2011-03-27T00:00:02 or use McLeopold's viewer. that should also still work 2011-03-27T00:00:05 antimatroid: why is that? 2011-03-27T00:00:27 it happens when you have the edge with a wall longer than 1 square i tihnk 2011-03-27T00:00:31 @seen bss003 2011-03-27T00:00:31 Frontier: I have not seen bss003. 2011-03-27T00:00:33 i'm pretty sure that breaks it 2011-03-27T00:01:26 sigh: how do you plan to fix it now? :P 2011-03-27T00:01:29 @seen bss03 2011-03-27T00:01:29 Frontier: bss03 was last seen in #aichallenge 1 week, 2 days, 1 hour, 39 minutes, and 37 seconds ago: amstan: My email is everywhere; I don't depend on it being secret to keep me safe from SPAM. 2011-03-27T00:02:27 antimatroid: yeah, I just came across the same case in my testing 2011-03-27T00:03:43 *** smippy has joined #aichallenge 2011-03-27T00:03:44 *** smippy has joined #aichallenge 2011-03-27T00:04:29 *** smellyhippy has quit IRC (Ping timeout: 260 seconds) 2011-03-27T00:04:37 very cool 2011-03-27T00:05:32 is there a doc describing what the conflict resolution rules are? 2011-03-27T00:05:35 spacebat: do check the specifications page https://github.com/aichallenge/aichallenge/wiki/Ants-Game-Specification 2011-03-27T00:05:38 thanks 2011-03-27T00:05:44 there is a link the the proposed options on that page 2011-03-27T00:05:55 and please do give us your feedback :) 2011-03-27T00:06:04 will do 2011-03-27T00:06:41 I got into planet wars halfway through the competition 2011-03-27T00:06:47 still play galcon occasionally 2011-03-27T00:06:53 yeah same, it's a great game 2011-03-27T00:06:58 i got into pw right at the start though 2011-03-27T00:07:46 my coworkers were using the perl starter pack and some of them timed out with too many fleets in flight 2011-03-27T00:07:57 spacebat: what language do you use? 2011-03-27T00:08:08 but I used it as an excuse to learn CLOS better in common lisp 2011-03-27T00:08:21 :) 2011-03-27T00:08:50 there was no worry about timing out at least with my puny tactics 2011-03-27T00:09:09 hopefully there will be a more recent version of SBCL this time around 2011-03-27T00:09:40 i don't know much (anything) about that kind of stuff, but people have been discussing updating things like that 2011-03-27T00:09:46 *** Chris_0076 has quit IRC (Ping timeout: 246 seconds) 2011-03-27T00:09:56 spacebat: apparently last contests perl start pack was not very well written :P 2011-03-27T00:10:14 yeah I got that impression :) 2011-03-27T00:10:16 they could always write their own :P 2011-03-27T00:10:36 *** Stocha has joined #aichallenge 2011-03-27T00:11:50 is ai-contest.com down? 2011-03-27T00:13:41 no route to hose 2011-03-27T00:13:43 host 2011-03-27T00:13:44 :) 2011-03-27T00:14:31 it puts the ants on its body or it gets the hose again 2011-03-27T00:14:35 it seems 2011-03-27T00:15:22 bah, who read the forums anyway :p 2011-03-27T00:15:31 *** suicidolt has quit IRC (Quit: ChatZilla 0.9.86-rdmsoft [XULRunner 1.9.2.16/20110323145543]) 2011-03-27T00:16:17 @seen stocha 2011-03-27T00:16:17 Stocha: stocha was last seen in #aichallenge 54 seconds ago: bah, who read the forums anyway :p 2011-03-27T00:17:26 antimatroid: ok, you can fix it by doing the adjacent cells first, then diagonal cells :P 2011-03-27T00:18:01 it's slower, but it's only like 10% slower instead of 2000% slower :P 2011-03-27T00:18:25 were diagonal squares in the AIM squares? 2011-03-27T00:18:32 no 2011-03-27T00:19:49 antimatroid: http://pastebin.com/j52RzXfw 2011-03-27T00:20:07 I'm not proposing we use this, just trying to demonstrate that it can be fixed :P 2011-03-27T00:20:49 umm 2011-03-27T00:20:53 why is ai-contest.com down? 2011-03-27T00:21:30 janzert: ping! any ideas why ai-contest.com is down? 2011-03-27T00:22:30 who's is in charge of ai-contest.com ? 