2011-04-10T00:03:38 amstan: about what things? 2011-04-10T00:03:48 everyone except stocha that i know wants damage for battle 2011-04-10T00:03:52 the stuff jeff has to decide about 2011-04-10T00:03:59 and i am yet to come across anyone that doesn't understand it right away 2011-04-10T00:04:04 mcleopold has been doing a lot of the ants managing 2011-04-10T00:04:09 he might be able to do this too 2011-04-10T00:04:19 *** dlila has quit IRC (Quit: Leaving) 2011-04-10T00:15:15 *** amstan_ has joined #aichallenge 2011-04-10T00:15:15 *** ChanServ sets mode: +o amstan_ 2011-04-10T00:18:35 *** amstan has quit IRC (Ping timeout: 260 seconds) 2011-04-10T00:21:45 *** Mathnerd314 has quit IRC (Ping timeout: 248 seconds) 2011-04-10T00:29:24 *** bss03 has quit IRC (Ping timeout: 252 seconds) 2011-04-10T01:19:26 *** bss03 has joined #aichallenge 2011-04-10T01:20:22 *** bss03 has joined #aichallenge 2011-04-10T01:33:10 *** Stocha has joined #aichallenge 2011-04-10T01:35:59 @later tell antimatroid i said i do SUPPORT the damage rule. Where am i against it ? 2011-04-10T01:35:59 Stocha: Job's done. 2011-04-10T01:36:39 @later tell amstan I do support the damage rule. 2011-04-10T01:36:39 Stocha: I think that worked... 2011-04-10T01:37:47 *** Stocha has quit IRC (Client Quit) 2011-04-10T01:42:46 *** fgump has joined #aichallenge 2011-04-10T01:50:06 *** Chris_0076 has quit IRC (Remote host closed the connection) 2011-04-10T02:15:23 *** dr- has quit IRC (Ping timeout: 252 seconds) 2011-04-10T02:22:47 *** dr- has joined #aichallenge 2011-04-10T03:03:48 *** _flag <_flag!~flag@69-165-173-172.dsl.teksavvy.com> has quit IRC (Quit: Lost terminal) 2011-04-10T03:28:50 *** delt0r_ has joined #aichallenge 2011-04-10T03:30:05 *** delt0r___ has quit IRC (Ping timeout: 246 seconds) 2011-04-10T03:36:55 *** Kender has joined #aichallenge 2011-04-10T03:39:11 *** Kender_away has quit IRC (Ping timeout: 246 seconds) 2011-04-10T03:39:12 *** Kender is now known as Kender_away 2011-04-10T03:48:02 *** Kender_away has quit IRC (Quit: HydraIRC -> http://www.hydrairc.com <- Po-ta-to, boil em, mash em, stick em in a stew.) 2011-04-10T04:13:44 *** p4p4p5 has joined #aichallenge 2011-04-10T04:29:24 *** dr- has quit IRC (Ping timeout: 276 seconds) 2011-04-10T04:37:59 *** fgump has quit IRC (Quit: Leaving) 2011-04-10T04:48:18 *** FireFly has quit IRC (Quit: swatted to death) 2011-04-10T04:56:37 *** dr- has joined #aichallenge 2011-04-10T05:18:35 *** dr- has quit IRC (Ping timeout: 240 seconds) 2011-04-10T05:23:14 *** woudshoo has joined #aichallenge 2011-04-10T05:30:08 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T05:30:55 *** dr- has joined #aichallenge 2011-04-10T05:32:59 *** p4p4p5 has joined #aichallenge 2011-04-10T06:13:00 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T06:13:48 *** p4p4p5 has joined #aichallenge 2011-04-10T06:27:18 has anyone that's around played with food collection 2011-04-10T06:27:54 i'm quite worried about how long my first food collection heuristic is 2011-04-10T06:28:14 people are going to be severely limited on cool/interesting strategies that are feasible with the current dimensions/no ants 2011-04-10T06:39:59 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T06:41:17 antimatroid: Hm, food collection takes sometime, but in my experience quite easy 2011-04-10T06:41:31 woudshoo: it's not as easy as i currently thought 2011-04-10T06:41:38 currently working on something that should speed it up quite a bit 2011-04-10T06:41:47 and i have a few things i could do that should cut down some more time 2011-04-10T06:41:48 *** p4p4p5 has joined #aichallenge 2011-04-10T06:41:57 previously thought* 2011-04-10T06:42:11 woudshoo: currently my food collection is as follows 2011-04-10T06:42:53 Hm, what are you thinking about. I am using a stupid greedy algorithm. I had some troubles with speed but that was because of a stupid mistake. I should time it again though, because it taks some time. 2011-04-10T06:43:49 I think it takes at most about 0.1 a 0.2 sec with about 300 ants. If there