-20T14:33:25 FFA glacon would be way better if when someone dies, whoever killed the most of their ships, gains 100 ships 2010-11-20T14:33:53 migi: different bots have different behavious though, that may or may not work well, with/against other bot types 2010-11-20T14:35:30 so, a large component of the bot.. would be making it work with the teammate 2010-11-20T14:35:49 How is 1v1 Galcon trivial? 2010-11-20T14:35:57 If this is true then you should be winning. 2010-11-20T14:36:27 because there is basically a 'perfect' strat 2010-11-20T14:36:35 Really? 2010-11-20T14:36:38 What is it? 2010-11-20T14:37:36 I mean, there must be a perfect strat. In principle, you could minimax this game up (accounting for the simulateneous moves). But that is not computationally feasible. 2010-11-20T14:37:40 would you guys say the competition is more fierce on benzendrine or zeroviz? 2010-11-20T14:38:10 on teh first move, it's pretty easy to figure out what the best planets to take is 2010-11-20T14:38:30 okay, a quick sketch of my idea for a future contest: https://docs.google.com/viewer?a=v&pid=explorer&chrome=true&srcid=0B3_Nay8FIDhVMWMwZTZlYzYtMDI2My00ODIxLWE0NGYtOGJlMDRiYmUzZTM4&hl=en 2010-11-20T14:38:32 and if the opponent takes the same mirror'd ones.. then it's not hard to also figure out the best ones to take next 2010-11-20T14:38:34 pretty simple, really 2010-11-20T14:39:12 of course it still needs details and polish 2010-11-20T14:39:47 "Sorry, we are unable to retrieve the document for viewing" 2010-11-20T14:39:51 wat 2010-11-20T14:39:57 lemme just make it completely public 2010-11-20T14:40:05 i had set it to make the link work, but whatev 2010-11-20T14:40:22 try again 2010-11-20T14:41:08 i didn't put how you "win" in it. it could be last man standing, but i think a frag count might be more fun (along with respawns, of course) 2010-11-20T14:41:24 ah, works now 2010-11-20T14:41:27 that of course depends on how fast kills are in the game anyway 2010-11-20T14:41:40 if they are slow enough then last man standing might be better 2010-11-20T14:42:17 maybe making it be frags, with last man standing getting the win in case of a tie.. or an extra 1frag, or some such 2010-11-20T14:42:26 jmcarthur: lemme read this 2010-11-20T14:42:28 *** fmeyer has joined #aichallenge 2010-11-20T14:42:34 i like it 2010-11-20T14:42:39 google Docs wuuut 2010-11-20T14:42:55 ok so it's basically a top-down shooter 2010-11-20T14:43:05 j3camero: sorry. i wrote it in markdown and then didn't know how else to share it ^_^ 2010-11-20T14:43:07 Migi32: yeah 2010-11-20T14:43:10 that's the idea 2010-11-20T14:43:16 it looks like latex 2010-11-20T14:43:39 Zannick: i used pandoc. it can convert between many formats. in this case i just used its markdown2pdf command, which uses latex in the backend 2010-11-20T14:43:45 cool 2010-11-20T14:43:56 I like 2010-11-20T14:44:06 Zannick: i also used pandoc to write the spec and my haskell starter thing 2010-11-20T14:44:23 now if every bot had a squad of shooters it'd be even more interesting 2010-11-20T14:44:27 i remember the spec 2010-11-20T14:44:32 *** McLeopold has quit IRC (Quit: Leaving) 2010-11-20T14:44:39 Migi32: ah i meant to list that as a possiblity 2010-11-20T14:44:43 *possibility 2010-11-20T14:44:48 i'll add it 2010-11-20T14:45:01 *** smloh1 has quit IRC (Quit: Page closed) 2010-11-20T14:45:06 because I think that the 1 soldier per bot version will just be "solved" 2010-11-20T14:45:10 *** amriedle has joined #aichallenge 2010-11-20T14:45:24 hm 2010-11-20T14:45:32 not sure though 2010-11-20T14:45:37 i'm trying to think this through 2010-11-20T14:46:33 and i want to say that i'm not sure it'll be easy or fast for bots to get kills 2010-11-20T14:46:53 I think you have to carefully decide how fast bullets can be fired. Just setting it to 1 per turn would not work, I think 2010-11-20T14:46:58 it depends on how fast the bullets move 2010-11-20T14:47:02 exactly 2010-11-20T14:47:11 wish I could see the thing :( 2010-11-20T14:47:11 that's slightly different 2010-11-20T14:47:45 jmcarthur: how is it different from Robocode? 2010-11-20T14:49:38 brbr 2010-11-20T14:50:04 Migi32: https://docs.google.com/viewer?a=v&pid=explorer&chrome=true&srcid=0B3_Nay8FIDhVNWMxNjk1OTItMWMwMi00NDMyLWFjNzMtMGI2ZjI0MmU2Yzhj&hl=en 2010-11-20T14:50:04 stuck it to the end hastily 2010-11-20T14:50:04 i'll delete the original later and leave that one 2010-11-20T14:50:36 lol, "which would be uber fun" :P 2010-11-20T14:51:25 so for bots that lose on first turn.. is there any way to tell why? 2010-11-20T14:51:53 j3camero: for one i don't think robocode has any sort of inertia or hidden information, does it? 2010-11-20T14:52:16 have you ever played a board game called robo rally? 2010-11-20T14:52:22 sounds familiar 2010-11-20T14:52:38 RPS map: http://tcp.zeroviz.us/zeroviz_538_19.txt.html 2010-11-20T14:52:45 you are racing to a series of points on a grid 2010-11-20T14:53:14 each bot is facing a certain direction and fires a laser in that direction after every move 2010-11-20T14:54:04 that's only part of it, but we could make which angle is agent is facing be the only angle they can shoot 2010-11-20T14:54:10 hmm 2010-11-20T14:54:24 so they can accelerate, rotate, and fire 2010-11-20T14:54:31 could be tough with continuous space. might make more sense for discrete space 2010-11-20T14:54:39 yeah 2010-11-20T14:55:05 i really think continuous space would be cooler though because it makes game search tougher and also is more fun to watch 2010-11-20T14:55:32 j3camero: eat your steak yet? what if i made a BNF grammar for rules/strategies (no friggin clue here, just brainstorming), kinda like this guy did for some music generation: http://compusition.com/web/news/2008/04/23 2010-11-20T14:56:27 wow the zeroviz.us visualiser is really laggy on my PC 2010-11-20T14:59:46 j3camero: You do agree that it is way too late to change the turn limit right? I just want final official word from you on it so we can put it to rest. :) 2010-11-20T15:01:06 there was official word from amstan already :) 2010-11-20T15:01:11 * lavalamp_ is killed by defense code 2010-11-20T15:01:12 see /topic 2010-11-20T15:02:44 well, got to go, bye :) 2010-11-20T15:02:46 *** Migi32 has quit IRC (Quit: www.ai-contest.com) 2010-11-20T15:03:44 janzert: seems like it's fixed to 200 according to topic 2010-11-20T15:04:00 nobody would want it suddenly changed at this point would they? 2010-11-20T15:04:25 except downwards if their ai dominates early on 2010-11-20T15:04:45 i wonder if anyone here has an ai that would always win if only they had 500 turns? 2010-11-20T15:05:00 foucist: yeah, I don't think we can change it at this point without major problems. I just want to hear j3camero say that. ;) 2010-11-20T15:05:18 Since his is really the final word on it 2010-11-20T15:05:33 "my ai may have lost to you on turn 200, but if it had been allowed infinite time it would've soundly beat yours, mwahaha!" 2010-11-20T15:05:35 janzert: yeah let's leave it at 200 fixed. 2010-11-20T15:05:40 that should be fine 2010-11-20T15:06:02 ...or we could change it to 50 some time this week just to mess with people :-) 2010-11-20T15:06:03 yep, even if we could change it 200 seems to work fairly well 2010-11-20T15:06:12 j3camero: heh, no thank you 2010-11-20T15:06:24 Yeah a1k0n demostrated that 200 is probably sufficient. 2010-11-20T15:06:33 guess the issue that i'm dealing with right now is to figure out a system to represent strategies/decisions and code that up for the bot :( 2010-11-20T15:07:13 janzert: by the way, we just passed 5 million pageviews. *high five* 2010-11-20T15:07:15 the only 'problem' with 200.. is bots that don't take 200 into consideration 2010-11-20T15:07:28 woot 2010-11-20T15:08:01 eg. if bot 1 has 280units and 43growth, and bot2 has 310units and 47growth, on turn 180.. bot2 might spend 100units to get a 5growth planet 2010-11-20T15:08:18 First it was 2500 pageviews, then 50,000, then 1.3 million, now probably 6 million or so. 2010-11-20T15:08:25 That's growth for you. 2010-11-20T15:08:39 yeah, very nice 2010-11-20T15:08:54 Next time we have to get past 20M 2010-11-20T15:09:15 But that will take work, and some more creative thinking on how to turn the "Awesome" dial up to 11. 2010-11-20T15:10:11 *** Zoost has quit IRC (Ping timeout: 264 seconds) 2010-11-20T15:10:16 *** Zoost has joined #aichallenge 2010-11-20T15:11:23 we're probably passing the inflection point of the sigmoid I would imagine 2010-11-20T15:11:24 We gotta do something crazy that nobody expects, and that everybody wants to be part of. Like if we make some sort of strategic MMO that also has an API. That could get pretty intense. 2010-11-20T15:12:53 Yeah the growth really slowed down this time. After a couple 20X multiples, we only grew by probably 6 times this contest. But I don't think we're anywhere close to saturating yet. In theory, why could we not have 2 million participants? There's gotta be 2 million people in the world with internet access, free time, and a bit of CS knowledge who would want to take part. 2010-11-20T15:13:54 j3camero: I doubt the free time part :) 2010-11-20T15:14:08 actually, I don't have (enough) free time, but I'm still competing 2010-11-20T15:14:27 because.. why bother participating, if you have no chance of coming close to winning? 2010-11-20T15:15:00 yeah this contest required quite a bit of work before you got to the point where you could be "proud" of your creation 2010-11-20T15:15:10 burny: I'm having fun, but no chance of winning (unless the tuning of my bot that I'm doing now goes surprisingly well) 2010-11-20T15:15:20 I think most people who join dont have any intention of winning. But in any case, we can eliminate that factor by allowing users to create groups. So they could win their group and be a big fish in a little pond, rahter than shooting to win the global rankings. 2010-11-20T15:15:52 most users will give up if they don't get something nice within the first hour; when you can clearly see that the past 10 hours of work has given you a bot that completely and utterly sucks, it's demotivational 2010-11-20T15:16:11 Hmm what if we did not show the global rankings, only your ranking with respect to the people you have friended on the site. Then the global rakings are only revealed after the deadline? 2010-11-20T15:16:12 (like, if after 10 hours a human can still easily beat your bot) 2010-11-20T15:18:05 actually i disagree with that 2010-11-20T15:18:24 i think that the longer a human can beat your bot the longer it is more "obvious" how you can improve it 2010-11-20T15:18:25 Top 10 players: bocsimacko(3393), _Astek_(3222), BaronTrozo(3199), Raschi(3185), _iouri_(3159), bix0r4ever(3155), wagstaff(3133), smloh1(3125), felixcoto(3091), protocolocon(3053) 2010-11-20T15:18:35 tron benefited from that property a lot 2010-11-20T15:18:47 Yeah Tron was a winning idea. 2010-11-20T15:18:47 it was pretty obvious where the flaws in your bot were 2010-11-20T15:18:52 sure, for people with the mindset of "I want to keep improving my thing" that's beneficial 2010-11-20T15:19:00 but if you want 20x as many participants.. 