2011-09-21T00:00:40 *** Keth_ has joined #aichallenge 2011-09-21T00:05:01 *** heinrich5991 has quit IRC (*.net *.split) 2011-09-21T00:05:02 *** Keth has quit IRC (*.net *.split) 2011-09-21T00:05:02 *** Maligo has quit IRC (*.net *.split) 2011-09-21T00:05:04 *** Chris_0076 has quit IRC (*.net *.split) 2011-09-21T00:05:04 *** delt0r_ has quit IRC (*.net *.split) 2011-09-21T00:05:04 *** dvladim has quit IRC (*.net *.split) 2011-09-21T00:05:04 *** Cyndre has quit IRC (*.net *.split) 2011-09-21T00:05:05 *** antimatroid has quit IRC (*.net *.split) 2011-09-21T00:05:06 *** janzert has quit IRC (*.net *.split) 2011-09-21T00:05:06 *** moondust has quit IRC (*.net *.split) 2011-09-21T00:05:07 *** acieroid has quit IRC (*.net *.split) 2011-09-21T00:05:07 *** Zannick has quit IRC (*.net *.split) 2011-09-21T00:05:07 *** kire has quit IRC (*.net *.split) 2011-09-21T00:05:08 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (*.net *.split) 2011-09-21T00:05:11 *** AlliedEnvy has quit IRC (*.net *.split) 2011-09-21T00:05:11 *** dr- has quit IRC (*.net *.split) 2011-09-21T00:05:11 *** mleise1 has quit IRC (*.net *.split) 2011-09-21T00:05:12 *** Accoun has quit IRC (*.net *.split) 2011-09-21T00:05:12 *** Eruonen has quit IRC (*.net *.split) 2011-09-21T00:05:13 *** nplus has quit IRC (*.net *.split) 2011-09-21T00:05:13 *** Lithosphere has quit IRC (*.net *.split) 2011-09-21T00:05:14 *** jmcarthur has quit IRC (*.net *.split) 2011-09-21T00:51:52 *** McLeopold has joined #aichallenge 2011-09-21T00:52:08 *** McLeopold has quit IRC (Client Quit) 2011-09-21T01:17:19 *** JokerJo has joined #aichallenge 2011-09-21T01:27:57 *** eashoka has joined #aichallenge 2011-09-21T01:40:35 *** bhasker has joined #aichallenge 2011-09-21T01:48:08 rather empty irc 2011-09-21T01:48:09 hmm 2011-09-21T02:17:42 *** onensora has joined #aichallenge 2011-09-21T02:31:30 *** amstan has quit IRC (Ping timeout: 260 seconds) 2011-09-21T02:35:24 *** JokerJo has quit IRC (Quit: Page closed) 2011-09-21T02:45:29 *** nux67 has joined #aichallenge 2011-09-21T02:55:22 *** ibdknox has quit IRC (Ping timeout: 252 seconds) 2011-09-21T03:01:12 *** Keth_ has quit IRC (Read error: Connection reset by peer) 2011-09-21T03:26:25 *** ltriant has quit IRC (Quit: Computer has gone to sleep) 2011-09-21T03:49:36 *** Chris_0076 has joined #aichallenge 2011-09-21T03:49:36 *** kire has joined #aichallenge 2011-09-21T03:51:32 *** Keth has joined #aichallenge 2011-09-21T03:51:32 *** AlliedEnvy has joined #aichallenge 2011-09-21T03:51:32 *** dr- has joined #aichallenge 2011-09-21T03:51:32 *** mleise1 has joined #aichallenge 2011-09-21T03:51:32 *** Accoun has joined #aichallenge 2011-09-21T03:51:32 *** nplus has joined #aichallenge 2011-09-21T03:51:32 *** Lithosphere has joined #aichallenge 2011-09-21T03:51:32 *** jmcarthur has joined #aichallenge 2011-09-21T03:51:43 *** Maligo has joined #aichallenge 2011-09-21T03:52:29 *** delt0r_ has joined #aichallenge 2011-09-21T03:52:29 *** Cyndre has joined #aichallenge 2011-09-21T03:52:29 *** antimatroid has joined #aichallenge 2011-09-21T03:52:29 *** janzert has joined #aichallenge 2011-09-21T03:52:29 *** moondust has joined #aichallenge 2011-09-21T03:52:29 *** Zannick has joined #aichallenge 2011-09-21T03:52:29 *** acieroid has joined #aichallenge 2011-09-21T03:52:58 *** heinrich5991 has joined #aichallenge 2011-09-21T03:58:00 *** bhasker has quit IRC (Ping timeout: 252 seconds) 2011-09-21T04:11:42 *** aerique has joined #aichallenge 2011-09-21T04:48:17 crap, my bots are not getting terminated properly 2011-09-21T04:48:38 *** Keth has quit IRC (Ping timeout: 260 seconds) 2011-09-21T04:52:33 *** Keth has joined #aichallenge 2011-09-21T04:54:54 deltor : terminated means ? method do_turns() doesn't exit ? or an issue with the output text ? 2011-09-21T04:55:53 *** Palmik has joined #aichallenge 2011-09-21T04:56:29 the bots stay live even when the engine is killed 2011-09-21T05:02:25 *** berak has joined #aichallenge 2011-09-21T05:03:36 where can I submit an update to a starter package so that it properly compiles/works with the current system? 