2011-03-27T00:22:33 no idea 2011-03-27T00:22:36 amstan :P 2011-03-27T00:22:42 Stocha: and j3camero 2011-03-27T00:22:42 not really 2011-03-27T00:22:51 sigh: http://pastebin.com/QCsFFiY4 2011-03-27T00:22:52 he is the one asking why it is down :p 2011-03-27T00:23:11 Stocha: i only have root on it... but i can't even ping it 2011-03-27T00:23:14 ssh down too? 2011-03-27T00:23:21 so.. it must be the hardware, i don't have access to that 2011-03-27T00:23:23 ahh 2011-03-27T00:23:32 contestbot: ping ai-contest.com 2011-03-27T00:23:32 pong 2011-03-27T00:23:38 How is the engine going. I saw a lot of talk about speed and optimization and so on. That i'm not sure i understood anything out of. 2011-03-27T00:23:51 or datacenter network 2011-03-27T00:23:53 our get_vision is broken 2011-03-27T00:23:54 Stocha: ask Mcleopold, he's in charge of that 2011-03-27T00:23:57 we're trying to work out how to fix it 2011-03-27T00:24:10 what does get_vision do ? 2011-03-27T00:24:18 janzert: he's hard rebooting it 2011-03-27T00:24:31 simultaneously paints out from a bots ants to find "visible" squares 2011-03-27T00:24:32 ok 2011-03-27T00:24:37 Dev.Opera suggests using "new String('text');" instead of "'text'" when you want to call methods on the string, as the raw type will have to be converted to an object for every single call otherwise. I really missed that when I wrote the parser. 2011-03-27T00:24:45 do you do that bitmap based ? 2011-03-27T00:24:51 or integer based ? 2011-03-27T00:24:57 ? 2011-03-27T00:25:07 I think i'm doing the same. But i'm not sure 2011-03-27T00:25:21 i didn't cared optimizing it, because it was already quite fast. 2011-03-27T00:25:36 so i'm not sure i understand what's doing on in there :p 2011-03-27T00:25:37 the naive way of doing that efficiently breaks with euclidean metric 2011-03-27T00:25:48 what do you mean ? 2011-03-27T00:25:54 what's the naive way ? 2011-03-27T00:25:58 http://pastebin.com/gscdvw3u 2011-03-27T00:26:01 you mean x2+y2=c ? 2011-03-27T00:26:38 Stocha: no, he means the algorithm that used to be in the engine, the breadth-first search one 2011-03-27T00:27:01 @later tell McLeopold here is a simple demonstration of vision breaking: http://pastebin.com/gscdvw3u 2011-03-27T00:27:01 sigh: I come to serve. 2011-03-27T00:27:41 antimatroid: I can't see how that example would fail? 2011-03-27T00:27:50 antimatroid: http://pastebin.com/QCsFFiY4 <--- this example 2011-03-27T00:28:37 i'm far from sure i understand what the problem is. 2011-03-27T00:28:53 I though a painting view would be aimed at known what an ant can see. 2011-03-27T00:29:01 Or what a team of ant can see. 2011-03-27T00:29:09 I see no way of getting wrong there. 2011-03-27T00:29:22 Or i totally missed a subtlety 2011-03-27T00:29:34 Or i missed what this is about entirely :p 2011-03-27T00:29:36 @later tell McLeopold the starter packs should have the simple by_ants method, we KNOW it works and it is not much slower. It is also much harder to patch starter packs that people have downloaded if there is a bug 2011-03-27T00:29:36 sigh: OK 2011-03-27T00:29:53 Stocha: both ants are on the same team 2011-03-27T00:29:56 in that example 2011-03-27T00:30:22 the numbers are just to differentiate how each square is reached by the algorithm 2011-03-27T00:30:42 do you mean that you have a function wich for a team_of_ant returns a set of seen squares; And that it doesn't work ? 2011-03-27T00:31:02 yes 2011-03-27T00:31:27 can you link me to the method, so i can have a look at the code ? :p 2011-03-27T00:31:43 we also have algorithms that do work :) 2011-03-27T00:31:51 sigh: looking at your latest now 2011-03-27T00:32:01 why is the one that doesn't work important then ? 2011-03-27T00:32:03 trying to wrap my head around what it's doing anyway :) 2011-03-27T00:32:06 janzert: don't bother, it doesn't work either 2011-03-27T00:32:09 oh 2011-03-27T00:32:27 I do have a straight algorithm that is working. 2011-03-27T00:32:34 there are some pathological cases where the current one that works is slow 2011-03-27T00:32:38 janzert: and interestingly it breaks in a way your tests don't catch 2011-03-27T00:32:39 Although i don't know if it would go well with python :p 2011-03-27T00:32:57 heh, really it takes more than two ants to break it? 2011-03-27T00:33:11 a.x 2011-03-27T00:33:13 .b. 2011-03-27T00:33:14 ..a 2011-03-27T00:33:27 vr2 = 2 2011-03-27T00:33:44 ahh, I wondered earlier about that case when I couldn't get the other one to break with vr2=6 2011-03-27T00:33:50 b can see x but is blocked because the a's reach the adjacent squares first 2011-03-27T00:33:56 *** fmeyer has quit IRC (Ping timeout: 264 seconds) 2011-03-27T00:34:42 what i use i think is : for(int x=0;x for(int y=0;y essentially my method sets up a priority queue ordered by euclidean distance (instead of manhattan distance)... and this example shows that still fails 2011-03-27T00:35:10 paint if x2+y2<=sqaure2 2011-03-27T00:35:25 Stocha: yes, that works, that's what is in the engine now 2011-03-27T00:35:38 what's wrong about it ? 