are more ants, I think I will limit the number of ants used for food gathering 2011-04-10T06:44:26 *** Apophis_ has quit IRC (Ping timeout: 276 seconds) 2011-04-10T06:45:05 create a grid of integers specifying when a square is available, -1 for not, 0 for starting ants 2011-04-10T06:45:06 iteratively perform a bfs from all "non-collected" food squares simultaneously (i'm currently doing that sequentially, so should speed it up quite a bit) until the current distance is >= current best distance, then if the current square is available and curDist + availableTime < bestDist, use the new information, then once i've chosen a move, move the ant, and update available for the new ant and where that ant will be 2011-04-10T06:45:25 i also want to try and improve that to avoid crossing paths 2011-04-10T06:45:29 but even then it's not optimal 2011-04-10T06:45:55 it can expand about 4-5 times faster than hunter bot though, and it doesn't move an ant currently unless that ant has been assigned a move to collect food 2011-04-10T06:47:17 i quite like that though, it works pretty damn well 2011-04-10T06:47:27 This is roughy what I do. I have a grid (map) keeping track of (food, distance) pairs. Initially this is empty. A worker queue also queues (food, location, distance). 2011-04-10T06:47:57 i think i'm killing myself by searching from each food square sequentially on each iteration 2011-04-10T06:48:00 i'm currently improving that 2011-04-10T06:48:54 An element is taken from the queue, it looks in the map of its neighbours. It a neighbours has a distance value > queue element distance + 1, add it the neighbour to the queue. 2011-04-10T06:49:13 do you move an already moved ant to get more food? 2011-04-10T06:49:22 If it discovers an ant, move the ant closer to the food. Important though: REMOVE all references for that food item from the queue and continue. 2011-04-10T06:49:26 like say you have *....aa.** 2011-04-10T06:49:33 will you send both a's right or one lefT? 2011-04-10T06:49:47 i will send the left one left in that situation 2011-04-10T06:51:54 It will send the left ant left, but only because (situation: Aa.*#' ) the algorithm first walks a -> * and than tries to walk A-># around 'a' and '#' so the left ant is closer 2011-04-10T06:52:05 ah okay 2011-04-10T06:52:14 can you see what i mean with mine though? 2011-04-10T06:52:29 it will update information each iteration to use "future ants" to collect food squares 2011-04-10T06:52:32 do you do that? 2011-04-10T06:52:37 that's the crux to my algorithm 2011-04-10T06:53:54 No, I don't update future ant creation. My algorithm is currently quite stupid and greedy. But honestly, I don't know how big a difference it will make. (as long as you don't get stuck and can find a path to the food.) 2011-04-10T06:54:09 it makes quite a difference i assure you 2011-04-10T06:54:28 otherwise you send more ants than necessary to collect certain food squares 2011-04-10T06:54:42 which could be off doing something else 2011-04-10T06:55:07 ah, i remember why i did it iteratively 2011-04-10T06:55:18 i don't want to go for a food square that has an enemy closest to it 2011-04-10T06:55:45 crap 2011-04-10T06:55:57 Ah, I ignore enemies. I have thought about avoiding enemies, but I changed my mind about that. 2011-04-10T06:56:19 you may then miss out on food squares you could have gotten 2011-04-10T06:56:34 and if the ant is that crap that it let you get the food, you probably can beat them anyway if you're half decent 2011-04-10T06:56:40 that's my personal rational 2011-04-10T06:58:07 a moment, got phone call 2011-04-10T07:02:09 antimatroid: about enemies, my thinking is that, (1) food squares with no enemy close to it, you will get anyway. (2) if you walk to the food and manage to get a 1-1 kill, the score differences won't change. 