2011-09-21T05:06:39 Either way it can be found at http://jnwhiteh.net/files/go_starter_package.zip and works with the version of Go installed on the servers 2011-09-21T05:06:50 I don't see them in the repo, or I'd submit a pull request =) 2011-09-21T05:21:13 *** mleise1 is now known as mleise 2011-09-21T05:29:53 *** sigh has joined #aichallenge 2011-09-21T05:36:43 delt0r : not due to a problem in the bot code I guess. its engine's responsibility to kill them off isn't it ? 2011-09-21T05:37:01 yea, should say.... in my engine 2011-09-21T05:37:12 seem to have fixed it 2011-09-21T05:38:30 deltor : BTW, I again see instances of bots loosing skill levels even after winning games. I logged a dfect for this long ago. looks like it has got re-opened. :-( 2011-09-21T05:41:42 are these in many player games eashoka 2011-09-21T05:42:20 you know i am not really "close to the metal", just a dev who likes to lurk and waste his time coding this up some times 2011-09-21T05:45:36 *** sir_macelon has joined #aichallenge 2011-09-21T05:48:43 delt0r : I just saw one game like that 2011-09-21T05:48:45 my pc clock drifts about 15min in just 22 days 2011-09-21T05:49:35 eashoka: so what it a 2 player game or 4 or what? 2011-09-21T05:49:39 was it 2011-09-21T05:51:15 delt0r : a 5 player game 2011-09-21T05:51:35 right, this is sort of where True Skill can get a bit funny 2011-09-21T05:52:05 it takes the orginal ranks and really only considers "ranked" draws... 2011-09-21T05:52:16 u could see in this profile. see the top game. winner has lost points! http://aichallengebeta.hypertriangle.com/profile.php?user=188 2011-09-21T05:53:47 there are not skill scores going down on that page 2011-09-21T05:53:51 no skill 2011-09-21T05:54:33 sorry, it has become the second game from top now. 2011-09-21T05:55:06 winner has lost 0.02 skill points 2011-09-21T05:55:24 well it shows a - 2011-09-21T05:56:01 yea it shows arrows only when the gain/loss is greater that 0.1 or something 2011-09-21T05:56:04 meaning the same skill... since there have quite a few resets, my guess 70.03 was not really the start score perhaps 2011-09-21T05:56:15 perhaps there is a reason for that 2011-09-21T05:56:32 esp if sigma is like 1 or something 2011-09-21T05:59:42 delt0r : it is actually a bit larger problem that 0.02, because actually the winner should have gained something. (even though a small fraction of a point). so that actual error has to be a larger one (for example : showing -0.02 in place of a +0.08 is actually a error of 0.10) 2011-09-21T06:01:03 eashoka: that is not correct... if they play a buch of people with a lot less skill than there own, then a win should not increase there skill 2011-09-21T06:01:16 as is expected from the model 2011-09-21T06:01:49 because we have a iterative approximate solver (only way to make it tracktable) you can "bleed" a little skill in this case 2011-09-21T06:01:59 round of error if you like 2011-09-21T06:02:03 off 2011-09-21T06:02:16 I really don't see anything unexpected sorry. 2011-09-21T06:02:49 also i don't think .02 will ever matter in the long run, since the *sigma* will show the spread 2011-09-21T06:03:13 delt0r : ah ok. so u think it's not a defect? alright then, I'm not familiar enough in trueskill to argue u in this :-) anyway, I have seen few other instances like that previously in many other profiles and have reported. 2011-09-21T06:03:34 if its like .1 i would worry. 2011-09-21T06:03:46 delt0r : in ELO ratings however, one never loose rating points after winning with a lower rank player. 2011-09-21T06:03:58 but at .02 when clearly the skill levels are very different... i am not surprised 2011-09-21T06:04:08 and here you would too 2011-09-21T06:04:25 a losse would mean loosing a lot of skill 2011-09-21T06:04:40 hence why you try and match similar skilled folks 2011-09-21T06:04:58 because a really good player has nothing to gain by playing low skill players 2011-09-21T06:10:12 delt0r : yea agreed. (however, ideal should be an unaltered skill level after winning with a weaker player) that brings me to the second qsn. currently in the tournament, it doesn't look like there is an attempt to match similar skilled folks. may be this is just for some testing purpose and not the final matching algorithm ? :-) 2011-09-21T06:14:06 *** onensora has quit IRC (Ping timeout: 252 seconds) 2011-09-21T06:16:08 no--its a beta 2011-09-21T06:16:30 they are debuging --they needed to force a few players to play always... 