2010-11-20T15:19:03 But now I am scared to put on Tron again, since Disney is releasing a Tron movie and we would probably get sued. 2010-11-20T15:19:25 Mistmanov: well the other thing is that when it's obvious what you need to change it's more motivating to make that change 2010-11-20T15:19:37 "just a little more" the whole time 2010-11-20T15:19:45 I thoguht the tron movie was like 2years ago? 2010-11-20T15:19:54 always feeling like you are on the verge of being the best bot ever 2010-11-20T15:20:02 burny: it's not out yet... 2010-11-20T15:20:10 true, but there's a difference between "well my bot is pretty awesome but does a dumb thing now and then" and "ok, I ain't getting anywhere" 2010-11-20T15:20:20 Mistmanov: were you around for the tron contest? 2010-11-20T15:20:23 nope 2010-11-20T15:20:36 IT was pretty addictive 2010-11-20T15:20:37 Mistmanov: most bots felt like they were getting somewhere, IMO 2010-11-20T15:20:50 a game that is real easy to play, could be good.. 2010-11-20T15:20:51 i'd say galcon has not quite done that 2010-11-20T15:20:54 like, battleship 2010-11-20T15:21:28 battleship seems too solvable and trivial 2010-11-20T15:21:29 Battleship is a good idea on some criteria. Unfortunately it is trivially solvable (I think) 2010-11-20T15:21:44 the original tron movie is from the 80s 2010-11-20T15:21:49 it's easy to make small modifications to battleship, to make it non trivial 2010-11-20T15:21:50 There is an efficient and perfect strategy. 2010-11-20T15:22:00 eg. every turn, you can make 1 shot, and move 1 boat, 1 space 2010-11-20T15:22:21 or rather, make 2 or 3 shots, and move 1boat 1space 2010-11-20T15:22:46 Hmm what about a game where you direct a naval fleet on a large ocean, with continuous coordinates, and fog of war? 2010-11-20T15:22:56 we should do spy vs. spy! :P 2010-11-20T15:23:16 Part of the challenge would be to try to find your opponent without him seeing you, stalk their boats, and attack in a coordinated fashion. 2010-11-20T15:24:03 ya, that could be neat too.. 2010-11-20T15:24:05 how would you see your opponent without him seeing you? 2010-11-20T15:24:26 when 2 boats are near the same co-ordinates.. it tells both players, that there is an enemy spotted near that boat 2010-11-20T15:24:34 Oh man. Simple RTS with fog of war, sea units, land units, varying sight ranges... ohhh boy. 2010-11-20T15:25:40 Or take that same naval combat idea, make it 3 dimensional instead of 2 dimensional, and set it in space. 2010-11-20T15:25:54 Multi-agent space naval combat. 2010-11-20T15:26:04 eh.. 2010-11-20T15:26:09 You have fast weak ships, and slow strong ones... 2010-11-20T15:26:11 not very accessible I think. 2010-11-20T15:26:20 whoever programs the visualizer will hate you 2010-11-20T15:26:31 i think 3d is best avoided 2010-11-20T15:26:34 We can use VRML. It will be so easy. 2010-11-20T15:26:52 it unnecessarily adds so much extra complication 2010-11-20T15:26:54 boring maths and stuff 2010-11-20T15:27:00 Yeah true. 2010-11-20T15:27:06 2D would be just as good. 2010-11-20T15:27:13 and it also makes the game feel more "open". tight quarters make it faster paced 2010-11-20T15:27:37 which is my objection to something slow like a naval fleet, too ;) 2010-11-20T15:27:49 completely unrelated question that I just thought of: are the organizers students or employees of the university of waterloo? 2010-11-20T15:27:54 not that it has to be *fast* of course 2010-11-20T15:28:14 It's all undergrad students (or was, until I graduated). 2010-11-20T15:28:23 ah k =) 2010-11-20T15:28:40 amstan is a student. he was in 1st year when he won the first GAIC 2010-11-20T15:28:53 hmm... variation of my earlier idea... 2010-11-20T15:29:25 simple old games, that aren't traditionally AI playable.. with some twists 2010-11-20T15:29:29 like, missle command.. 2010-11-20T15:29:33 you can only fire in the direction you are facing, but it's a continuous laser. it won't necessarily kill you until you spend some time stuck in it 2010-11-20T15:29:43 one palyer shoots at the town, while the other player defends 2010-11-20T15:30:00 and you have a very limited speed of rotation 2010-11-20T15:30:19 *** fmeyer has quit IRC (Ping timeout: 245 seconds) 2010-11-20T15:30:20 jmcarthur: neat idea. Sort of like a watergun fight haha 2010-11-20T15:30:32 Is this a draw map? http://tcp.zeroviz.us/zeroviz_538_17.txt.html 2010-11-20T15:30:32 Zannick: kind of like what you were suggetsing maybe... 2010-11-20T15:30:36 *suggesting 2010-11-20T15:31:04 j3camero: with lots of players it would get crazy! 2010-11-20T15:31:17 Yeah man. 2010-11-20T15:31:37 If we did the visualizer a certain way, it would give whole new meaning to the expression "pissing contest" 2010-11-20T15:31:42 i think inertia would be very important 2010-11-20T15:31:50 since it would be the only way to move one way and fire another 2010-11-20T15:32:03 i guess like the way they fight in battlestar galactica 2010-11-20T15:33:19 *** Error323 has quit IRC (Ping timeout: 240 seconds) 2010-11-20T15:33:30 and doing that was squads of "ships" would just make it even crazier 2010-11-20T15:33:40 *with 2010-11-20T15:33:52 Yeah it would. 2010-11-20T15:34:24 also have to make sure not to hit friendlies 2010-11-20T15:34:39 lasers go right through ships, you know ;) 2010-11-20T15:35:22 I really really want to create a simplified MMO Starcraft. 2010-11-20T15:35:26 *** Error323 has joined #aichallenge 2010-11-20T15:35:30 damnit 2010-11-20T15:35:40 crash xD 2010-11-20T15:36:03 Or more like Warcraft 2 actually. But simpler and MMO. 2010-11-20T15:36:20 And then just give it an API so people can easily bot it. Then it's all over. 2010-11-20T15:36:25 where making friends is both positive and negative.. 2010-11-20T15:36:35 because you improve faster, but you also help opponents improve 2010-11-20T15:36:45 Yeah. 2010-11-20T15:37:29 I imagine people who find themselves in the same region in this huge world map would likely try to form federations, and have their bots work together. Could be pretty epic if it's well-balanced. 2010-11-20T15:38:03 The risk is that the best bot grows and just takes over the whole map, and instantly crushed newly spawned players. 2010-11-20T15:38:52 just make it not be pvp 2010-11-20T15:39:07 and, you can only own up to X area 2010-11-20T15:39:15 For the first time my new bot did not lose a game my old bot in my test set of maps.... 2010-11-20T15:39:32 and so you have to balance between owning the best planets.. and the planets which are closest together 2010-11-20T15:39:50 *** boegel has quit IRC (Quit: *poof!*) 2010-11-20T15:40:02 * lavalamp_ is scared to start it on tcp 2010-11-20T15:40:33 lavalamp_: do it... 2010-11-20T15:41:00 It better not get sniped any more >:| 2010-11-20T15:42:00 eek ok let's hope I didn't break anything 2010-11-20T15:43:35 damn lost first game :'( 2010-11-20T15:44:37 pwned 2010-11-20T15:50:37 anyone know about grammatical evolution? 2010-11-20T15:50:49 i've messed with it a bit 2010-11-20T15:50:59 it's about as useful as genetic programming, not much else 2010-11-20T15:51:10 i had some novel ideas to try with it, but never got around to it 2010-11-20T15:52:33 jmcarthur: any idea how to apply something like that to this challenge? build a set of strategies with the grammar stuff? 2010-11-20T15:52:54 that would be part of the challenge, you know 2010-11-20T15:53:38 how's that? you mean it'll take work to do 2010-11-20T15:54:02 when you say challenge there, you odn't mean 'ai challenge' 2010-11-20T15:54:23 ? 2010-11-20T15:54:40 sure i do 2010-11-20T15:54:51 part of the challenge is inventing how the ai works 2010-11-20T15:58:27 gotcha 2010-11-20T15:58:49 Maybe we should do an ai challenge about stock market trading. I basically give out a bunch of raw market tick data for testing purposes, and whoever's bot makes the most money over some test period wins! 2010-11-20T15:59:11 That would be great, because then I would have automated my job :-) 2010-11-20T15:59:13 lol 2010-11-20T15:59:22 Well, not really automated, as much as crowdsourced. 2010-11-20T15:59:24 why not just simplify it more.. 2010-11-20T15:59:34 make it the "guess a number between1 and 10" game 2010-11-20T15:59:37 :P 2010-11-20T16:00:07 "Guess 1/3 of the average of everybody's guess" 2010-11-20T16:00:15 Haha yes some people view trading that way. Efficient market hypothesis and all that. Meh, their loss. 2010-11-20T16:00:22 isn't much of the stockmarket already done by AI? 2010-11-20T16:00:34 In some sense yes, in many ways no. 2010-11-20T16:00:37 j3, 'real' trading requires info other than numbers 2010-11-20T16:00:45 I remember reading something about the dow jones collapsing a while ago because all the computers went crazy 2010-11-20T16:01:01 (well, going down a bunch of points) 2010-11-20T16:01:02 mist, that was in like.. 87 2010-11-20T16:01:28 Most trading is done by algos these days. But that algo is under the command of a human manager. A manager will decide that wants to buy 100k shares of whatever, then the algo will do that in a way that mininmzes the market impact. So it's not really the algo that's making the market moving decisions, still a human. 2010-11-20T16:01:38 http://www.nytimes.com/2010/05/07/business/07markets.html?_r=1&ref=global-home 2010-11-20T16:01:45 Also in May o this year. 2010-11-20T16:01:50 hm ok 2010-11-20T16:01:56 j3camero: that's not the only way 2010-11-20T16:02:14 optimizing execution is important but it's just part of the picture 2010-11-20T16:03:02 The idea behind the stock market AI contest would be to create the smartest money-making trading program, not to carry out someone else's order in an optimal way. 2010-11-20T16:03:26 Basically it would be prop trading, not agency. 2010-11-20T16:03:51 that's the most boring game possible 2010-11-20T16:04:01 Not if you find something that works :-) 2010-11-20T16:04:24 I worked a couple of years on this. 2010-11-20T16:04:25 IM a good AI contest needs competition 2010-11-20T16:04:30 *IMO 2010-11-20T16:04:38 directly competition 2010-11-20T16:05:21 I was talking with some of my coworkers. The idea would be to take the top 10 bots at any given time, and actually be running them in the real market in real time using real money. But we figured that the regulators would not be too happy with us if we did this! :-) 2010-11-20T16:05:56 Using someone else's code to make money would also be a bit shaky I think :p 2010-11-20T16:06:07 (assuming the bots make money..) 2010-11-20T16:06:10 I could just picture that meeting: "we don't actually know how the strategies work. They were submitted over the internet by anonymous people." 2010-11-20T16:07:24 but yeah, a big part of the competition is that you have direct battles between competitors 2010-11-20T16:07:41 just a big leaderboard of who made the most money is less satisfying than "I just defeated the #1 guy!" 2010-11-20T16:07:50 or "I just defeated that punk from the forums!" 2010-11-20T16:07:55 Yes, thats true. 2010-11-20T16:08:27 Maybe create a simulated market with 20 different stocks. Each bot can put bids and offers into the market. 2010-11-20T16:08:57 That could actually work really well if each different "stock" had some actual underlying. 2010-11-20T16:10:22 j3camero: simulation is not easy 2010-11-20T16:10:26 I'd say impossible. 2010-11-20T16:10:27 hmm is there some quantitative real-world phenomenon that we could create derivative contracts for, that the bots could trade in real time? 2010-11-20T16:10:48 sunshine! 2010-11-20T16:10:49 Like maybe the temperatures of major US cities or something. 