2011-04-10T07:02:43 So it seems a toss up at the moment. However I haven't done anything about fighting yet. 2011-04-10T07:02:43 it might have to be done for speed anyway 2011-04-10T07:02:51 i haven't done fighting either 2011-04-10T07:03:02 food collection was meant to be FAST so i can trees for battles 2011-04-10T07:03:24 i want to use multiple trees based on which ants can a/effect each other 2011-04-10T07:03:46 then all other ants will be moved towards the closest square that hasn't been seen in the last x turns 2011-04-10T07:03:58 probably not iteratively at this point :( 2011-04-10T07:04:34 Interesting tidbit that last thing. I was thinking something similar, keeping a map of probability of food being present, the longer you haven't seen it the higher it gets. Together with a map of enemy probability. 2011-04-10T07:05:06 you could use markov chains to get probabilities of ants locations 2011-04-10T07:05:12 Yes, excatly! 2011-04-10T07:05:13 but it'd be rank and probably not all that useful 2011-04-10T07:05:23 i considered that too :) 2011-04-10T07:05:58 i want to get a tcp server going already :P 2011-04-10T07:06:21 *** Eruonen has joined #aichallenge 2011-04-10T07:06:43 I think it might payoff for the enemy distribution, if you expand your area and the middle is not visible, you at least know there is no enemy ant present. So deployment of forces might change on that map. 2011-04-10T07:07:12 yeah 2011-04-10T07:07:17 it would be really cool to see such a bot win 2011-04-10T07:07:52 i originally wanted to try and model the probabilities of moves the enemies might make with planet wars, but it was all too messy to ever get that far 2011-04-10T07:08:05 based on their past moves etc. 2011-04-10T07:08:18 *** needsch has joined #aichallenge 2011-04-10T07:09:58 I think it will also be crucial to try to attack as many different enemies as possible. If you manage to engage 3 enemies while the other bots only engage 1 or 2 you have a real advantage in the scoring rules. 2011-04-10T07:10:42 i plan to use a game tree to avoid anything where i don't kill more than i lose until it's getting towards the end at which point i'll start trying to kill off ants 1:1 2011-04-10T07:11:12 although one is obviously worried about how well the game trees will work 2011-04-10T07:11:53 my fallback plan might have to be move ants near enemies towards their closest friend 2011-04-10T07:12:04 that's an easy battle strategy to implement though 2011-04-10T07:12:10 I was thinking different, if you can kill 1:1 go for it! For example, if I kill 100 ants 1:1 with enemy B and you kill 2:1 but only 50 times with me, I have about 130 points and you 100 2011-04-10T07:12:24 but with 4 other players it's not that simple 2011-04-10T07:12:32 you might have another enemy with a better kill ratio 2011-04-10T07:12:39 4+ players 2011-04-10T07:13:48 At the moment, I think fighting is really hard. (Just tried on paper.) I think just making big lines of ants and move them in a synchronized manner might be good. 2011-04-10T07:13:58 Depends on how narrow the corridors are 2011-04-10T07:14:05 game trees :P 2011-04-10T07:14:20 i have an idea for diverting ants around "traps" (blocked, corridors) 2011-04-10T07:14:55 Well I am worried that if you have about 100 ants on both sides simultanuously in battle the game trees might be unwieldly 2011-04-10T07:15:08 yep definitely 2011-04-10T07:15:41 i plan to partition "battling ants" into the sets of ants that can effect each other, go through for as long as possible, then revert back to moving towards the closest friend or whatever other heuristic 2011-04-10T07:16:19 I was thinking of picking random close enemy ants and gang up on them. 2011-04-10T07:16:39 (obviously not completely random) 2011-04-10T07:16:44 that's an alright naive idea 2011-04-10T07:16:59 i think game trees are win if we can get them going in time 2011-04-10T07:17:10 big if though 2011-04-10T07:17:40 there are cool options for backward induction on the n-player game tree too 2011-04-10T07:18:24 sometimes offering potential cooperation situations, where one could use tit for tat as to whether they are willing to cooperate with the player 2011-04-10T07:18:57 yeah, very big if. Also suppose someone just position a big