2011-09-21T06:16:49 so in the last 24 hours or so there are some pretty bad match ups 2011-09-21T06:17:06 Its not what would be expected in the live version 2011-09-21T06:17:18 Its is also still getting worked on 2011-09-21T06:17:55 in the true skills system the match quality is a measure of how likely a draw is btw 2011-09-21T06:18:55 *** olexs has joined #aichallenge 2011-09-21T06:19:20 also not true skills is bayesian and that its is approximate... but i think its a good set of trade offs. 2011-09-21T06:19:27 s/not/note 2011-09-21T06:19:41 what is with me in the last few days... I can't type 2011-09-21T06:21:20 delt0r : ah u r way better than me in terms of typos :-) interesting. so the skill calculation and match pairings goes very much hand in hand in true skill ? 2011-09-21T06:23:38 delt0r : In ELO based systems they mostly use swiss system for pairing but generally the pairing logic is important only to decide on a tournament winner and not the rating level 2011-09-21T06:31:52 true skill and ELO are related. true skill is kind of ELO on steroids. They both need a lot of games played and in our case we have n player games. Draws are kinda stuck on the end of both systems more or less 2011-09-21T06:32:51 ass for optimal matches ... i don't even think that is known for sure, but it has been asserted that similar is better since you get more information about "local rank" 2011-09-21T06:32:59 ass->as 2011-09-21T06:33:32 and don't forget each player gets a sigma --while in elo is the same for everyone... IIRC 2011-09-21T06:34:06 Both use approximations pretty heavily to make the problem tractable 2011-09-21T06:34:07 *** bhasker has joined #aichallenge 2011-09-21T06:34:27 esp trueskill 2011-09-21T06:34:45 for example assuming the posterior is a Gaussian 2011-09-21T06:36:11 afternoon 2011-09-21T06:36:25 the true distribution will depend on your rank, and will be closer to some sort of extreme distribution (Weibul) when you are last/first 2011-09-21T06:37:10 eashoka: sorry... i generally don't ping if i think its just us doing the talking... 2011-09-21T06:37:20 bhasker: hi 2011-09-21T06:37:48 hi, I have created a snippet for an algo proposal as in https://github.com/aichallenge/aichallenge/issues/247#issuecomment-2019057 2011-09-21T06:38:06 https://gist.github.com/1231764 - what do you think of it? 2011-09-21T06:44:49 what the point sir_macelon 2011-09-21T06:51:03 hi delt0r 2011-09-21T07:05:54 delt0r_: to level the rankings for close scores with a parametrizable factor 2011-09-21T07:06:10 but why? 2011-09-21T07:06:42 have you read the issue #247? 2011-09-21T07:10:21 the problem is bots don't get any information about scores during the game, so it is impossible to plan game strategy for that 2011-09-21T07:12:11 and since score determines final ranks, it is rather unfair if a bot turns out to lose bacuse it didn't know it was behind in score 2011-09-21T07:12:27 example: http://aichallengebeta.hypertriangle.com/visualizer.php?game=196413 2011-09-21T07:13:22 @monologue 2011-09-21T07:13:22 sir_macelon: Your current monologue is at least 5 lines long. 2011-09-21T07:18:27 delt0r : sorry, had to run out. thanks for all explanations. :-) 2011-09-21T07:20:05 *** Palmik has quit IRC (Remote host closed the connection) 2011-09-21T07:20:15 sir_macelon : I have proposed a little improvement for the scoring formula. (given as comments on the issue) like to know your views on it as well. I think that also will help address the problem you have mentioned just now. 2011-09-21T07:20:41 sir_macelon: your version of unfair is not the same as mine 2011-09-21T07:21:13 eashoka: I'll have look at it 2011-09-21T07:21:32 delt0r_: what do you mean by that? what is your version? 2011-09-21T07:21:39 the bot does not have global information. Therefore it can't know if its wining... unless perhaps it takes over the whole map. So the bot must take that into account 2011-09-21T07:22:02 my version is that the rules are the same for everyone and everybot, so it is fair 2011-09-21T07:22:14 fair does not mean easy 2011-09-21T07:24:23 sir_macelon, delt0r : I think I'm also talking about the same issue but in a different view. In the game sir_macelon provided, the real reason for the anomoly is that scoring algorithm does not acknowledge who the victor is (in normal battles this is mostly the party which has larger remaining force at the end of the battle). We don;t take remaining ants into the equation. 