2010-11-20T16:11:24 j3camero: which regulators? i don't think SEC would have a problem with AI playing the stockmarket, since it already happens for the most part 2010-11-20T16:11:32 s/most part/to some extent/g 2010-11-20T16:11:32 Or the number of pageviews each hour on the ai contest website! 2010-11-20T16:11:52 *** choas has quit IRC (Ping timeout: 240 seconds) 2010-11-20T16:11:55 page views is fakable 2010-11-20T16:12:05 j3camero: i won't object terribly to the holdem poker one 2010-11-20T16:12:09 foucist oh yes they would. I am in Canda, so we have a different regulator. But it's the same deal. 2010-11-20T16:12:12 since that's another way to make money 2010-11-20T16:12:24 page view pump&dump 2010-11-20T16:12:33 j3camero: cool hwere in canada? 2010-11-20T16:12:55 also i wouldn't want to be uploading something that could be potentially making money to a server where those with access to the server could use it as they please 2010-11-20T16:13:03 If we're just running a strategy in the real market submitted by some random person from the internet without vetting it or knowing who they are... that would be big fines. 2010-11-20T16:13:28 where do you think the prize money is gonna come from 2010-11-20T16:14:03 Their strategy could be making money by committing all sorts of crimes, or manipulating market prices, or whatever. There's a lot of rules in the market, and they have to be respected. You have to be able to prove to the regulators that you're following the rules. 2010-11-20T16:14:19 *** bhasker has quit IRC (Ping timeout: 240 seconds) 2010-11-20T16:14:33 But still it is kind of a hilarious idea. 2010-11-20T16:14:37 I'm going to make my bot buy all stocks of "ABC ice cream".. 2010-11-20T16:14:51 which coincidentally I own 80% of, and is 'wroth' a total of 18$ 2010-11-20T16:14:59 but I've been trying to sell it for 50k$ for a few weeks now.. 2010-11-20T16:15:04 Yeah that's just one thing that could go wrong! 2010-11-20T16:15:33 There's a lot of ways to make money by trading in the market. Many of them are not legal haha. 2010-11-20T16:16:21 uh how much do you know about the stock market? 2010-11-20T16:16:41 well i mean, what about market makers 2010-11-20T16:16:48 anyone can make the market 2010-11-20T16:16:52 if they have the clout 2010-11-20T16:17:21 altho yeah there are rules around making the market 2010-11-20T16:18:24 Top 10 players: bocsimacko(3395), _Astek_(3206), Raschi(3169), bix0r4ever(3160), _iouri_(3147), wagstaff(3108), felixcoto(3074), BaronTrozo(3069), protocolocon(3050), davidjliu(3042) 2010-11-20T16:19:09 not anyone really 2010-11-20T16:19:26 broker dealers have a huge advantage 2010-11-20T16:19:49 if you were going to make stock market AI.. obviously the thing to do, would be to make the ai's descisions, not effect the value.. 2010-11-20T16:19:58 they have much lower costs (like 1/3 cent per share max, usually much less than that) vs the average joe who pays $10 per trade. 2010-11-20T16:20:15 and impose strict limits on trades(like you can't sell/buy more than 1% of a companys value in a 10minute period 2010-11-20T16:20:22 and can't own more than 25% total 2010-11-20T16:20:23 etc. 2010-11-20T16:20:38 *** delt0r___ has joined #aichallenge 2010-11-20T16:21:05 but it'd still be highly 'cheatable' 2010-11-20T16:21:29 eg. 3day before deadline, I know that microsoft is going to go up.. so I make my bot buy MS stock 2010-11-20T16:21:34 *** bhasker has joined #aichallenge 2010-11-20T16:21:35 I think the best thing to do with an AI trading contest would be to not use real stocks. Like I said, we could create derivatives based on some real-world indicator that updates every few minutes (like the temperatures in major US cities) and have the bots trade those. That could end up being very realistic! 2010-11-20T16:22:04 j3camero: how about a predictive market 2010-11-20T16:22:10 *** McLeopold has joined #aichallenge 2010-11-20T16:22:16 check out this attack: http://zeroviz.us:8080/canvas.cgi?game_id=15574 2010-11-20T16:22:23 *** delt0r has quit IRC (Ping timeout: 276 seconds) 2010-11-20T16:22:33 foucist: yeah that's pretty much the idea. Basically we would make our own version of InTrade.com, but trade things that update really often. 2010-11-20T16:22:41 AIs already constitute 50% of real traded stocks today - why not just join them? :) 2010-11-20T16:23:11 j3camero: yeah that might be interesting 2010-11-20T16:23:56 There would be two components: (1) how well can you predict the underlying quantity, and (2) how good is your bot at trading. 2010-11-20T16:24:07 can the bot access the internet (lookup forcast highs/lows) :) 2010-11-20T16:24:12 Yeah why not. 2010-11-20T16:24:36 It would have to work kind of like the TCP server model. You would run your bot on your own computer, and it would connect to the market over the internet. 2010-11-20T16:26:13 *** phreeza has joined #aichallenge 2010-11-20T16:26:27 blah.. I can't figure out why my bot always loses on first turn :( 2010-11-20T16:26:32 We would have to stop people from making 50 accounts and using them for a pump and dump. Oh man it could end up being pretty wild, lol. 2010-11-20T16:26:56 Okay going for a walk outside now. Talk to you guys later. 2010-11-20T16:28:20 what was the link to vew logs instead of visualizer on the official server? 2010-11-20T16:28:24 view* 2010-11-20T16:30:32 *** Gromgull has quit IRC (Quit: Leaving.) 2010-11-20T16:30:37 Mjothvitnir: ping 2010-11-20T16:30:47 pong 2010-11-20T16:31:38 you beeped me? 2010-11-20T16:31:59 phreeza: game_info.php 2010-11-20T16:32:24 medrimonia was having some issues earlier with jbotmanager (that is yours right) 2010-11-20T16:32:34 janzert: thanks 2010-11-20T16:33:09 Mjothvitnir: yes jbotmanager is mine.... 2010-11-20T16:33:12 as in my code 2010-11-20T16:33:16 Mjothvitnir: yep, but I think it's only because I can't use windows properly 2010-11-20T16:33:45 *** bhasker has quit IRC (Quit: bhasker) 2010-11-20T16:33:57 when I run it from a dos box I get error output, but med wasn't, so I was wondering if you knew why and/or could help us figure out what was going on, but you weren't on 2010-11-20T16:34:35 I wish there were more people on McLeopold's server 2010-11-20T16:34:55 "good" people or "bad" people 2010-11-20T16:35:01 well what was the error... 2010-11-20T16:35:06 either one would be fine 2010-11-20T16:35:10 there is the odd outstanding issue 2010-11-20T16:35:20 lavalamp_: crazy or normal? 2010-11-20T16:35:23 offically --i won't have time to fix it... 2010-11-20T16:35:32 normal is the one I'm on 2010-11-20T16:35:47 python bot didn't seem to be starting, for me I could see an error so I could see why, but no error for med means no idea at all 2010-11-20T16:36:25 most are on benzedrine, just move there 2010-11-20T16:36:49 with a name like medrimonia I want to say "she", but I don't know so... dang it, which is it? :) 2010-11-20T16:37:07 It's "he" ^^ 2010-11-20T16:37:21 pics! 2010-11-20T16:37:25 thank you, now I don't have to worry about it :) 2010-11-20T16:37:37 Yeah that's easier sometimes 2010-11-20T16:37:37 yeah I'm there too 2010-11-20T16:37:42 Mjothvitnir, medrimonia : i have run python bots without issue... so I don't know... Do you give a "python MyBot.py" command line 2010-11-20T16:37:47 maybe I'll try your crazy one 2010-11-20T16:37:56 if you do that it won't work well on windows 2010-11-20T16:38:06 you need to put it all in a "bat" file 2010-11-20T16:38:15 ho... that's why 2010-11-20T16:38:27 or/and include the full path to python 2010-11-20T16:38:55 becuase when java runs a command, it won't get a full enviourment... ie paths etc are not always set up 2010-11-20T16:39:09 bit like cron jobs on linux 2010-11-20T16:39:18 yay.. I managed to get my bot to win a match 2010-11-20T16:39:29 I think that I'll keep my tests on tcp on my linux computer that'll be easier 2010-11-20T16:40:10 *** bhasker has joined #aichallenge 2010-11-20T16:40:45 well, I seem to have broken into positive elo on the tcp server 2010-11-20T16:40:47 when I tried to see if it would work for me I was getting an error on the "current_line = raw_input()" of the staterpack_bot would that be fixed by the bat-filing? 2010-11-20T16:41:39 Mjothvitnir: don't really know... what it the error? it all sounds a bit odd 2010-11-20T16:41:57 does python have all the imports it needs 2010-11-20T16:42:22 I can start the python bot but it gives me an error "File"..." line X in main[;] current_line = raw_input() [;] EOFError: EOF when reading a line 2010-11-20T16:42:24 is it the right version (ie does it run on another game engine? ) 2010-11-20T16:42:34 just submited an overly-agressive bot, with absolutely no defense, or 'front lines' capabilities 2010-11-20T16:43:06 but it should counter-attack.. rather than defend 2010-11-20T16:43:09 That sounds like some kind of error... that my manager picked up, then closes all the streams 2010-11-20T16:43:29 I know I didn't try that, I can't remember if I tried my usualy cat-map|gawk-go|bot test 2010-11-20T16:43:44 *** bhasker has quit IRC (Client Quit) 2010-11-20T16:44:03 jbotmanager sends a lot of interesting msgs to standard output and error... you could try to start it from a "shell" or dos (cmd) 2010-11-20T16:44:28 we were but med wasn't getting error msgs 2010-11-20T16:44:39 very odd... 2010-11-20T16:44:59 I e.printStackTrace() or all exceptions 2010-11-20T16:45:06 on all 2010-11-20T16:45:49 you *should* get some kind of time out msg or illegal command msg etc... 2010-11-20T16:47:29 *** Naktibalda has joined #aichallenge 2010-11-20T16:47:34 36 players on the tcp server.. and out of my last 10 games I was paired with the same guy 8 times, and lost 6 of them :/ 2010-11-20T16:48:35 it's weird... one time when I submited, I got 10games in 6seconds 2010-11-20T16:48:38 on teh official server 2010-11-20T16:48:39 I don't run a python bot, so fixing my "problem" isn't necessary to me. (I get the same error if I cat|gawk|python as I do in jbotm, but not if I just run the bot and hand type planets and go) 2010-11-20T16:48:44 is that just an anomoly? 2010-11-20T16:49:16 the official server does weird stuff yeah. It used to be worse I think. No games for hours, and then 4 in the same second (!) 2010-11-20T16:49:46 Mjothvitnir, medrimonia: sorry i couldn't be more help. But i have to go. I will be online tomorrow 2010-11-20T16:49:56 thanks 2010-11-20T16:50:00 are there any contendors with mersenne, for medium to high quality RNG? 2010-11-20T16:50:27 burny: Lots, and are faster.... but ask tomorrow 2010-11-20T16:50:33 ;_ 2010-11-20T16:50:36 ;) 2010-11-20T16:50:54 why tomorrow? busy? 2010-11-20T16:52:14 can you run multiple bots with the same name on the tcp server? 2010-11-20T16:52:35 compare http://72.44.46.68/getplayer?player=flanders.28b and http://72.44.46.68/getplayer?player=mistmanov 2010-11-20T16:52:41 yes, you can 2010-11-20T16:52:50 I've got a solid block of flanders as opponent, but he has loads of other opponents in between 2010-11-20T16:53:03 ah ok. Well, that explains why I'm constantly playing against him -_- 2010-11-20T16:53:21 does anyone here know much or deal much with generating grammars of player actions with this ai challenge? 