block of ants and let them sit. I am worried that the game tree will just decide it is not worth attacking, and thinks it is best to just sit there. So the evaluation function should take more into account than just ants killed. (Area?) 2011-04-10T07:19:34 yeah, i have an idea for trying to place "temporary water squares" around squares that would be stupid to move into 2011-04-10T07:19:39 ah, tit for tat. Well, interesting, how are thinking detecting 'cooperation vs defection (cheating)' 2011-04-10T07:19:41 so one can divert ants around "traps" 2011-04-10T07:20:34 otherwise the ants will build up there and just wait 2011-04-10T07:20:38 which is obviously a waste 2011-04-10T07:20:41 I was thinking for that to make a map with a desirability value and a danger value and use path finding to guide the ants to the desirable squares 2011-04-10T07:21:12 Also as a random other idea, I was thinking of deviding the ants in classes, fighter ants, food gathering ants. 2011-04-10T07:21:53 Yes, building up is tricky. I will first go for fight fight fight, even if I lose some. Just to break down the defence. 2011-04-10T07:23:05 *** JamesMG has quit IRC (Read error: Operation timed out) 2011-04-10T07:24:10 antimatroid: Your food algorithm just sunk in. Clever!! 2011-04-10T07:24:29 :) i quite like it 2011-04-10T07:24:59 it'd be even better with a map for what turns a location will be occupied on ants paths, and take that information into account when searching 2011-04-10T07:25:10 but i'll worry about that once i can do what i already am fast enough 2011-04-10T07:25:55 http://pastebin.com/tDpVPiJw 2011-04-10T07:26:04 there's a replay file of it against two null bots and hunter bot 2011-04-10T07:27:45 w.r.t. avoiding future ant position, It really depends on the maps though. The current map I use for testin is quite open. So I see ants as obstacles. It still sees a path and next turn it will see the space is empty. However if the maps become more maze like this will be really bad. 2011-04-10T07:28:44 mine will currently try and go around the ant instead of "waiting", making it calculate stuff incorrectly 2011-04-10T07:28:50 bbiab, gotta cook some food 2011-04-10T07:36:40 i like watching it expand 2011-04-10T07:38:32 I am trying to get watch your replay file, but I need to fetch and merge from git to see if the bugs I am encountering in the Replay.js are fixed 2011-04-10T07:39:48 it loads for me 2011-04-10T07:40:52 lol, i was using the old visualiser until then 2011-04-10T07:41:11 hunterbot beat me by one point, but not too surprising with 2 players and me doing nothing about getting ants near each other 2011-04-10T07:42:57 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T07:43:52 *** p4p4p5 has joined #aichallenge 2011-04-10T07:46:17 *** Naktibalda has joined #aichallenge 2011-04-10T07:48:03 Ha, I know that experience, my current version sometimes had about 700 ants and was beaten by a hunterbot having 50. At the moment I am actively trying to spread out my ants. But the algorithm does not work as I expect. 2011-04-10T07:53:45 i had to tell this one to stop before the time limit to stop it from timing out 2011-04-10T08:00:27 turns out i wasn't even checking that the food squares were closer to me than an enemy 2011-04-10T08:04:29 What I notice is that the results vary wildly between runs. Most of the time my current version 2011-04-10T08:04:50 beats the Hunterbots comfortably, but also quite often it does not. 2011-04-10T08:06:14 are you doing 1v1? 2011-04-10T08:06:41 if the food bonus doesn't come into play with 1v1, then the result is who fought the best, nothing else 2011-04-10T08:06:55 whereas with more players, if you get eliminated early then you stop scoring 2011-04-10T08:07:03 so you need to worry about growth of your colony as well 2011-04-10T08:07:23 No, I am doing 3 hunterbots against me. 2011-04-10T08:07:28 ah okay 2011-04-10T08:07:37 got a replay file for me? :P 2011-04-10T08:09:20 Yeah, I am uploading it right now. (never used pastebin before.) 