2011-09-21T07:24:26 *** bhasker has quit IRC (Ping timeout: 252 seconds) 2011-09-21T07:24:48 delt0r_: well maybe indeed fair is not the right word here, but maybe incosistent 2011-09-21T07:24:55 by a simple improvments to the scoring algorithm we can rectify that and still make sure the better bots in battle is rewarded. 2011-09-21T07:27:25 *** delt0r_ has quit IRC (Ping timeout: 276 seconds) 2011-09-21T07:27:26 delt0r_: I also think that the bot should have some kind of information about the current rank during the game to be able to react on that 2011-09-21T07:28:27 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has joined #aichallenge 2011-09-21T07:28:30 issue 247 comments have all the explanations but in short : score = (kills_made - ants_lost) * weight + remaining_ants_at_the_end Note that there is no bonus for spawning! 2011-09-21T07:29:40 eashoka: I like your idea for the scoring 2011-09-21T07:29:51 but you don't weight the remaining ants? 2011-09-21T07:30:27 first weight can control the second. consider its a weight of 1 2011-09-21T07:30:48 shouldn't this weight also be adjustable? 2011-09-21T07:31:27 if we increase the first wight then that means more prominance to battles and less for remaining ants. if we lower the first weight beyone 1, that will increase the remaning ants significance 2011-09-21T07:32:11 so mathematically, just one weight is enough, but yea we could have too, that will make it more readable as well. 2011-09-21T07:32:20 too -> two 2011-09-21T07:32:20 I also have another idea: the bot would recieve a new line of input from the engine with its current rank, then it would know he is behind eventhough it sees it has more ants in game 2011-09-21T07:34:04 sir_macelon : the very idea why a new input is needed is because bots can't calculate for themselves the score isn't it ? that's because score is less natural right now. (no acknowledgement for the overall battle winner with the larger army) 2011-09-21T07:35:09 if we solve that prob, then bots may not need a new input. it can just go about killing, spawning and defending. increasing the relative strength of its army and everything will turn out alright. 2011-09-21T07:35:23 eashoka: exactly, the idea behind is that the bot MUST actually know if it is winning or loosing at any point in the game, to be able to react on that 2011-09-21T07:35:46 that was possible in every previous contest, why now it is not? 2011-09-21T07:36:45 how scoring is done is rather a background topic as it is just an indirect solution 2011-09-21T07:37:32 *** onensora has joined #aichallenge 2011-09-21T07:37:37 like delt0r mentioned, if the bot can see the map he can find out enemy strength and his own. if the scoring is natural then bots can reasonably calculate their position isn't it ? 2011-09-21T07:37:57 what I also find weird from the engine pov that bot doesn't get precise information about movement of enemy ants 2011-09-21T07:38:48 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (Ping timeout: 252 seconds) 2011-09-21T07:39:01 sir_macelon : do you think fog of war is bad ? 2011-09-21T07:39:08 eashoka: the problem is that how would you check if your scoring is natural? 2011-09-21T07:39:09 *** delt0r_ has joined #aichallenge 2011-09-21T07:39:57 eashoka: I don't think so, but as I expressed there should a way for the bot to know if it is winning or not 2011-09-21T07:40:16 I meant natural in the sense that it adhers to normal battle resolutions. (larger army at the end of the battle doesn't lose it :-) ) 2011-09-21T07:41:46 but how this scoring theme would deal with bots avoiding fight and gathering ants, I can imagine there can be some odd situations at game end 2011-09-21T07:42:04 I think we both agree on the exact problem (scenarios like in the game you provided should not happen) we just proposing 2 different solution to it right ? and there could be other solutions as well (such as increasing the turn limit or set it dynamically adjusted)) 2011-09-21T07:42:49 I don't have anything against having both solutions 2011-09-21T07:42:54 sir_mac : in the improvement I suggested. we can increase the battle weight and then (kills - losses) will have more significant say over the final result. 