2011-03-27T00:35:48 nothing, we thought we had a faster method 2011-03-27T00:35:50 you can optimize it a lot also 2011-03-27T00:35:59 although i had no reason to do so yet 2011-03-27T00:36:17 it's pretty much the fastest method i know. 2011-03-27T00:36:33 Stocha: it becomes slow when a lot of ants are packed together 2011-03-27T00:36:36 because you can go with clever tricks about incrementaly computing x2 and y2 2011-03-27T00:36:41 it takes O((viewradius*2)**2 * num_ants) time 2011-03-27T00:36:46 because you check the same square multiple times 2011-03-27T00:36:49 ah 2011-03-27T00:36:49 of course the constant is very low 2011-03-27T00:36:56 i see 2011-03-27T00:37:11 Of course i used 64 bits integer to get it go faster with many ants. 2011-03-27T00:37:46 Although i got slower the larger the map is. 2011-03-27T00:37:57 I still think it should be possible to do O(number_of_squares_in_view) but the question is if the constant overhead can be low enough 2011-03-27T00:38:05 How many ants can you support at the moment ? 2011-03-27T00:38:30 janzert: yes, I have a candidate, but it is more complicated 2011-03-27T00:38:49 great :) 2011-03-27T00:38:57 why are you doing optimization when the engine has not been finished yet ? 2011-03-27T00:39:19 the engine itself is basically finished 2011-03-27T00:39:23 although maybe it's just fun. 2011-03-27T00:39:28 all the supporting stuff isn't 2011-03-27T00:39:31 it's just interesting 2011-03-27T00:39:36 okay 2011-03-27T00:39:36 and yeah, it's an interesting thing to look at 2011-03-27T00:39:48 I thnk the current method in the engine is fine 2011-03-27T00:39:50 janzert: http://pastebin.com/j52RzXfw <-- forgive the duplicated code 2011-03-27T00:40:03 So want something linear into the number of seen squares ? 2011-03-27T00:40:04 antimatroid pasted a case, but I can't see how that breaks this 2011-03-27T00:40:14 sigh: sorry trying to catch up on the convo 2011-03-27T00:40:17 got caught on the phone 2011-03-27T00:40:24 antimatroid: no worries 2011-03-27T00:41:04 what did you try ? 2011-03-27T00:42:33 sigh: ignore my example, i think your new method works 2011-03-27T00:42:54 I think it works too, but I've been wrong a few times now :) 2011-03-27T00:43:07 yes, i think it works for everything that's broken stuff so far :) 2011-03-27T00:43:28 hehe 2011-03-27T00:43:39 man, should start making unit tests for this 2011-03-27T00:45:32 what testing strategy would you use ? 2011-03-27T00:45:44 and it takes half the time of by_ant for a 10x10 ant grid with a spacing of 1 2011-03-27T00:45:45 nice 2011-03-27T00:45:58 I'm not sure i underestood the idea behind the algo yet 2011-03-27T00:47:30 i guess i do not understand python wel :p 2011-03-27T00:47:34 Stocha: search squares outwards from all ants simultaneously using a priority queue where squares are weighted by Euclidean distance 2011-03-27T00:48:09 this is not the most pedagogical example to look at, it's the end result of me hacking on fixes till it worked :P 2011-03-27T00:48:14 sigh: you can just use a queue i'm pretty sure 2011-03-27T00:48:37 priority queue is more correct 2011-03-27T00:48:58 do they have much difference in speed? 2011-03-27T00:49:34 what is a square weighted by euclidiant distance ? means squares are given a value equal to distance from closest ant ? 2011-03-27T00:49:44 yeah 2011-03-27T00:49:56 oh I thought you were just taking issue with the name. priority queue is the name of an abstract data structure, it says nothing about implementation 2011-03-27T00:50:15 the actual code uses stacks at the core 2011-03-27T00:50:30 but overall it implements a priority queue 2011-03-27T00:50:32 What's the idea for weighting the squares then ? 2011-03-27T00:50:46 it doesn't sound easy to do, that weighting 2011-03-27T00:50:59 so you don't repeatedly check the same visibly square more than once 2011-03-27T00:51:17 how do you obtain the weight matrix ? 