2011-04-10T08:10:32 http://pastebin.com/XS31EPcw 2011-04-10T08:10:47 My bot has the most ants at the end (blue I think.) 2011-04-10T08:11:01 i just copy paste to pastebin 2011-04-10T08:11:15 codepad is good if you have a single file you want to show the output of too (it'll run it) 2011-04-10T08:12:38 i think there's way too many ants on the maps 2011-04-10T08:12:51 not that game, all of them 2011-04-10T08:12:53 Never heard of codepad. www.codepad.org? com? 2011-04-10T08:13:16 .org 2011-04-10T08:13:44 Yes, I agree, there are some many ants that you don't have to worry about not having enough ants. 2011-04-10T08:14:10 yeah, and you can basically just funnel them all to the front lines 2011-04-10T08:14:11 *** KEth_ has joined #aichallenge 2011-04-10T08:14:32 KEth_: how are you doing food collection? I have a replay file for you :P 2011-04-10T08:14:40 Hm? 2011-04-10T08:14:40 http://pastebin.com/tDpVPiJw 2011-04-10T08:14:42 Exactly, that was my reason for wanting to engage in all 1-1 loss battles. 2011-04-10T08:14:51 What's that? 2011-04-10T08:14:55 the ants are doing nothing but running my food collection 2011-04-10T08:14:59 that's a replay file 2011-04-10T08:15:27 in visualizer open copy_paste.html and paste that to there 2011-04-10T08:15:35 is red my obt? 2011-04-10T08:15:36 bot? 2011-04-10T08:15:47 it's me vs hunter bot 2011-04-10T08:16:09 red is yours that stops? 2011-04-10T08:16:58 yep 2011-04-10T08:17:05 you mind if I run it with my bot? 2011-04-10T08:17:07 i am only moving an ant that is to be used for food collection 2011-04-10T08:17:16 I'm quite curious to see the results 2011-04-10T08:17:19 i would like to, but not really sure how to give you a runnable version 2011-04-10T08:17:31 we need a tcp server :( 2011-04-10T08:17:32 python? 2011-04-10T08:17:35 you could send "pyc" 2011-04-10T08:17:36 c++ 2011-04-10T08:17:38 Oh 2011-04-10T08:17:40 send me exe then? 2011-04-10T08:17:47 I did my one in C++ as well 2011-04-10T08:17:56 will it run though? 2011-04-10T08:18:00 of course it will 2011-04-10T08:18:03 pm me your email 2011-04-10T08:19:22 let me know once sent 2011-04-10T08:19:32 done 2011-04-10T08:19:41 well it's sending 2011-04-10T08:19:50 there we go 2011-04-10T08:20:03 it will be super slow :P 2011-04-10T08:20:06 i'm working on that 2011-04-10T08:20:07 It's not .exe? 2011-04-10T08:20:15 i compiled on ubuntu 2011-04-10T08:20:19 lol 2011-04-10T08:20:22 I'm on windows 2011-04-10T08:20:24 hence my question :P 2011-04-10T08:20:35 well, I thought you will compile under win 2011-04-10T08:20:45 nah i can't be bothered restarting 2011-04-10T08:20:51 and i'm fairly new to ubuntu 2011-04-10T08:21:11 mkay, that's fine. 2011-04-10T08:21:15 feel free to compile under windows any time 2011-04-10T08:21:58 what it's doing is iteratively moving ants to the closest food square, but then possibly using that ant and the new ant from their future squares with the following food squares 2011-04-10T08:22:05 accounting for when they will be available 2011-04-10T08:22:20 yeah, we'll see how it turns out some day 2011-04-10T08:22:47 are you using any pathing algorithm? 2011-04-10T08:23:06 just breadth first searches from food squares 2011-04-10T08:23:22 I see 2011-04-10T08:23:24 could've been improved i noticed 2011-04-10T08:23:29 i'm doing it sequentially (as i was planning to not move to a food square closer to an enemy), but it's slow as hell 2011-04-10T08:23:36 it could be much better 2011-04-10T08:23:38 sometimes I see your ant doing 3 moves but it could do 2 2011-04-10T08:23:44 really? 