2011-09-21T07:44:12 but then suppose you have a very aggressive bot, who fights everyone has many kills, but eventually is eliminated later, he could have so big score, keeping him as a leader till the game ends 2011-09-21T07:44:21 how the scoring would deal with it? 2011-09-21T07:45:44 both those situation you mentioned will be handled by finding the proper value for the weight. we may need to run the scoring algo against current data to seehow it behaves and experiment with different wegihts 2011-09-21T07:46:08 even now, the system depends on a weight, (2 bonus points for a kill and 1 for a spawn) 2011-09-21T07:46:43 eashoka: I believe we won't find an ideal weight, because it may turn out every weight would have a flawed scenario 2011-09-21T07:46:47 any formula will depend on correct values for the weights 2011-09-21T07:47:33 sir_macelon : true. anamoloies will happen in any case. all we can do is to minimise them 2011-09-21T07:48:03 I think much easier would be just to let to know the bot if it is winning or not, then the scoring system doesn't matter really 2011-09-21T07:48:52 sir_macelon : you have a point. what needs to be passed is the score in that case, not the rank. 2011-09-21T07:49:03 as the score is also quantitative 2011-09-21T07:49:26 not necessarilly, because the bot shouldn't know the score of other bots 2011-09-21T07:49:49 even it shouldn't know what is its current rank 2011-09-21T07:50:00 just to know if it is on the first place or not, that would be enough 2011-09-21T07:50:17 without exposing too much information to it about the game progress 2011-09-21T07:51:31 if ur objective to be achieved, I think scores of all bots should be provided to it in every turn (which is not necessarily a bad thing) one problem I have with that is the point delt0r raised. bots should be made out to find info. 2011-09-21T07:52:19 one can argue that by winning the control of the entire map, one can see all remaining enemy, one knows how many of it's ants got lost and how many killed etc.... 2011-09-21T07:52:58 well indeed you can find out eventually the whole map, if you explore it totally, how many opponents there are and how many ants do they have, but you will never know their score 2011-09-21T07:53:30 I have nothing against your solution it's good. wish there is a way for a bot to fully calculate all by itself........ 2011-09-21T07:53:39 some opponents may get eliminated before you ever see them as well 2011-09-21T07:53:40 and if you got to know all the scores, then you would get to know the number of players, which is supposed not to be known from the start 2011-09-21T07:53:40 etc. 2011-09-21T07:54:20 eashoka: there fog of war should be removed if you would like to know everything 2011-09-21T07:54:28 antimatroid: you have a point 2011-09-21T07:54:45 but I like for of war :-) 2011-09-21T07:54:56 ok, how about this : 2011-09-21T07:55:23 basically the bot shouldn't get a free info about player count on the map as it would enable it to easier calculate the map symmetry 2011-09-21T07:56:27 yes. 2011-09-21T07:57:24 *** bhasker has joined #aichallenge 2011-09-21T07:58:14 sir_macelon : can you explain again the last line please 2011-09-21T07:58:28 *** olexs has quit IRC (Ping timeout: 260 seconds) 2011-09-21T07:58:29 *** olexs1 has joined #aichallenge 2011-09-21T07:59:05 eashoka: you mentioned that the engine could provide about bot scores every turn, yes? 2011-09-21T07:59:23 *provide info about 2011-09-21T08:00:00 the number of those scores would tell the bot how many players there are in the game 2011-09-21T08:00:17 and this kind of info the engine is trying to hide 2011-09-21T08:00:50 I believe you know why, don't you? 2011-09-21T08:00:57 yea, I arrived at that since you initially suggested bots should be provided their ranks 2011-09-21T08:01:19 but I agree now with your later comment.... 2011-09-21T08:01:39 only the 1st or not should be provide. so only a boolean value. true or false. 