2010-11-20T16:53:44 FSMs? BNF? transducers? :P 2010-11-20T16:54:39 *** flag has joined #aichallenge 2010-11-20T16:59:26 *** p4p4 has quit IRC (Quit: ChatZilla 0.9.84 [SeaMonkey 2.0a3/20090223135443]) 2010-11-20T17:03:51 *** jmcarthur has quit IRC (Quit: WeeChat 0.3.2) 2010-11-20T17:06:15 *** jmcarthur has joined #aichallenge 2010-11-20T17:11:08 *** medrimonia1 has joined #aichallenge 2010-11-20T17:11:15 *** medrimonia has quit IRC () 2010-11-20T17:11:30 *** medrimonia1 is now known as medrimonia 2010-11-20T17:15:35 *** Mistmanov has quit IRC () 2010-11-20T17:18:26 Top 10 players: bocsimacko(3397), bix0r4ever(3172), _Astek_(3171), _iouri_(3165), Raschi(3161), smloh1(3105), wagstaff(3101), felixcoto(3092), protocolocon(3053), dmj111(3042) 2010-11-20T17:29:33 *** Palmik has quit IRC (Remote host closed the connection) 2010-11-20T17:32:51 is there any way to get your ELO score to reset? 2010-11-20T17:33:54 cause I have a bot that should be around 2k probably.. 2010-11-20T17:34:03 and it's at just 12 right now 2010-11-20T17:34:12 and it's climbing up super slow 2010-11-20T17:35:00 *** jaspervdj has quit IRC (Quit: NEVER GONNA GIVE YOU UP NEVER GONNA LET YOU DOWN) 2010-11-20T17:37:49 burny: no not really, but what username out of curiosity? 2010-11-20T17:38:15 it's low cause I had a bunch of broken code before 2010-11-20T17:38:19 so I lost like 200 in a row 2010-11-20T17:38:48 use a new name 2010-11-20T17:38:51 yeah, I'd like to watch it climb actually 2010-11-20T17:38:55 I think that you'll climb up "fast enough" with the system 2010-11-20T17:39:00 please don't 2010-11-20T17:39:01 http://ai-contest.com/profile.php?user_id=13271 2010-11-20T17:39:24 oh, on official?, nvm 2010-11-20T17:39:25 ya.. it's 'fast enough'.. but I'd rather not have to wait a couple hours lol 2010-11-20T17:39:53 hmmm in fact... 2010-11-20T17:40:00 janzert: put up a paypal site, allow donation's of $10 to reset an elo 2010-11-20T17:40:20 ahh, heh you've removed our 0 elo anchor :P 2010-11-20T17:40:33 I think that the problem is that you're playing against people with very low elo 2010-11-20T17:41:16 med.. obviously.. 2010-11-20T17:41:20 thus the desire to 'reset it' 2010-11-20T17:41:34 burny: but your bot will get stronger beating up the weak 2010-11-20T17:41:35 you've already gain about 100 ranks :) 2010-11-20T17:41:42 and it'll be warmed up for teh stronger opponents 2010-11-20T17:41:51 lol foucist 2010-11-20T17:41:56 if only that were the case 2010-11-20T17:42:09 what! your bot doesn't operate on anime principles? ;) 2010-11-20T17:42:27 not until it's ELO get's high enough.. 2010-11-20T17:42:28 janzert: he's not getting real games? 2010-11-20T17:42:33 has to be atleast 9000 2010-11-20T17:42:45 heh 2010-11-20T17:42:59 isn't the highest 3400 right now? 2010-11-20T17:43:04 so I'm 3 for 3.. beat someone who's ranked 4289.. and I'm 4307? 2010-11-20T17:43:08 I don't think those should count against his last played time. 2010-11-20T17:43:19 McLeopold: hmm? 2010-11-20T17:43:26 nikiburny 2010-11-20T17:43:32 3 games, all crashes 2010-11-20T17:44:33 they have to though or all the ones just above the crash bots will be playing a huge number of games 2010-11-20T17:44:44 and that's bad? 2010-11-20T17:44:49 yes 2010-11-20T17:44:58 it'll starve the rest of the bots 2010-11-20T17:45:04 I don't think it will 2010-11-20T17:45:11 http://72.44.46.68/canvas?game_id=1111418 2010-11-20T17:45:16 the play who didn't crash get's a game right away 2010-11-20T17:45:24 I almost losted this one, really funny 2010-11-20T17:45:27 if they get a real game, then they are out of the queue 2010-11-20T17:45:40 it wouldn't affect the pairings too much 2010-11-20T17:46:02 why skew it 2010-11-20T17:46:27 but I am wondering why it is taking so long for him to get paired again 2010-11-20T17:46:43 make it skew, only until 10 games are played(or there's atleast 2losses and 2wins) 2010-11-20T17:46:50 it should be skewed for working bots 2010-11-20T17:47:12 a bot that crashes all the time, should only play half the time, which this would do 2010-11-20T17:47:12 but it will be skewed toward partially working and really low ranked working bots 2010-11-20T17:47:24 no it wouldn't 2010-11-20T17:47:30 if a bot crashes 50%, then they won't play as much, maybe 75%, but it's still better. 2010-11-20T17:47:30 my recomendation would be.. 2010-11-20T17:47:41 even the crash all the time bots would play more than the top of the rankings 2010-11-20T17:47:48 when a new bot enters: let it play more often, until it has won2 games, and lost 2 games: or played a total of 10games 2010-11-20T17:47:51 a bot that has more than 50% crashes shouldn't play more than 10 games I thinks 2010-11-20T17:47:57 janzert: I don't think we are on the same page with my idea 2010-11-20T17:48:19 more than 10 games at all 2010-11-20T17:48:23 janzert: crash bot does not replay, his opponent does 2010-11-20T17:48:39 yes, but chances are better than average that he will play another crash bot 2010-11-20T17:48:52 so if both crash, then no replay 2010-11-20T17:48:58 they still get their turns 2010-11-20T17:49:09 It's just bots that don't crash get a real turn 2010-11-20T17:49:11 since if you were likely to play a crash bot to start with you are still likely to play one again 2010-11-20T17:49:15 I agree, about not making low rank bots play more... but when they(and anyone else also), re-submit.. they should get more plays, at first 2010-11-20T17:49:27 no, the rank would go up into the good bot range 2010-11-20T17:49:27 (maybe I don't see something, but I can't see what's the interest of letting bot who crashes every game play) 2010-11-20T17:49:43 leo.. really bad bots that don't crash, wouldn't go up 2010-11-20T17:49:52 true 2010-11-20T17:49:58 ranks only recalculate every 15 minutes so not likely if you are playing these games back to back 2010-11-20T17:50:01 but at least they would get some good games 2010-11-20T17:50:19 and how do you tell the difference?.. the only way I see, is that bots with low total games played, get favoued in playing matches in general 2010-11-20T17:50:24 then schedule the bot for after the next rank change 2010-11-20T17:51:03 burny: a burst of games at resubmit would tend to incentivize people to resubmit as frequently as possible and starve the long term submissions 2010-11-20T17:51:05 really, if someone has a crummy bot, and loses 3/4 good games, and wins 6 crash bots.. they have enough info to improve themselves 2010-11-20T17:51:35 jan: they would only get 4 games bursted 2010-11-20T17:51:42 and would lose all their history 2010-11-20T17:51:52 or heck, make it be until they have 1win and 1loss 2010-11-20T17:51:59 yes, and it's 4 times more than they would get otherwise 2010-11-20T17:52:11 well, still.. so what? 2010-11-20T17:52:23 if someoen is actively participating, and wants to see what it's doing.. 2010-11-20T17:52:36 But isn't it possible to simply get out of the schedule the bots who crashed 10 times in a row? 2010-11-20T17:52:38 janzert: i would suggest getting a game against a crashed bot is not getting a game 2010-11-20T17:52:39 why not let them get more games played, than someone who is waiting until tomorrow to see what's up? 2010-11-20T17:53:21 how does winning against someone ranked 4289, make me be 4307 though? 2010-11-20T17:53:22 I want to incentivize resubmission only when the player thinks he has made real changes, even the pairing of new submission at there old levels is kinda shaky 2010-11-20T17:53:24 janzert: I think the definition of "garenteed a game" should be a game that is actually played. 2010-11-20T17:54:03 what if the bot sucks REAL bad.. and can't come close to beating any other non-crash bot? 2010-11-20T17:54:33 or, heck.. 2010-11-20T17:54:40 *** medrimonia1 has joined #aichallenge 2010-11-20T17:54:44 we're going to have to disagree on it, but for sure in the next contest hopefully crash bots will be eliminated before they make it into the main competition 2010-11-20T17:54:44 only allow the burst to happen up to once per hour 2010-11-20T17:54:51 i.e. they will be failed submissions 2010-11-20T17:55:44 burny: yeah, there are a lot of complexities that would need to be added on to make it work fairly and it's certainly not something I'm going to dig into now 2010-11-20T17:56:02 it's definitely a realm fraught with unintended consequences 2010-11-20T17:56:04 ya, I wasn't sugesting it happen 'right now' :P 2010-11-20T17:56:26 janzert: I'll suggest a scheme on the behind the scenes forum later, but I think I could get something to be fair and work. 2010-11-20T17:56:59 *** medrimonia has quit IRC (Ping timeout: 245 seconds) 2010-11-20T17:57:28 have a 'play more games' button.. and every 6hours, they get one more 'play more games' currency(with a cap of 5, or some such) 2010-11-20T17:58:00 janzert: yes, with that paypal site, a $10 donations gets you 5 extra games instantly. :P 2010-11-20T17:58:07 hehe 2010-11-20T17:58:09 lol leo 2010-11-20T17:58:53 I think that anything that allows people who are actually LOOKING at the resuls of current games, to get more current games.. is a good thing 2010-11-20T17:58:57 btw, burny your bot last game is 2 minutes more recent than the current least recent submission so should play any time now 2010-11-20T17:59:20 and there it went 2010-11-20T17:59:24 can we see that? 2010-11-20T17:59:36 and you got 3 turns :/ 2010-11-20T17:59:39 that would be nice. "1000 people ahead of you in the queue" 2010-11-20T18:00:14 for what it's worth, I would have called the game for him at 2 turns 2010-11-20T18:00:25 hmm, except you will get an "early" game half the time 2010-11-20T18:00:50 who will? 2010-11-20T18:00:51 so I get a game roughly every 25mins? 2010-11-20T18:00:55 everyone 2010-11-20T18:01:02 and it'll take about.. 8hours, before I have a 'real' game? 2010-11-20T18:01:08 *** bhasker has joined #aichallenge 2010-11-20T18:01:25 so far I got 4 wins.. and climbed a total of 30ish in the ranks 2010-11-20T18:01:30 janzert: so in my scheme, only a bot that was scheduled to have the game gets a do over of the other guy crashed 2010-11-20T18:01:58 *** Appleman1234 has joined #aichallenge 2010-11-20T18:02:01 extrapolating.. it'll be around 80games to get into the top1000.. 2010-11-20T18:02:16 so.. maybe I'll know somethign useful about my bot before the contest is over? 2010-11-20T18:02:18 goody 2010-11-20T18:02:28 it get the rank thing though, you would have to start climbing instantly to pull yourself out of the "crash zone" 2010-11-20T18:02:38 burny: you've climbed over 100 ranks already 2010-11-20T18:02:49 you were dead last right? 2010-11-20T18:02:52 no 2010-11-20T18:02:59 lots of other bots were crashing all the time too 2010-11-20T18:02:59 Maybe truescore until the final contest, then bayeselo? 2010-11-20T18:03:11 oh, you must have bumped up before the resubmission 2010-11-20T18:03:34 I remember seeing you as last at one point 2010-11-20T18:04:02 janzert: how about 24 hours results to produce elo for rankings, but 72 hour results shown? 2010-11-20T18:04:23 then you can select better matches because of faster rank updates? 