2011-04-10T08:23:49 that's a bug 2011-04-10T08:23:52 of course 2011-04-10T08:23:53 :P 2011-04-10T08:24:00 because you're not testing diagonally 2011-04-10T08:24:06 you pickup food when you're diagonal to it 2011-04-10T08:24:11 oh, yeah i'm ignoring that atm 2011-04-10T08:24:23 so you're wasting one move quite often 2011-04-10T08:24:39 I shouldn't be helping :) 2011-04-10T08:24:40 yeah that's alright, i'll worry about that once the spawn radius is settled 2011-04-10T08:24:42 Haha 2011-04-10T08:24:50 i'm fairly keen on that being 1 anyway 2011-04-10T08:25:01 i don't like spawning over water 2011-04-10T08:29:38 *** woudshoo has quit IRC (Remote host closed the connection) 2011-04-10T08:32:46 *** woudshoo has joined #aichallenge 2011-04-10T08:50:09 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T08:51:52 *** p4p4p5 has joined #aichallenge 2011-04-10T08:58:56 *** mceier has joined #aichallenge 2011-04-10T09:00:45 *** skynetz has joined #aichallenge 2011-04-10T09:15:55 *** Stocha has joined #aichallenge 2011-04-10T09:17:23 i have some simple food collection algorithm that seems robust time wise. 2011-04-10T09:17:53 by robust i mean : i can do it under a second :p 2011-04-10T09:18:50 how does it scale with map size, nr of food and nr of ants? I haven't had any problems time wise and although it is not perfect it is good enough for now. 2011-04-10T09:19:05 i didn't really made any measure 2011-04-10T09:19:10 nor did i do any optimizations 2011-04-10T09:19:16 it's roughly a prototype 2011-04-10T09:19:18 *** Eruonen has quit IRC (Ping timeout: 252 seconds) 2011-04-10T09:19:24 and the bot is freely available for testing :p 2011-04-10T09:19:45 It can be used to compare your to what i like to call the reference method :p 2011-04-10T09:19:53 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T09:20:07 *** Eruonen has joined #aichallenge 2011-04-10T09:20:22 Oh, i thought maybe you have a feeling for it. Mine is roughly O (#food x #avr dist to food^2) 2011-04-10T09:20:45 Mine depends on map size for sure. 2011-04-10T09:21:01 and it depends on number of food to grab. 2011-04-10T09:21:07 *** dlila has joined #aichallenge 2011-04-10T09:21:35 It's a simple greedy algorithm : launch the nearest ant to the nearest food, if no ant is already going to collect it. 2011-04-10T09:21:49 *** p4p4p5 has joined #aichallenge 2011-04-10T09:22:03 i would love to have comparison of it against more advanced strategy. 2011-04-10T09:22:42 I think we are all inventing similar methods. antimatroid and I have similar strategies. His is quite clever if I understood it correctly 2011-04-10T09:23:12 is it an advanced one, or basic greedy one ? 2011-04-10T09:23:23 (i'm reading the log, but didn't reach that part yet :p) 2011-04-10T09:23:36 Stocha: slightly more advanced, but really slow atm 2011-04-10T09:23:48 but it compares it with the hunter bot. That's not a good reference. 2011-04-10T09:23:52 there is a replay file for you to watch 2011-04-10T09:23:53 Mine is a basic greedy one, closet ant to closest food, discard those from consideration, and continue 2011-04-10T09:24:00 can't you compare it with my bot antimatroid ? 2011-04-10T09:24:09 that would be better than using hunterbot as a reference. 2011-04-10T09:24:09 Stocha: maybe tomorrow 2011-04-10T09:24:16 i'm only looking at food collection 2011-04-10T09:24:23 yeah 2011-04-10T09:24:29 but if the map is big enougth 2011-04-10T09:24:29 i'm not moving an ant unless it needs to be used for collecting food 2011-04-10T09:24:38 you can watch how it does in the beginning 2011-04-10T09:24:44 and have some sort of ratio. 2011-04-10T09:25:01 ah you don't do any exploration ? 2011-04-10T09:25:14 well my ref bot isn't good at exploring. 2011-04-10T09:25:15 put it this way, i make it to like 500 ants by the time hunter bot as 120 2011-04-10T09:25:24 yeah 2011-04-10T09:25:28 and i don't explore at all, just collect what's visible 2011-04-10T09:25:30 but hunter bot is disastrous 2011-04-10T09:25:52 chances are your current speed is no more than the simple greedy strat. 2011-04-10T09:26:01 because you don't use free ant to accelerate it. 2011-04-10T09:26:06 it's smarter than that 2011-04-10T09:26:23 yeah i understand it's smarter 2011-04-10T09:26:31 if it moves an ant, it'll consider the future location of that ant and the new one (accounting for when it's available) to continue collecting 2011-04-10T09:26:34 because it use what i call the "rebound food collection" probably :p 2011-04-10T09:26:47 yeah that's it 2011-04-10T09:27:09 still, if you do not use your free ant for exploring, you would end up with same food gathering speed as the greedy strategy. 