2011-09-21T08:01:54 that sounds good. 2011-09-21T08:02:34 eashoka: exactly, that was an enahncement for initial idea :) 2011-09-21T08:02:47 now we are on same page. :-) 2011-09-21T08:03:30 so to summarize the changes I would like to see in the engine in terms of input provided: 1) one line with info winning/loosing; 2) additional info about direction of movement of enemy ants 2011-09-21T08:03:44 eashoka: glad to see that :) 2011-09-21T08:04:04 I thought with 'providing the rank' we would eventually end up with something like providing scores of all bots :-) phew! 2011-09-21T08:04:41 ok with 1) 2011-09-21T08:04:59 why 2) ? that goes against the spirit of fog of war isn't it ? 2011-09-21T08:05:00 eashoka: well even with the rank provided you won't get the full info 2011-09-21T08:05:23 eashoka: why is that? 2011-09-21T08:05:37 but rank will tell how many players are there and the score is just more quantifying info on rank. they are basically same. 2011-09-21T08:06:16 eashoka: it only will tell you there are at least that many players as your current rank, but you don't know more 2011-09-21T08:07:00 for example: you are now third, doesn't mean it is a 3 player game 2011-09-21T08:07:32 if I were given as rank 3 then i AT LEAST know it is not a 2 player game, which will drastically change my approach 2011-09-21T08:08:07 eashoka: that's why I eventually proposed the boolean value 2011-09-21T08:08:34 but it was at least better than providing every score 2011-09-21T08:09:03 yea thats why I eventually liked that. :-) 2011-09-21T08:09:51 great :) 2011-09-21T08:10:30 about 2) currently it is a bit hard to determine if the enemy ant is not moving if it has adjacent friendly ants 2011-09-21T08:11:02 you won't get to know anything more than what it in your LOS 2011-09-21T08:11:13 *what is in 2011-09-21T08:12:12 or it is hard to keep track of the enemy movement in your LOS 2011-09-21T08:13:25 you need to find out from which square the ant has come and sometimes it is not too easy 2011-09-21T08:13:39 if you would like to chase a particular ant 2011-09-21T08:13:40 *** Blkt has joined #aichallenge 2011-09-21T08:13:55 good afternoon everyone 2011-09-21T08:14:14 hi Blkt 2011-09-21T08:14:24 hi there 2011-09-21T08:15:02 *** bhasker has quit IRC (Ping timeout: 252 seconds) 2011-09-21T08:15:10 Blkt: o/ 2011-09-21T08:15:19 hi aerique 2011-09-21T08:15:21 how are you? 2011-09-21T08:15:40 at work so not too well :-) 2011-09-21T08:18:33 I'm at work too! 2011-09-21T08:18:38 started september 5th 2011-09-21T08:18:59 Blkt: cool, what & where? 2011-09-21T08:19:17 my first job using neural networks in industry (Monza, Italy) 2011-09-21T08:19:27 actually, my very first job 2011-09-21T08:19:47 it is unusually rare to be able to work with AI in Italy 2011-09-21T08:20:22 sounds good! 2011-09-21T08:20:35 sir_macelon : hey, gtg now. if you do get to write somewhere on exactly, how u like to see that implemented, send me pointer to there as well pls. would like to hear more on that. it's a nice chat. hve a good day! 2011-09-21T08:20:42 yes it does! 2011-09-21T08:20:45 hve a good day/evening all! 2011-09-21T08:20:50 Blkt: what kind of work? 2011-09-21T08:20:51 cya 2011-09-21T08:21:05 eashoka: bye! 2011-09-21T08:21:11 like i assume you're applying neural networks to stuff, what stuff? :P 2011-09-21T08:21:19 *** eashoka has quit IRC (Quit: Page closed) 2011-09-21T08:21:32 antimatroid: we need to do forecasting, I'm using nnets on timeseries 2011-09-21T08:22:05 weather, finance, health? 2011-09-21T08:22:22 computer load 2011-09-21T08:22:33 it should be easier 2011-09-21T08:22:39 then finance, weather and health I mean 2011-09-21T08:22:45 than 2011-09-21T08:22:49 wtf... 2011-09-21T08:23:12 congrats :) 2011-09-21T08:23:16 thanks :D 2011-09-21T08:24:18 also, I asked if I could use Clojure for this job and they said yes 2011-09-21T08:26:38 sir_macelon: I like the lack of info... that makes it challenging... in planet wars you had complete information. I not really a fan of such "games" 2011-09-21T08:26:52 except chess 2011-09-21T08:27:09 i do not want bots to know how many ants there are or its rank 2011-09-21T08:27:41 bots should try and maximize its score on what it does know. 2011-09-21T08:29:47 the dimensions give you some information about the number of players 2011-09-21T08:29:54 *** mcstar has joined #aichallenge 2011-09-21T08:30:19 we could fix dimensions to maximise the number of divisors below 10 and use that as a set of players if we wanted to fully hide it (i think 120 is best here from memory) 2011-09-21T08:30:40 but otherwise i think there's enough symmetries that people will find it hard enough to exploit symmetry 2011-09-21T08:32:07 delt0r: I know what you mean, but I don't see a point why we should hide a simple info for the bot if it is winning or not 2011-09-21T08:32:36 delt0r: anyway people will find a way the workaround this and some of them already did that 2011-09-21T08:33:07 why not drop fow too, right? 