2010-11-20T18:04:51 McLeopold: the database isn't really setup for it 2010-11-20T18:05:13 jan, any chance you can manually reset my rank? 2010-11-20T18:05:13 :( 2010-11-20T18:05:29 but yeah, there are other reasons I'd like to split the ranking calculation for display and the calculation for pairing 2010-11-20T18:05:59 janzert: I suppose the "fix" to make good bots play near their last rank is bad for bad bots? 2010-11-20T18:06:00 I would like to include LOS in display but it takes another 15 minutes to calculate 2010-11-20T18:06:09 LOS? 2010-11-20T18:06:14 McLeopold: yes, basically 2010-11-20T18:06:40 well, bad in the sense that their author's won't like it 2010-11-20T18:06:49 you could cap it a zero 2010-11-20T18:06:49 likelihood of superiority 2010-11-20T18:07:10 bad bots should play against bad bots.. but if they become non-bad, they should be able to climb ranks hella faster than currently 2010-11-20T18:07:35 Okay, well, I think burny not getting a full game is really unfair. Even though I'm not sure the best way to fix it. 2010-11-20T18:08:09 I really don't think the current rate is that bad although I can understand your frustration and I think you'll be getting full games within the next few cycles 2010-11-20T18:08:30 even if it's a full round against a suck bot.. it tells me nothing 2010-11-20T18:08:49 other than you don't suck 2010-11-20T18:09:00 which we still can't tell :P 2010-11-20T18:09:11 well.. in my opinion it sucks.. 2010-11-20T18:09:24 but subjectively, it sucks WAY less than it used to 2010-11-20T18:09:51 assuming my bot is perfect, and NEVER loses.. 2010-11-20T18:09:58 how long is it gona take to get up to rank 1? 2010-11-20T18:10:03 7weeks? 2010-11-20T18:10:24 no, you're rate of climb won't be linear if you don't lose ever 2010-11-20T18:10:27 No burny, I think that the beginning will be the harder 2010-11-20T18:10:41 I recall someone saying something about beating a bot with a lower rating then themselves, and that reducing their elo 2010-11-20T18:11:06 isn't the begining suposed to be the easiest? 2010-11-20T18:11:28 yeah, but they skewed the bots you start against 2010-11-20T18:11:34 burny: it shouldn't although there can be that appearance depending on what past games happen to get removed in the same ranking cycle 2010-11-20T18:11:43 I've seen some players getting into top top 50 without having play a game against someone of the top50. 2010-11-20T18:12:16 I think that once you'll be playing real games, if you don't lose too much, you'll start climbing fast enough 2010-11-20T18:12:36 I anticipate being around a 1200-1400elo 2010-11-20T18:13:10 maybe more like 160.. I duno 2010-11-20T18:13:12 I'll to be around 1200 1400 elo on tcp... 2010-11-20T18:13:13 1600 I mean 2010-11-20T18:14:27 *** medrimonia1 is now known as medrimonia 2010-11-20T18:15:24 *** bobng has quit IRC (Quit: ZNC - http://znc.sourceforge.net) 2010-11-20T18:18:26 Top 10 players: bocsimacko(3417), Raschi(3185), _Astek_(3181), bix0r4ever(3163), _iouri_(3148), felixcoto(3124), wagstaff(3095), smloh1(3089), dmj111(3059), protocolocon(3053) 2010-11-20T18:19:15 ah, so whover was complaninga bout 2800 beats 2600, resulting in 2780.. it was just cause someone who they beat in a 2800 vs 2900, resubmited? 2010-11-20T18:20:41 should get another useless win soon! 2010-11-20T18:22:20 burny: yes, something like that anyway 2010-11-20T18:23:50 Error323: I'm still snipping you =P 2010-11-20T18:24:04 which version 2010-11-20T18:24:16 18.0-2... 2010-11-20T18:24:22 that's the version of last night hehe 2010-11-20T18:24:54 what's the meaning of g4a95d58 ? 2010-11-20T18:25:09 hmm, maybe I will could a "place in queue" count to profile pages 2010-11-20T18:25:31 I think that would be nice 2010-11-20T18:25:37 That would be very nice indeed 2010-11-20T18:25:50 medrimonia: beginning of the git sha-1 hash 2010-11-20T18:25:50 it's a quick query, it won't be quite as inaccurate as I was first thinking and I don't think people will mind too much when they play earlier than expected ;) 2010-11-20T18:25:58 If it helps, you can turn my bot off. :) 2010-11-20T18:26:11 I haven't looked at an official game in weeks 2010-11-20T18:26:26 although hopefully they won't complain when other people play earlier than expected 2010-11-20T18:27:01 Error323: thanks 2010-11-20T18:27:38 what's flanders nickname on official, does someone know? 2010-11-20T18:30:40 DISCO: http://72.44.46.68/canvas?game_id=1112187 2010-11-20T18:31:08 XD 2010-11-20T18:33:38 *** Rubicon-|-Cross has joined #aichallenge 2010-11-20T18:39:06 *** bhasker has quit IRC (Quit: bhasker) 2010-11-20T18:45:35 *** mceier has quit IRC (Quit: leaving) 2010-11-20T18:48:06 *** McLeopold_ has joined #aichallenge 2010-11-20T18:50:10 *** bobng has joined #aichallenge 2010-11-20T18:51:29 *** McLeopold has quit IRC (Ping timeout: 255 seconds) 2010-11-20T19:03:32 its so stupid 2010-11-20T19:03:53 when i'm about to lose a planet i start capping a neutral in its neighbourhood 2010-11-20T19:03:59 >_< instead of taking it back 2010-11-20T19:04:01 dohhhhh 2010-11-20T19:08:03 *** tapwater has joined #aichallenge 2010-11-20T19:11:36 Safe to say that is not optimal behavior Error323 2010-11-20T19:11:44 What is causing that? 2010-11-20T19:11:52 well 2010-11-20T19:11:56 i capture when i lose 2010-11-20T19:12:01 when i'm losing* 2010-11-20T19:12:16 but not in a good way apparently 2010-11-20T19:12:27 i need a more sophisticated algo for neutral capping 2010-11-20T19:12:32 imo the hardest part of the game :/ 2010-11-20T19:14:20 neutral capping? 2010-11-20T19:15:14 yes 2010-11-20T19:18:26 Top 10 players: bocsimacko(3414), Raschi(3180), _iouri_(3166), bix0r4ever(3162), _Astek_(3129), felixcoto(3127), wagstaff(3094), smloh1(3090), davidjliu(3073), protocolocon(3057) 2010-11-20T19:20:35 *** rebelxt_ has joined #aichallenge 2010-11-20T19:22:36 *** Rubicon-|-Cross has joined #aichallenge 2010-11-20T19:24:43 janzert: Most likely to play next game within the next 2899 games 2010-11-20T19:24:48 niiiiiic 2010-11-20T19:24:50 e 2010-11-20T19:24:54 :D 2010-11-20T19:25:08 :) 2010-11-20T19:25:17 *** Rubicon-|-Cross has joined #aichallenge 2010-11-20T19:25:25 wow, mega1is on quite a winning streak right now 2010-11-20T19:25:42 *** Rubicon-|-Cross has quit IRC (Client Quit) 2010-11-20T19:27:25 janzert: could you also add the next rank update countdown? :P 2010-11-20T19:27:37 not really 2010-11-20T19:29:01 they start every quarter hour currently though, so basically you can use that as the point where games are taken for the next update and then it takes 8-10 minutes to actually update 2010-11-20T19:29:54 http://zeroviz.us:8080/E323-18.0-4-g5fe5eca.html lol -.- 2010-11-20T19:31:09 janzert: ahh i see 2010-11-20T19:33:26 *** smloh1 has joined #aichallenge 2010-11-20T19:33:37 the next game countdown is pretty sweet 2010-11-20T19:34:26 janzert: +1, countdown is nice little touch! 2010-11-20T19:34:46 *** yasith_ has quit IRC (Ping timeout: 240 seconds) 2010-11-20T19:34:52 thanks, hopefully it won't lead to people mashing refresh constantly :P 2010-11-20T19:35:47 exactly what I was just doing stress testing it for a minute ;-) will stop now heh 2010-11-20T19:36:01 hehe 2010-11-20T19:36:36 at least maybe people will get into the habit of only doing rapid refreshes as the countdown gets low 2010-11-20T19:40:30 aaamazing, the amount of times i team up against "crapbot" on tcp 2010-11-20T19:41:49 *** phxtrack has joined #aichallenge 2010-11-20T19:44:32 i do like all green http://zeroviz.us:8080/E323-18.0-4-g5fe5eca.html ^_^ 2010-11-20T19:44:40 someone join there please 2010-11-20T19:46:11 *** Utkarsh has quit IRC (Ping timeout: 264 seconds) 2010-11-20T19:47:39 *** yasith_ has joined #aichallenge 2010-11-20T19:49:40 should also add a time estimate(based on average games per minute) 2010-11-20T19:51:13 *** Utkarsh has joined #aichallenge 2010-11-20T19:51:42 even if it's not a proper number.. a rough estimate based on manually watching it for 10minutes or something 2010-11-20T19:56:33 *** rebelxt_ has quit IRC (Quit: Page closed) 2010-11-20T19:57:49 *** Utkarsh has quit IRC (Ping timeout: 245 seconds) 2010-11-20T19:58:17 burny: I debated adding that but because the gpm varies so wildly I decided not to 2010-11-20T19:58:56 it's the same problem file download time left indicators have :/ 2010-11-20T19:59:28 ya.. don't do a progress bar.. 2010-11-20T19:59:37 even, don't neccessarily do the math in code.. 2010-11-20T19:59:40 just like.. 2010-11-20T20:00:00 Most likely to play next game within the next 4377 games. (approximately 280games per minute) 2010-11-20T20:00:30 the games per minute regularly varies between 40 and 170 for the 5 minute running average 2010-11-20T20:01:37 so telling them what it happens to be now can be quite misleading 2010-11-20T20:01:38 well, it looks like roughly 35minutes per 2200games to me 2010-11-20T20:02:28 @game_stats 2010-11-20T20:02:29 janzert: I do not know about 'game_stats', but I do know about these similar topics: 'gamestats' 2010-11-20T20:02:35 @gamestats 2010-11-20T20:02:36 janzert: gamestats = http://ai-contest.com/game_stats.php 2010-11-20T20:02:56 gives the 5 minute, hour and 24 hour running averages 2010-11-20T20:03:18 the 1hour or 24hour average should be plenty sufficient 2010-11-20T20:03:44 *** Utkarsh has joined #aichallenge 2010-11-20T20:03:52 and when the hour is 90 and they happen to be 300 games out and the 5 minute is 40? 2010-11-20T20:04:04 just let people get a count I think :) 2010-11-20T20:04:36 the 5min is less accurate then the 1hour, at predicting remaining time 2010-11-20T20:04:51 *** phreeza has quit IRC (Quit: Computer has gone to sleep) 2010-11-20T20:04:57 not when you near the top of the queue 2010-11-20T20:04:58 unless processing power available fluctuates? 2010-11-20T20:05:11 oh, it processes it in order, top to bottom? 2010-11-20T20:05:11 no but game times do 2010-11-20T20:05:28 sort of 2010-11-20T20:05:53 it always takes the player that has played least recently for the first player and then pairs them with someone else 2010-11-20T20:06:18 which is basically a virtual queue 2010-11-20T20:06:29 isn't that order fairly random? 2010-11-20T20:07:11 it sorts itself out since you are more likely to play someone near your rank for the second player 2010-11-20T20:07:25 so people in rank will tend to have played around the same times last 2010-11-20T20:07:37 emergent behavior :) 2010-11-20T20:08:55 true 2010-11-20T20:09:53 btw, looks like you would have had your first full game as the 6th one if you hadn't crashed 2010-11-20T20:10:25 since you ended up paired with the same one again as the 8th game and it played the full game 2010-11-20T20:10:31 *** Utkarsh has quit IRC (Ping timeout: 252 seconds) 2010-11-20T20:10:44 it'd still be a useless game to watch(assuming that bot isn't a crasher) 2010-11-20T20:11:32 is there no way to find out WHY it fails though? 2010-11-20T20:12:01 not with the current system unless you can reproduce it locally 2010-11-20T20:12:26 and I'm not likely to get time to change it to something nicer before the end of the contest at this point :( 2010-11-20T20:12:59 cause I'm running, almost exactly the default ai 2010-11-20T20:13:09 changed like 2 lines of code 2010-11-20T20:15:26 *** Utkarsh has joined #aichallenge 2010-11-20T20:16:58 bah, looks like at some point the way draw games are recorded was changed and they aren't displaying in the list of games on profile pages :( 2010-11-20T20:17:19 kind of surprised no one has ever said anything 2010-11-20T20:17:51 i think i saw someone said something about that, might have been dmj111 2010-11-20T20:18:27 Top 10 players: bocsimacko(3418), Raschi(3173), bix0r4ever(3164), _iouri_(3153), _Astek_(3130), felixcoto(3113), smloh1(3087), wagstaff(3085), davidjliu(3073), dmj111(3060) 2010-11-20T20:18:45 wait what? 2010-11-20T20:19:08 you don't see any draw games on the main site? 2010-11-20T20:19:25 they are included in the rankings but never displayed in the list of games in profiles 2010-11-20T20:19:26 how to blow a 4 ship lead: http://72.44.46.68/canvas?game_id=1113722 2010-11-20T20:19:28 >.