2011-04-10T09:27:34 no, it's still better 2011-04-10T09:27:44 okay 2011-04-10T09:27:53 it'll just be even better with exploration and proper fighting 2011-04-10T09:27:55 i guess it can split two ant in an efficient way 2011-04-10T09:28:09 exactly 2011-04-10T09:28:13 you should compare the two method then :p 2011-04-10T09:28:37 is the slownest proportional to food seen ? 2011-04-10T09:28:49 Stocha: at the moment it seems that just going after visible food already forces quite a bit of exploring in the beginning. Later on or on different maps it seems more important. 2011-04-10T09:29:20 yeah, i'm being lazy 2011-04-10T09:29:30 it really isn't hard to update to what should be faster, reddit is distracting 2011-04-10T09:31:26 woudshoo: that's awfully dependent on map settings ... 2011-04-10T09:32:18 Stocah: Yes, indeed, are there any maps which are more challenging? (I could make my own of course, hm maybe, after I am done making my own visualizer) 2011-04-10T09:32:46 probabilities are hard to handle when you don't know the rules :p 2011-04-10T09:32:53 woudshoo: you are making a visualiser... 2011-04-10T09:33:03 did you use jbotmanager at all for planet wars? 2011-04-10T09:33:07 woudshoo : there are map generators built in 2011-04-10T09:33:09 i would love to release something like that officially 2011-04-10T09:33:16 you just have to plug them 2011-04-10T09:33:20 i know keth did that. 2011-04-10T09:33:29 i did'nt do any testing with the official engine myself. 2011-04-10T09:33:33 keth just changed frontiers 2011-04-10T09:33:50 he used the map generators for sure 2011-04-10T09:34:01 i have seen game with asymetric map 2011-04-10T09:34:01 antimatroid: well, actually more for debugging purposes. I am wanting to visualize internal state of my bot. 2011-04-10T09:34:38 i think everybody needs some way to look at internal state. 2011-04-10T09:35:12 ASCII being the worst for it i suppose :p 2011-04-10T09:35:24 still enough in some cases. 2011-04-10T09:35:24 i like ascii 2011-04-10T09:35:51 when you compare with more visual method, ascii becomes weak :p 2011-04-10T09:36:00 but it can be deployed faster in most cases. 2011-04-10T09:36:01 Well, the visualizer will nothing be too fancy, most probably a collection of images that can be played back and forward with a slideshow program 2011-04-10T09:37:15 i do wish c++ had tuples 2011-04-10T09:37:18 If you want do have exploratin right, you need to use food probability for unknown squares, as well as ground probabilities 2011-04-10T09:37:29 wich will most probably be unavailable. 2011-04-10T09:37:34 stocha, the only thing your bot does not consider is moving into direction of where most food is 2011-04-10T09:37:40 rather, you move into nearest food 2011-04-10T09:37:54 keth: my bot is a simple greedy strategy 2011-04-10T09:37:58 yeah, I know 2011-04-10T09:37:59 i have that issue 2011-04-10T09:38:04 look at antimatroid strategy for more advanced ones. 2011-04-10T09:38:22 It's fine 2011-04-10T09:38:26 I am okay :) 2011-04-10T09:38:27 i have made prototypes where it does go with the densest patch of food 2011-04-10T09:38:33 doesn't sound too promising though 2011-04-10T09:38:41 nah i don't trust that 2011-04-10T09:38:55 Stocha: I am planning on using probabilities. Even if you don't know the exact probabilities, just extrapolated ones based on visible data is better than nothing. 2011-04-10T09:39:00 i've also considered going for the food squares closest to me that are closest to an enemy 2011-04-10T09:39:00 The real food collecting algorithm, goes in pair with exploring though. 2011-04-10T09:39:38 woudshoo, that doesn't too pratical. I have no idea how to do it. If map are very diversified it wont work 2011-04-10T09:39:52 for example : are unseen near water more likely to be water ? 