2011-09-21T08:33:10 * mcstar disagrees 2011-09-21T08:33:12 delt0r: by throwing it all in x last turns to maximize the score as the remaining ants are not being counted at all 2011-09-21T08:34:03 delt0r: this leads to rather artificial and odd behavior, than a neat game strategy 2011-09-21T08:35:45 delt0r: soon you'll notice everyone will go for on for all in last x turns, I don't think it is nice to encourage such "strategy" 2011-09-21T08:35:48 i really don't see it that way.. and right now if you get 2 points per ant killed. killing sprees are totally worth it 2011-09-21T08:36:00 why not? 2011-09-21T08:36:18 what is "wrong" with that strategy? 2011-09-21T08:36:40 everybody will do that then 2011-09-21T08:37:20 it won't be a strategy eventually it will be a must if you want to increase your score just before the game ends 2011-09-21T08:38:28 sir_macelon: we discussed NOT sending maxturns to the bots, at least i think its a good idea, not sure if it will happen 2011-09-21T08:38:29 there's nothing fancy about that strategy and I don't like it 2011-09-21T08:39:23 mcstar: it's a way to fix this, but then the bot will get even less information 2011-09-21T08:39:36 thats a pro for me :) 2011-09-21T08:39:51 id like a minimally invasive engine so to speAK 2011-09-21T08:40:22 I don't have anything against limiting the info a bot will receive 2011-09-21T08:41:06 but I strongly believe a bot has to know if it is winning or not, to plan its moves accordingly 2011-09-21T08:41:16 in that spirit id like to propose to limit viewradius to spawnradius! 2011-09-21T08:41:26 (lol, kidding) 2011-09-21T08:41:52 without it, there's too much luck on which bot decides to go for all or nothing 2011-09-21T08:42:15 trying to predict the game end, if the game is going pretty even 2011-09-21T08:42:35 sir_macelon: How you go about killing other ants will be far from boring/simple/easy/plain or anything else 2011-09-21T08:42:38 i think the bot should gather his info, and it already knows quite much, the size of the map for example, so it can make assumptions on how good its assumptions are 2011-09-21T08:42:43 one on one gets you *nothing* 2011-09-21T08:43:03 mcstar: yes 2011-09-21T08:43:11 that is what its all about IMO 2011-09-21T08:43:31 anyway at 1000 turns with the size maps we have... i doubt games won't finish 2011-09-21T08:44:51 that im not so sure of 2011-09-21T08:44:57 really depends on the actual bots 2011-09-21T08:45:17 (if you meant some kind of all-but-one elimination) 2011-09-21T08:45:21 with a low food spawn rate games can't be decided in 1500 turn even 2011-09-21T08:45:57 its toatally possible that some logic lead to never ending games 2011-09-21T08:46:02 is that true for smarter bots? 2011-09-21T08:46:04 with no apparent winners 2011-09-21T08:46:37 ie. bots that actually path find for food and enemies 2011-09-21T08:46:58 antimatroid: it depends on how easily you let yourself get killed 2011-09-21T08:47:17 the "smarter" the bots, the longer it can live 2011-09-21T08:47:22 they* 2011-09-21T08:47:48 kill it with fire! 2011-09-21T08:47:53 :) 2011-09-21T08:48:04 i say burn the whole damn house 2011-09-21T08:50:10 delt0r: ono on one gets you *nothing*? I don't agree. 2011-09-21T08:50:15 but theres a point in there somewhere, i guess people already mentioned this, its rather trivial, but for a long game, where bots try to avoid battle, and they play quite similarly, and 1 bot wins by a small margin, it absolutely shouldnt get any bonuses 2011-09-21T08:50:17 delt0r: take a look at this game: http://aichallengebeta.hypertriangle.com/visualizer.php?game=186863 2011-09-21T08:50:55 1 ant on 1 ant 2011-09-21T08:50:57 delt0r: one of the bots goes for all one on one and eventually wins the game having lest ants 2011-09-21T08:50:58 he meant 2011-09-21T08:51:10 mcstar: I know 2011-09-21T08:51:22 just see the game 2011-09-21T08:51:36 or last 50 turns rather 2011-09-21T08:51:50 now every bot will have to do that 