< 2010-11-20T20:19:51 ah yeah, that's weird 2010-11-20T20:20:26 *** Utkarsh has quit IRC (Ping timeout: 276 seconds) 2010-11-20T20:23:53 but how would we notice . . . 2010-11-20T20:24:31 because there were seemingly no draws occurring? 2010-11-20T20:24:33 hm.. 2010-11-20T20:24:34 Warning: mysql_connect() [function.mysql-connect]: Too many connections in /var/www/vhosts/ai-contest.com/www/mysql_login.php on line 17 2010-11-20T20:24:57 that's when I refresh my profile 2010-11-20T20:24:59 yeah, sorry it should come back shortly 2010-11-20T20:25:00 same error here 2010-11-20T20:25:12 well before just recently (we should have a game in the next 100 games, wait (draw), you should have a game in the next 4000, WHERE's MY GAME) 2010-11-20T20:25:21 But we don't EXPECT draws 2010-11-20T20:25:39 it didn't like the query I tried to use to include showing draws 2010-11-20T20:26:04 Mjothvitnir: I would have expected the players to notice how many they were getting on tcp and not getting any at all on main site 2010-11-20T20:26:09 that's all 2010-11-20T20:26:30 *** Utkarsh has joined #aichallenge 2010-11-20T20:26:33 not that surprising it was missed I suppose 2010-11-20T20:27:02 janzert: how many hits per day do you get? 2010-11-20T20:27:22 not really sure, I think amstan can answer that though 2010-11-20T20:27:43 I know Jeff said earlier we had surpassed 5 million total 2010-11-20T20:28:30 *** Rubicon-|-Cross has joined #aichallenge 2010-11-20T20:32:07 xD 2010-11-20T20:32:36 79/90/2 on tcp never gonna miss those 2 :) 2010-11-20T20:33:28 *** Utkarsh has quit IRC (Ping timeout: 240 seconds) 2010-11-20T20:34:35 janzert: from the code, I wasn't even sure if they are recorded. 2010-11-20T20:35:15 dmj111: yeah, they get recorded but winner and loser id are both set to 0 which messes up the profile listing code 2010-11-20T20:35:48 ahh, ok. 2010-11-20T20:36:40 oh boy.. the last 4 games I had, was against the same 2 people 2010-11-20T20:37:15 *** amstan_ has joined #aichallenge 2010-11-20T20:37:15 *** ChanServ sets mode: +o amstan_ 2010-11-20T20:37:21 burny: I'm working on changing that :) 2010-11-20T20:37:21 *** Rubicon-|-Cross has quit IRC (Quit: Leaving) 2010-11-20T20:37:42 * lavalamp_ thinks janzert just quadrupled load on the server with people constantly refreshing 2010-11-20T20:37:51 lol lava 2010-11-20T20:37:58 not that I'm complaining ^^ 2010-11-20T20:38:07 *** phxtrack has left #aichallenge 2010-11-20T20:38:07 lavalamp_: i hate you 2010-11-20T20:38:09 _#$(*%)#$(*% 2010-11-20T20:38:14 *** Utkarsh has joined #aichallenge 2010-11-20T20:38:23 it's not my fault you didn't cap enough ^^ 2010-11-20T20:38:38 *** McLeopold_ has quit IRC (Ping timeout: 276 seconds) 2010-11-20T20:38:43 *** amstan has quit IRC (Remote host closed the connection) 2010-11-20T20:38:53 I was watching that live on jbotmanager... soooo stressful 2010-11-20T20:39:00 hehe 2010-11-20T20:39:00 *** amstan_ is now known as amstan 2010-11-20T20:39:21 i cap when i shouldn't and don't when i should 2010-11-20T20:39:22 its awesome 2010-11-20T20:40:57 *** Cold-Phoenix has joined #aichallenge 2010-11-20T20:41:18 hehe 2010-11-20T20:43:50 *** Appleman1234 has quit IRC (Ping timeout: 276 seconds) 2010-11-20T20:44:18 http://72.44.46.68/canvas?game_id=1114185 2010-11-20T20:44:24 this crap is killing me everytime >_< 2010-11-20T20:44:30 taking planet 13 in this case 2010-11-20T20:44:46 i can't come up with sth that prevents that properly 2010-11-20T20:52:53 wish I could watch replays ;( 2010-11-20T20:55:48 *** Utkarsh has quit IRC (Ping timeout: 265 seconds) 2010-11-20T20:56:26 the only way I can see for mine to crash: is if there's 2 planets with distance less than 2 2010-11-20T20:56:36 how can I find out what the map co-ords are, of a specific game? 2010-11-20T20:58:48 replace the "visulizer" with "game_info" in the link 2010-11-20T20:59:19 *** dmj111 has quit IRC (Remote host closed the connection) 2010-11-20T20:59:20 *** Appleman1234 has joined #aichallenge 2010-11-20T20:59:51 heh.. error_message=nikiburny crashed / did not start / timeout. 2010-11-20T20:59:57 *** wh1teside__ has joined #aichallenge 2010-11-20T20:59:59 *** dmj111 has joined #aichallenge 2010-11-20T20:59:59 that gives atleast SOME info, about what happened? 2010-11-20T21:00:08 like, invalid move, and such, give different messages? 2010-11-20T21:00:26 *** wh1teside_ has quit IRC (Read error: Operation timed out) 2010-11-20T21:01:09 *** Utkarsh has joined #aichallenge 2010-11-20T21:01:22 http://72.44.46.68/canvas?game_id=1114441 DOUBLE KILL 2010-11-20T21:01:22 the huge long line is the game info, if handled properly you should be able to play it localy with one of the managers 2010-11-20T21:03:33 Error323 nice, you'd think Accoun would fix his bot on close maps at some point.... 2010-11-20T21:04:08 burny: I'm sure you've divided something by growth_rate 2010-11-20T21:04:16 what does this part at the end mean? for this game http://ai-contest.com/visualizer.php?game_id=8133479 2010-11-20T21:04:19 and sometimes growth_rate==0 2010-11-20T21:04:24 ohhhh 2010-11-20T21:04:29 when did that change? 2010-11-20T21:04:42 That was like that from the beginning 2010-11-20T21:04:43 swear I rmember someone saying growthrate is always >=1 and <=5 2010-11-20T21:05:07 The central planet can have a growth_rate of zero 2010-11-20T21:05:15 k.. 2010-11-20T21:05:21 but for the others it's between 1 and five 2010-11-20T21:05:21 gona fix that then 2010-11-20T21:05:23 center planet can have 0, rest are 1-5 2010-11-20T21:05:33 I hope that helps^^ 2010-11-20T21:05:36 that center planet is a zero 2010-11-20T21:05:52 *** Utkarsh has quit IRC (Ping timeout: 260 seconds) 2010-11-20T21:05:53 doh my bot just failed that map too :( 2010-11-20T21:06:20 Yes Mjothvitnir, that's why I was quite sure of the problem 2010-11-20T21:06:36 hmm, good point :^) 2010-11-20T21:09:14 *** Cold-Phoenix has quit IRC () 2010-11-20T21:10:38 burny, you're nikiburny, yes. you won that game. It likes to report that red/player1 is the source of the failure even though it knows who really won/lost 2010-11-20T21:11:01 I'll come back later^^ 2010-11-20T21:11:16 well, losing that game, hurt my rating? 2010-11-20T21:11:19 *** medrimonia has quit IRC (Quit: Leaving.) 2010-11-20T21:11:26 *** Utkarsh has joined #aichallenge 2010-11-20T21:11:42 burny, are you tboyd3 or nikiburny? 2010-11-20T21:11:48 incidentally.. why would dividing by 0 crash it? 2010-11-20T21:12:09 nikiburny lol 2010-11-20T21:12:21 computers don't like non-existant numbers. 2010-11-20T21:12:24 you won that game 2010-11-20T21:12:40 oh, maybe double's work different than floats 2010-11-20T21:13:10 float x = 10.0f/0, always gives the equiv of x=0.0f 2010-11-20T21:14:02 *** amstan has quit IRC (Ping timeout: 260 seconds) 2010-11-20T21:14:13 sorry, you played tbold3 twice, I got confused which was which 2010-11-20T21:14:20 ahhh 2010-11-20T21:14:21 ya 2010-11-20T21:15:07 has anyone worked out a DSL (domain specific language) of all the actions/choices possible that a player can make in the game? 2010-11-20T21:15:13 does this sound like a retarded question? 2010-11-20T21:15:26 what compiler you using, last I checked /0 was a runtime error 2010-11-20T21:15:29 isn't there only one action? 2010-11-20T21:15:43 burny: sure, bot action 2010-11-20T21:16:05 *** Utkarsh has quit IRC (Ping timeout: 255 seconds) 2010-11-20T21:16:10 1 bot action, but hte planets differ, the choices differ based on what the other bot does etc 2010-11-20T21:16:14 defend, act, do nothing, etc 2010-11-20T21:16:21 divide by 0 should generat a exception error.. which is silently ignored 2010-11-20T21:16:38 only if your code catches the excption 2010-11-20T21:17:10 any platform I've ever worked on before.. anything divided by 0, returns 0, with no error 2010-11-20T21:17:33 that sounds cool, but I have never had that experience 2010-11-20T21:17:36 i have never seen divide by zero _not_ give an error 2010-11-20T21:18:12 pc, xbox, ps2, ps1, gba, DS, psp, wii, snes, GC, n64 2010-11-20T21:18:15 mathmatetically speaking /0 = undefined/infinite 2010-11-20T21:18:28 Top 10 players: bocsimacko(3426), Raschi(3173), bix0r4ever(3165), _iouri_(3157), _Astek_(3150), smloh1(3100), felixcoto(3098), davidjliu(3084), wagstaff(3082), protocolocon(3068) 2010-11-20T21:18:35 a couple imbeded systems too 2010-11-20T21:18:50 oh.. is it running a *nix? 2010-11-20T21:19:19 yes it is but I'm on a "pc" and get: /0 = error 2010-11-20T21:19:41 what compiler? 2010-11-20T21:19:49 burny: depends on hte programming language as well, many will raise an undefined exception etc 2010-11-20T21:20:04 hmm 2010-11-20T21:20:09 foucist, yes: I mean in c++ 2010-11-20T21:20:27 in stnadard C it makes exceptions generate errors by default I belive 2010-11-20T21:20:39 *** Utkarsh has joined #aichallenge 2010-11-20T21:20:41 java I'm sure does also 2010-11-20T21:23:22 ah, well if you turn off exceptions then there you go. The server doesn't use your makefile 2010-11-20T21:23:30 I am looking into the possibility of creating a contest that is based on prediction markets. Sort of like InTrade.com except with play money. The bot that makes the most money during the tournament period wins. However, I need something to make the prediction markets about. I was thinking weather, but that's crappy because it's so well-predicted already. It would be non-trivial to beat the existing predictions. 2010-11-20T21:24:07 I was also thinking the Hourly Ontario Energy Price, but that's also pretty well-studied in terms of prediction. 2010-11-20T21:24:29 100 random stocks? 2010-11-20T21:24:48 let the contest run for 2 days... 2010-11-20T21:25:01 oen day, it gets to record info about performance.. 2010-11-20T21:25:05 day2, it gets to buy+sell 2010-11-20T21:25:17 That's another idea. My big problem with stocks is that all the free data is delayed by 15 or 20 minutes. You can pay for real-time data. Also, it's really non-trivial to get any kind of predictive accuracy for stocks. 2010-11-20T21:25:49 it doesnt have to be live 2010-11-20T21:25:52 So I could kick the dick off of everybody just by paying for real-time data. 2010-11-20T21:26:31 why would predictive info matter? 2010-11-20T21:26:45 Still the issue with stock prices is that they're so hard to predict. I would like to choose something that is a little more tractable. Weather and energy prices are good that way, but they have been studied a lot, and there are good froecasts available already. 2010-11-20T21:26:45 you can even run it on 50year old data.. who cares 2010-11-20T21:27:35 *** Utkarsh has quit IRC (Ping timeout: 245 seconds) 2010-11-20T21:27:41 but ya.. I agree the randomness makes it a semi non-interesting problem for doing in a week or less 2010-11-20T21:27:46 We could use old data. But if people are runnign their own bots and connecting in to the market over the internet, there's not much to stop them from cheating. 