2011-04-10T09:39:57 yeah, there is also food proritizing that would depend on enemy ants, if it can be safely taken later on and so on 2011-04-10T09:40:12 true 2011-04-10T09:40:22 I can't spoil how my system works 2011-04-10T09:40:23 but it does that 2011-04-10T09:41:07 i'm afraid game trees for fightings will be too slow. Unless you have something very clever indeed. 2011-04-10T09:41:17 Game trees would be fast honetsly 2011-04-10T09:41:25 would they ? 2011-04-10T09:41:29 if you know how to do it properly, using stack with pointers for example, rather than list 2011-04-10T09:41:32 the branching factor is quite tricky 2011-04-10T09:41:37 espescially if you do it in C++ 2011-04-10T09:42:17 KEth_: is it? battles between 100 ants vs 100 ants each can do about 3-5 valid moves gives 3^100 combinations at least, unless you do something smart. 2011-04-10T09:42:18 except the branching factor is huge and game trees are massive 2011-04-10T09:42:41 well branching factor is 5 max, but for the number of ants, that's massive 2011-04-10T09:43:11 but you can partition sets for game trees 2011-04-10T09:43:21 and do that for as long as possible before reverting to heuristics 2011-04-10T09:43:46 woudshoo: I have been dealing with massive real-time physics simulation engines, I'm quite familiar with large amount of combinations. There are really plenty of ways to optimize. 2011-04-10T09:44:16 woudshoo: There are plenty on my head and I can imagine tree data wouldn't be slow to go up to 10 possible moves. 2011-04-10T09:45:33 including enemy moves 2011-04-10T09:46:48 KEth_: Hm, interesting. I always thought that physic engines were more in the O(objects^2) complexity. [Note I know nothing whatsoever about this.] Maybe I need to check out the literature. 2011-04-10T09:47:07 n2 might not be a problem. If it's fast enough 2011-04-10T09:48:00 and you limit the maximum number of ants you put into your tree somehow 2011-04-10T09:48:14 i' really wonder how tree do with simultaneous turns games also. 2011-04-10T09:48:23 it becomes quite noisy i would guess 2011-04-10T09:48:58 O(objects^2) are really the most basics of collision detection. Take mpm grid based simulation as an example. You can have infinite number of particles, pretty much until you ran out of ram, rather than cpu calculations. 2011-04-10T09:49:07 because you theoretically can have rock/paper/scisor situation. 2011-04-10T09:49:37 mpm grid based, have irregular grids, or regular ones ? 2011-04-10T09:49:47 irregular grids? 2011-04-10T09:49:57 mm, i'm not clear :p 2011-04-10T09:50:02 I have shown you the pictures, remember? 2011-04-10T09:50:06 it works the same way. 2011-04-10T09:50:12 Let's you want to have collision detections, in a thing with big empty parts 2011-04-10T09:50:24 *** needsch has quit IRC (Read error: Connection reset by peer) 2011-04-10T09:50:32 uh well, i remember some pictures. 2011-04-10T09:50:33 *** needsch has joined #aichallenge 2011-04-10T09:53:01 the idea is I can imagine tree data be stored and computed on bitmap, rather than calling functions. Bitmap then can be minimized to 1d array or stack structure. So you end up with having seriously fast framework and you compute everything without a single call to any function, just pure maths like physics does. 2011-04-10T09:53:21 physics engiens* 2011-04-10T09:53:23 *** amstan_ has quit IRC (Ping timeout: 246 seconds) 2011-04-10T09:54:40 KEth_: As I said, I don't know anything of physics engines, and I don't see how the techniques would apply. If you pull it of, I will be seriously impressed. 2011-04-10T09:54:57 i have seen a prototype of his bot. 2011-04-10T09:55:03 it made me very happy to look at it :) 2011-04-10T09:55:12 It was mediumly efficient 2011-04-10T09:55:17 with some strength and weakness 2011-04-10T09:55:25 very different from the main thread of thinking though. 2011-04-10T09:55:27 impressive. 2011-04-10T09:58:34 yeah we badly need a tcp server. Amstan on the forum said : there is already all the code in there. Let's wait for people to share it :p 2011-04-10T09:58:42 i understood it this way at least. 2011-04-10T09:59:59 *** p4p4p5 has quit IRC (Read error: Connection timed out) 2011-04-10T10:00:13