2011-09-21T08:52:59 another example: http://aichallengebeta.hypertriangle.com/visualizer.php?game=177970 2011-09-21T08:53:19 i dont get it 2011-09-21T08:53:26 mathis could have done the same for example 2011-09-21T08:53:28 7 high ranked bots and game wasn't decided after 1500 turns - all bots have survived 2011-09-21T08:53:29 but he didnt 2011-09-21T08:53:31 so what? 2011-09-21T08:53:39 that's the point 2011-09-21T08:53:48 everybody know should be doing this 2011-09-21T08:54:02 that is not a strategy that is a must 2011-09-21T08:54:03 greentea smartly attacked those ants, which was suicidal too, and mathis i think tried to avoid its enemies 2011-09-21T08:54:09 to bias the scoring system 2011-09-21T08:54:16 so greentea could gather points over mathis 2011-09-21T08:54:19 if only one bot does this, he gets a lot of +2 bonuses from 1-on-1 fights, while the "other side" +2s get split between all enemies he fights 2011-09-21T08:54:19 I dont; think the beta server is a good reflection of what good bots will be 2011-09-21T08:54:31 is not folks a game effort 2011-09-21T08:54:34 IMO 2011-09-21T08:54:37 for the most part 2011-09-21T08:55:14 you might be wrong here 2011-09-21T08:55:17 "is not folks a game effort" ? 2011-09-21T08:55:23 the beta is 4 months old now 2011-09-21T08:55:42 delt0r: help me i have trouble understanding that line 2011-09-21T08:55:43 and there are pretty advanced bots already 2011-09-21T08:56:17 noway, sir_macelon dont you tell me that youre game logic doesnt fit on 1 screen 2011-09-21T08:56:25 -e 2011-09-21T08:56:48 the beta page even says don't put "real" entries there 2011-09-21T08:57:06 that is for governmental workers 2011-09-21T08:57:14 with supersecret ai works 2011-09-21T08:57:28 delt0r: and you think people are not doing this? 2011-09-21T08:57:37 A game effort.. ie its not the comp where folks will be putting more hours etc and be far more competitive 2011-09-21T08:57:53 since rules and game scoring can all change 2011-09-21T08:57:54 etc 2011-09-21T08:58:00 mcstar: I am developing my bot like since april 2011-09-21T08:58:45 so it is rather complicated and the room for improvement is more and more limited 2011-09-21T08:59:17 yeah, well, i think i started right about then 2011-09-21T08:59:44 i worked up a couple of versions, than recently i gave it a rewrite 2011-09-21T09:00:08 and the bot that i will be using in the real contest will be an almost total rewrite if i can manage 2011-09-21T09:02:32 either way I really dislike this post scoring adjustment, and i don't like the idea of giving rank info or no. of players info 2011-09-21T09:02:55 if you want more combat just bump up the kill score 2011-09-21T09:03:24 well, if you know that which enemy belongs to which hive, can lead to sophisticated decisions, im somewhat inclined to think that is a good thing 2011-09-21T09:04:24 if they can't handle that much hidden info, then i am doubtful of the level of sophistication tbh 2011-09-21T09:04:54 what do you mean hidden info? 2011-09-21T09:05:20 if the engine doesnt send the id's of enemy ants, you have little way of knowing who are you fighting with 2011-09-21T09:05:53 recognizing previous actions of encountered ants doesnt give you the ability to predict an ant you just encountered 2011-09-21T09:06:07 ie scores and numbers of players 2011-09-21T09:06:18 or ranks/number of players 2011-09-21T09:06:41 they dont actually receive scores 2011-09-21T09:06:54 *** bhasker has joined #aichallenge 2011-09-21T09:06:56 but what is your opinion on enemy id's? 2011-09-21T09:07:06 yea--but this is what was suggested above and i really don't like it.. 2011-09-21T09:07:24 (sure, they shouldnt receive scores :) ) 2011-09-21T09:07:25 delt0r: current situation leads to the bahvior like in the game I posted. Do you find it acceptable then? 2011-09-21T09:07:38 enemy id's? what do you mean. currenty you do get each ants "colour" or team... 2011-09-21T09:07:51 thats is an id 2011-09-21T09:07:56 from where i come from 2011-09-21T09:08:31 why not? --really why not... why this strong expectation of how things should turn out, and then change anything to "fix it" when they don't? 2011-09-21T09:10:06 i think this is just a different preferance of people: some consider full-informationgames more interesting than the opposite 2011-09-21T09:10:28