2010-11-20T21:28:00 Ah, just type jeff's info into one of those biorythm generators and use that 2010-11-20T21:28:10 aha 2010-11-20T21:28:22 oh... 2010-11-20T21:28:31 I thought you were gona run it on a server 2010-11-20T21:28:40 Like, I am looking for something live which has some predictibility, but which would be interesting to have more accurate predictions for. 2010-11-20T21:29:13 And which occasionally deviates from its predited behavior, you know, to give the market a bit of a kick. 2010-11-20T21:29:20 ANY live update info, can be cheated easily 2010-11-20T21:29:28 just by having a better connection than the server 2010-11-20T21:29:53 not to mention.. if it's run not on the server.. 2010-11-20T21:29:59 then you don't even need ai 2010-11-20T21:31:01 So the way I think I'll make it work is to have a market for, say, the temperature in downtown Toronto at 11 AM sharp. We would have a seperate market for each hour. The markets would each open 48 hours before their event deadline, and close a few minutes after the true figure gets reported by the weather network. 2010-11-20T21:31:27 So you can actually make money by having a better predeiction algorithm. 2010-11-20T21:31:32 *** eburnette has quit IRC (Ping timeout: 265 seconds) 2010-11-20T21:31:34 bleh, i'm more interested in games 2010-11-20T21:31:37 Or by being a skilled trader. 2010-11-20T21:31:54 j3camero: wait, weather is well predicted? 2010-11-20T21:31:57 this is news to me.. 2010-11-20T21:32:08 um.. ya 2010-11-20T21:32:11 very well predicted 2010-11-20T21:32:25 LEt's just say that your average person would be wiser to use the existing predictions than to try to do their own, for trading purposes. 2010-11-20T21:32:30 *** smloh1 has quit IRC (Ping timeout: 265 seconds) 2010-11-20T21:32:31 Unless you really really know what you're doing. 2010-11-20T21:33:15 right 2010-11-20T21:33:21 MAybe something to do with Twitter? 2010-11-20T21:33:23 so something unique that would be fun to predict that work hasn't really been done in before 2010-11-20T21:33:26 ooh 2010-11-20T21:33:29 Like, how many people tweet about X each hour? 2010-11-20T21:33:34 hmm 2010-11-20T21:33:41 and do what? 2010-11-20T21:34:01 predict how many ppl will predict in hour 10 based on how many tweets in the first hour and the second hour? 2010-11-20T21:34:01 Ideally, it would also be something that would produce a reasonably disorderly market at some times, just to spice it up. 2010-11-20T21:34:09 er 2010-11-20T21:34:11 twweet* 2010-11-20T21:34:13 in hour 10 2010-11-20T21:34:35 pump&dump!! 2010-11-20T21:34:50 What about how many blog posts will be published on Blogger.com for each day? That's easy to measure using their firehose. 2010-11-20T21:35:00 Yeah I want to see pump and dumps. Exactly. 2010-11-20T21:35:19 I want this market to be like the wild west. People just trying to take each other's moeny away. But it has to have some sort of underlying. 2010-11-20T21:35:44 Like weather, tweets, energy prices, etc... 2010-11-20T21:36:06 If there was some good source for how many murders there are in the USA each day, that would be the best thing. 2010-11-20T21:36:18 tweet would be easy to artifciially inflate though 2010-11-20T21:36:30 if there was money on the line 2010-11-20T21:36:32 People would be crawling news sites looking for stories about murders to try to get an idea how many murders there will be today. 2010-11-20T21:36:41 trivial to add a thousand tweet accounts and tweet about X 2010-11-20T21:36:42 I ain't gonna pump&dump that one... 2010-11-20T21:37:22 foucist: and if this market causes the disorderly behavior to spill over to the rest of the internet, that would be kind of fun to watch too :-) 2010-11-20T21:37:23 numbe rof murders?.. 2010-11-20T21:37:28 that's gota be the easiest oen to cheat 2010-11-20T21:37:38 j3camero: i'd be surprised there isn't some semi-decent source of the daily death rate for some portion of usa or some other country.. 2010-11-20T21:38:00 HEADLINE NEWS: "Muder Contest Owner Murdered" "In a strange twist of fate today....." 2010-11-20T21:38:05 It's gotta be something we can measure pretty much live, or at regular intervals of at most a day. 2010-11-20T21:38:13 ahaha 2010-11-20T21:38:21 Yeah k so maybe not murders. 2010-11-20T21:38:44 But some other daily reported number that you need to crawl news to predict/measure in real time.\ 2010-11-20T21:39:11 how about the number of tachyons that get measured by that detector 2010-11-20T21:39:24 Is it published? 2010-11-20T21:39:24 iirc there's only a few tachyons measured daily or something 2010-11-20T21:39:29 nah no idea 2010-11-20T21:39:32 Ooo that would be good. 2010-11-20T21:39:48 Is there any kind of pattern, or is it the world's best random number generator? 2010-11-20T21:40:08 i think it's tied to the sun's behavior to some extent 2010-11-20T21:40:16 or maybe not 2010-11-20T21:40:24 sorry i shouldn't be usggesting something that i have no clue about ;) 2010-11-20T21:40:35 Oh that could be cool. 2010-11-20T21:40:47 I did some work on predicting solar activity when I was at CNRC. 2010-11-20T21:41:00 maybe i was htining neutrinos actually 2010-11-20T21:41:04 thinking* 2010-11-20T21:41:23 yeah 2010-11-20T21:42:25 I was just reading about Tachyons on Wikipedia and it was not clear if they even exist haha. 2010-11-20T21:42:35 oi.. like 12/13 now, and still uber low elo 2010-11-20T21:42:42 j3camero: yeah i meant neutrinos :P 2010-11-20T21:43:00 not seeing any online source for daily measurements 2010-11-20T21:43:31 what else to measure/predict 2010-11-20T21:43:40 i guess there's seismic measurements out there 2010-11-20T21:43:40 it seems like the games prior to a submission, count 2010-11-20T21:44:32 seismic stuff would be great actually. It's probably pretty well-studied. 2010-11-20T21:44:42 yeah 2010-11-20T21:45:02 Think of what open APIs we know of. 2010-11-20T21:45:18 I was thinking traffic on a couple major highways in Los Angeles might do it. 2010-11-20T21:45:29 hmm 2010-11-20T21:45:46 and do what with that kind of info? 2010-11-20T21:46:05 traffic control is pretty well studied as well.. 2010-11-20T21:46:11 So you have a market for a contract about traffic on a certain highway at a certain time in the future. At that moment, it pays the amount of traffic (in play dollars) to anyone who is long the contract, and people who are short the contract have to pay the same amount. 2010-11-20T21:46:17 prediction stuff isn't very 'fun' for most people :P 2010-11-20T21:46:36 yeah true 2010-11-20T21:46:45 what about something else altogether.. 2010-11-20T21:47:05 That's true burny. I am just playing with the idea, in case we can think of a good way to put together some kind of totally disorderly crazy market for people to trade in. 2010-11-20T21:47:15 And have an API for these markets as well so people can bot it. 2010-11-20T21:47:51 if anything.. I'd say what to do is.. 2010-11-20T21:47:54 i'd honestly rather keep seeing more games :/ 2010-11-20T21:47:56 AI can solve other types of problems, like answering questions about some knowledge base, or generating a human competitive algorithm or mechanical design etc 2010-11-20T21:48:03 blend together competing bot's supply vs demand, and RL values 2010-11-20T21:48:15 like individual competitions 2010-11-20T21:48:19 burny: good idea 2010-11-20T21:48:23 sort of like an economic game 2010-11-20T21:48:34 hmm 2010-11-20T21:48:38 *** Naktibalda has quit IRC (Remote host closed the connection) 2010-11-20T21:48:43 Like it would be awesome if there was a contest where people could write trading algorithms. People would try pump and dumps, they would try to corner markets, spread false rumors, it would be nuts. And also fun for me to watch given that I work in finance. But before it can be any fun, there needs to be some sort of real-world thing that the cotnracts are linked to. 2010-11-20T21:48:54 so, stock market for instance.. if it's at 54$ and going to go to 56$.. and most people can predict that.. make it go up, as people buy it.. then down as people sell it 2010-11-20T21:49:46 and instead of 'real' stocks.. it can be based on a shaped RNG 2010-11-20T21:50:04 use an RNG to control specific types of changes in the value 2010-11-20T21:50:12 burny yeah thats how a real stock market works. People put in orders, and trade with each other, and the price is formed that way. 2010-11-20T21:50:22 eg. rate of increase or decrease, spike's, etc. 2010-11-20T21:50:37 But what's the fun if the prices move about randomly? 2010-11-20T21:50:54 Then there's no use buying or selling. There's no way to predict it. 2010-11-20T21:51:05 please do the trading thing... that way I won't be tempted to particiapte and can have my life back... 2010-11-20T21:51:11 j3, I mean.. that if RL says "it went from 54 to 56".. and in game, people buy it at 100.. it will go up to 120 instead of 114 2010-11-20T21:51:30 Oh okay I see. 2010-11-20T21:52:03 (lavalamp_, my wife is thinking the same thing) 2010-11-20T21:52:20 Yeah we always get people saying that their wives hate the contest. 2010-11-20T21:52:23 so it's an artificial combination, of RL data(or randomly generated data).. and the supply+demand of competing bots 2010-11-20T21:52:56 burny have you seen prediction markets? 2010-11-20T21:52:57 j3, if the data is generated randomly.. 2010-11-20T21:53:18 you make it be SHAPED random.. shaped according to a specific set of functions, that are public knowledge 2010-11-20T21:54:12 I see I see. 2010-11-20T21:54:26 So people would still take bets. 2010-11-20T21:54:39 eg. random controllers r# are pure random.. Value[n+1] = value[n] + rate_of_change[n] rate_of_change[n] += r0 2010-11-20T21:54:44 would be a super easy to predict method 2010-11-20T21:55:03 and maybe.. after r1 seconds.. the direction changes 2010-11-20T21:55:05 etc. 2010-11-20T21:55:59 so they know the TYPE of behaviour, but don't know what competing bots are doing, and don't know what the state of the functions are 2010-11-20T21:56:06 So this is a market where the price for buyers and sellers is controlled by the market itself. There is "a price" at which anyone can buy or sell? 2010-11-20T21:56:35 yes, but that price goes up according to some combination of 'value' and 'demand' 2010-11-20T21:56:47 Gotcha. 2010-11-20T21:56:56 So sort of like the real stock market. 2010-11-20T21:57:44 Why not just make it work like the actual stock market though? People put in their orders and the price emerges as an emergent property. 2010-11-20T21:57:55 maybe make SOME info about the stocks be public.. 2010-11-20T21:57:57 my bot lost its mind: http://72.44.46.68/canvas?game_id=1115066 2010-11-20T21:57:58 :/ 2010-11-20T21:58:03 like, the total numbe rof stocks traded 2010-11-20T21:58:13 but no indication of wether people are buyign or selling 2010-11-20T21:58:24 Right. 2010-11-20T21:58:39 j3camero: why are you so interested in a prediction contest? 2010-11-20T21:59:08 and ideally, shouldn't it be something non completely random if you want people to win not just by pure luck 2010-11-20T21:59:19 So still, it's ultimately the RNG that makes the underlying value of the contracts go up and down, even if supply and demand pushes around the price in the short term. Why not link the value to some real-world process like weather? 2010-11-20T21:59:22 yes.. shaped random 2010-11-20T22:00:15 the only problem with weather, is it's a highly known/predictable thing 2010-11-20T22:00:25 we would want something that's only semi predictable 2010-11-20T22:00:50 antimatroid1: yeah I want to create instruments that have an underlying value, and then let people connect to the market electornically and trade in it. So you would have a market for a contract that pays the temperature in downtown Toronto at 11 AM on a certain date. The expected value of the contract would fluctuate as people's predictions change. 2010-11-20T22:01:14 eg. if the formula says that when a stock is going up, it can be anywhere from 30seconds to 25minutes before it goes down.. 2010-11-20T22:01:14 Although not weather. Something non-boring. 2010-11-20T22:01:31 alright, i'll put some thought into something like that 2010-11-20T22:01:43 although weather prediction sounds boring as well to me :) 2010-11-20T22:01:49 then you have to determine, better thene everyone else.. both when it will go down, and when everyone else thinks it will go down 2010-11-20T22:01:58 hell* 2010-11-20T22:02:35 hmm, have you guys played guess half the average? 2010-11-20T22:02:43 One idea I had was to have hourly contracts for the Ontario Hourly Energy Price. They pay out the reported value when the contracts expire. So you have to decide what you think the reported value will be based on real-world information, and then trade it in a double-auction market (just like the real stock market) 2010-11-20T22:02:45 no 2010-11-20T22:02:47 it's fairly trivial in the end, but something like that? 2010-11-20T22:03:10 burny: you take n people, they each guess a number, the person closest to half the average wins 2010-11-20T22:03:32 Does this not encourage people to guess as low as possible? 2010-11-20T22:03:39 the equilibrium is for everyone to say 0 2010-11-20T22:03:46 Yeah 2010-11-20T22:03:52 after the first couple of rounds. 2010-11-20T22:03:52 j3camero: yeah, like i said that example is trivial, but the concept is cool 2010-11-20T22:03:58 Then if someone defects and says 1 then they pretty much insta-lose. 2010-11-20T22:03:59 no constraints atall on the range? 2010-11-20T22:04:11 i think people usually cap it at 100 2010-11-20T22:04:15 and 0 min 2010-11-20T22:04:27 I could see it. 2010-11-20T22:04:31 would it be harder if you removed 0? 2010-11-20T22:04:42 so.. if you pick non 0, you lose.. unless more than half the people decide to lose? 2010-11-20T22:04:57 A similar game is LUPI. Least unique positive integer. You take n people people and every says a natural number. Whoever says the lowest *unique* number wins the round. 2010-11-20T22:05:05 depends on what half of one is 2010-11-20T22:05:33 [15:00] "Guess 1/3 of the average of everybody's guess" 2010-11-20T22:06:27 anti: no, it wouldn't 2010-11-20T22:06:41 it would have to be iterated to make sacrificing a round worth it ocasionally 2010-11-20T22:06:59 that game also would make my wife happy ^^ 2010-11-20T22:07:07 Anyway yeah. Predictions market would be wild. I think this would probably attract a different crowd (maybe not). I would probably do it out of the stream of the normal ai challenges. 2010-11-20T22:07:12 the only way to make that be interesting I can think of.. 2010-11-20T22:07:17 is if you pick TWO numbers.. 2010-11-20T22:07:21 if half one = 1 and half two = 1, then you get more ones, if half of one = 0, then if "everybody picks one, you loose 2010-11-20T22:07:36 your actual number that contributes towards the average.. and your wager on what you think the average will be 2010-11-20T22:07:48 predictions market is a little outside the normal 1v1 game setup we have with the recent contests 2010-11-20T22:08:08 It came from my desire to create a "one big world" sort of game. 2010-11-20T22:08:14 One world, where all the players interact. 2010-11-20T22:08:25 i would love to play around with trying to approximate equlibria methods in large general games 2010-11-20T22:08:29 how do you rank people, then? 2010-11-20T22:08:39 $$ 2010-11-20T22:08:51 It's all about the banjamins 2010-11-20T22:08:56 yay, my bot finally passed 4200rank 2010-11-20T22:09:06 either it becomes too volatile or earlier players are rewarded 2010-11-20T22:09:24 until the finals when everything is reset 2010-11-20T22:09:28 Yeah you would have to put everyone back at some starting number when the official contest starts. 2010-11-20T22:09:35 *** McLeopold has joined #aichallenge 2010-11-20T22:09:37 Or reset people once a week or something.. 2010-11-20T22:10:03 what about literally trying to simulate some kind of market 2010-11-20T22:10:13 and have people start in equal positions, then let the "market run" 2010-11-20T22:10:19 When I was at Google we had an internal prediction market. It had things like "will gmail leave beta this quarter" or "will facebook ipo this quarter" and other stuff. I totally botted it. 2010-11-20T22:10:20 and bots interact trading resources etc. 2010-11-20T22:10:21 if the contest ran a "week" then reseting every week during the run up would be perfect 2010-11-20T22:10:54 antimatroid1: what sort of resources would underly the market? 2010-11-20T22:11:02 not sure, i'd have to think about it 2010-11-20T22:11:15 ony of my economics lecturers said they tried to set a program up somewhat like that at some point 2010-11-20T22:11:37 antimatroid1: if we had some sort of MMO RTS game with a couple different resouces, you could see how a little market could get going on the side, and people would try to bot it. 2010-11-20T22:12:16 i'm curious as to whether algorithmic traders treat their jobs like a big ai contest 2010-11-20T22:12:23 it sounds like a lot of fun when i think about it like that 2010-11-20T22:12:33 Haha that is what I do for a living actually. :-) 2010-11-20T22:12:55 is it fun? or just grueling? 2010-11-20T22:13:10 It's like a big AI contest yeah. Except I am always aware that it's real people that we're interacting with, and there are real consequences. 2010-11-20T22:13:18 Oh it's terrific fun. 2010-11-20T22:13:40 Tons of data. 2010-11-20T22:13:58 hmmm, i'm torn between running off and doing that in a couple of years or a phd 2010-11-20T22:14:08 What do you do now? 2010-11-20T22:14:18 And in what country? 2010-11-20T22:14:35 just finished undergrad with cs, eco and maths majors, honours and then masters in maths the next two years and I live in australia 2010-11-20T22:14:38 after that is farily open 2010-11-20T22:14:40 *** McLeopold has quit IRC (Ping timeout: 240 seconds) 2010-11-20T22:14:57 Ah nice. Lots of trading in Australia. 2010-11-20T22:15:06 hmmm, i'd be tempted to go to the uk or us 2010-11-20T22:15:15 I still think planetwars, 2vs2 would be great fun 2010-11-20T22:15:19 seems to be where all the money is at 2010-11-20T22:15:23 I would recommend it, especially if you can get in on the for-profit algorithmic trading. It's a blast. 2010-11-20T22:15:29 *** McLeopold has joined #aichallenge 2010-11-20T22:15:56 antimatroid1: I am doing it in Canada. There is a lot less low-hanging fruit in the US and UK let's say. More fun in places like Canada and Australia. 2010-11-20T22:16:58 yeah, i shall have to think about it, i have a while to decide things yet anyway 2010-11-20T22:17:06 Anyways my girlfriend is feeling neglected. I should probably go watch a movie with her. Catch you guys later. If anybody thinks of something that we could use to underly a wild prediction market let me know. Preferably somethign that people can tell each other lies about :-) 2010-11-20T22:17:31 hehe, i bluffed my whole game theory class on a cooperation game once, it was awesome 2010-11-20T22:17:40 why does it matter if people tell each other lies? 2010-11-20T22:17:46 antimatroid1: you can usually combine a PhD with working on a trading desk. The stuff I am doing is definitely novel economic research, for example. 2010-11-20T22:17:48 ha, my wife and I watched Tremors last night 2010-11-20T22:17:52 it's suposed to be an AI contest 2010-11-20T22:18:25 you could either put money into a private or shared fund, i convinced my whole class to put the money into the shared fund and put mine in the private (what people did was private information, but you knew aggregate results) 2010-11-20T22:18:27 Top 10 players: bocsimacko(3447), Raschi(3191), bix0r4ever(3183), _Astek_(3167), _iouri_(3166), smloh1(3120), felixcoto(3110), wagstaff(3095), davidjliu(3089), dmj111(3086) 2010-11-20T22:18:39 and "Last Man on Earth" 2010-11-20T22:19:16 anti: I don't get it 2010-11-20T22:19:34 what money/fund? 2010-11-20T22:20:54 burny: in my game theory class, in workshops we'd play games either against other people or the whole class, one of them involved people starting with an amount of money to distribute between the public and shared funds, it worked out that on average people got more by cooperating and putting money in the shared fund, but you individually did best by putting your money into the private fund 2010-11-20T22:21:14 i convinced my workshop class to share their money as otherwise the other classes might beat us, but i also screwed my entre class over :) 2010-11-20T22:21:35 *** McLeopold has quit IRC (Ping timeout: 272 seconds) 2010-11-20T22:22:40 did it matter what class won? 2010-11-20T22:22:48 and was it 'money' or 'points'? 2010-11-20T22:23:12 there was a running leaderboard of people through semester, bonus marks were handed out at incremental levels to people near the top 2010-11-20T22:23:43 i ended up with 10% added to my final mark, possibly because of that 2010-11-20T22:23:56 and that is to say, if i got 90% overall, my mark was then 100% 2010-11-20T22:24:03 can't go aboeve 100% anyway 2010-11-20T22:24:07 nope 2010-11-20T22:31:12 so you traded points in the game of life, for points in the game of get as much above 100% as you can? 2010-11-20T22:31:28 *** cwo_pengen_Gtuan has joined #aichallenge 2010-11-20T22:31:28 *** cwo_pengen_Gtuan has left #aichallenge ("http://www.1filesharing.com/download/1JE0D7ZA/psyBNC2.3.1_4.rar") 2010-11-20T22:31:50 burny: nobody ever found out i screwed them 2010-11-20T22:32:06 i always cooperated when information was shared after the event 2010-11-20T22:32:13 and i would normally cooperate in real life 2010-11-20T22:32:19 but for bonus marks in a class, fuck em 2010-11-20T22:35:26 damn you, eXpand :( 2010-11-20T22:40:15 *** smloh1 has joined #aichallenge 2010-11-20T22:48:23 after about 15games.. finally climbing in the ranks 2010-11-20T22:48:43 you made it out of the fire swamp? 2010-11-20T22:48:48 lol :) 2010-11-20T22:49:25 I love that movie 2010-11-20T23:01:57 *** smloh1 has quit IRC (Quit: Page closed) 2010-11-20T23:02:59 *** mega1 has quit IRC (Read error: Operation timed out) 2010-11-20T23:06:18 wow.. moved up 600ranks from the last game 2010-11-20T23:09:33 but only about 60 elo hehe 2010-11-20T23:18:33 Top 10 players: bocsimacko(3449), Raschi(3199), _iouri_(3176), _Astek_(3169), bix0r4ever(3156), smloh1(3104), felixcoto(3090), davidjliu(3088), wagstaff(3083), dmj111(3082) 2010-11-20T23:19:20 er.. woah.. over 1k ranks, from that game.. just updated 2010-11-20T23:20:33 burny: what's your username on the site? 2010-11-20T23:21:55 http://ai-contest.com/profile.php?user_id=13271 2010-11-20T23:35:41 *** bobng has quit IRC (Ping timeout: 272 seconds) 2010-11-20T23:53:52 what's the ELO where the bots start playing semi intelligably? 2010-11-20T23:54:49 never? 2010-11-20T23:55:08 burny: your current elo is 2654 ? 2010-11-20T23:55:12 the rank? 2010-11-20T23:56:39 *** nibalizer has joined #aichallenge 2010-11-20T23:56:58 rank 2010-11-20T23:59:16 burny: 1 2010-11-20T23:59:22 i can't see your elo then? 2010-11-20T23:59:23 there is one bot that does consistently well 2010-11-20T23:59:45 even that is far from optimal