2011-03-27T00:51:27 or just the weight i mean. 2011-03-27T00:51:35 you build it as you search 2011-03-27T00:51:52 i don't get the linearity of it yet. 2011-03-27T00:51:58 Stocha: when you reach a square you are guaranteed that it was reached the fastest way 2011-03-27T00:52:14 sigh: the horizontal directions loop and diagonal directions loop are identical right? 2011-03-27T00:52:20 the problem with the naive way was it didn't guarantee you reached it the fastest way 2011-03-27T00:52:21 just double checking before I combine it 2011-03-27T00:52:26 janzert: yes, they are 2011-03-27T00:52:55 janzert: they can't be done at the same time though 2011-03-27T00:53:07 yeah, just found that out :P 2011-03-27T00:53:10 :D 2011-03-27T00:53:34 So you point outward from every-ant, and put a weight to each squared reach then. 2011-03-27T00:54:31 Stocha: what language are you most familiar with? 2011-03-27T00:54:41 java i guess 2011-03-27T00:55:05 or anything using int[] and while loops :p 2011-03-27T00:55:25 hmm... antimatroid you have a c++ impl of vision? 2011-03-27T00:55:52 i have always been bad at reading code also i guess :p 2011-03-27T00:55:53 not the new one 2011-03-27T00:55:58 i have an old one somewhere 2011-03-27T00:56:19 So we still might discuss the raw algorithms anyway 2011-03-27T00:56:39 actually, have a look at this: https://github.com/aichallenge/aichallenge/blob/epsilon/ants/dist/starter_bots/python/ants.py#L198 2011-03-27T00:56:43 i don't see where there is a cutoff 2011-03-27T00:56:57 this is the old broken one, but it is well commented and the idea is the same 2011-03-27T00:57:19 Stocha: http://pastebin.com/VmzKyGhv 2011-03-27T00:57:41 and there's a c++ version :) 2011-03-27T00:59:22 i don't think there exists a screen wide enough for me :( 2011-03-27T00:59:47 :P 2011-03-27T01:00:47 imagine going back to <20" crts :\ 2011-03-27T01:01:29 so at likely common game ant spacing by_ant is still the fastest correct method :/ 2011-03-27T01:01:29 I'm working on my 15" laptop just fine :P 2011-03-27T01:01:36 janzert: yes 2011-03-27T01:02:11 we could always use a different metric 2011-03-27T01:02:49 i used to have MacPlus when i first learned turbopascal and c. 2011-03-27T01:02:50 janzert: if you want to get really hacky, you can update vision can be a count instead of a boolean, and you update it as a ant moves or dies/spawns 2011-03-27T01:02:55 the screen was not that big. 2011-03-27T01:03:23 I didn't grab the principle of the algorithm 2011-03-27T01:03:28 you're meaning incrementally update it? 2011-03-27T01:03:34 is there an article somewhere describing the principle ? 2011-03-27T01:03:34 janzert: when an ant moves you only need to update the difference of the previous and new vision 2011-03-27T01:03:37 yeah, that would be an option 2011-03-27T01:03:37 janzert: yes 2011-03-27T01:03:50 but then the vision code leaks out into the entire engine 2011-03-27T01:03:57 ah so it's based over ant movement ? 2011-03-27T01:04:11 between turns. 2011-03-27T01:04:13 and although it spreads the cost out I'm not sure it would actually be faster? 2011-03-27T01:04:17 Stocha: my tron bot was based around it 2011-03-27T01:04:22 janzert: why not? 2011-03-27T01:04:30 I'm assuming most all ants will move every turn 2011-03-27T01:04:42 there was an euclidian distance in tron ? 2011-03-27T01:04:45 you still have to do the same work per ant right? 2011-03-27T01:04:46 no 2011-03-27T01:04:46 kill/spawn ant is O(visionradius**2), moving an ant is O(visionradius) 2011-03-27T01:04:59 this works for movement as well, you just block your search with water 2011-03-27T01:05:42 the evauluation function for my tron bot for non wins/loses was basically to do the difference between the number of squares each player could reach first 2011-03-27T01:06:31 you don't want to see the code for that, i'm not even sure i still have it, but it was all in one file and just awful 2011-03-27T01:06:56 i'm more interested into grabbing the principle. 2011-03-27T01:07:30 hang on 2011-03-27T01:07:32 in any case incremental updating seems like it would get messy 2011-03-27T01:07:38 So it's like, for each ant you maintain a list of seen squares. 2011-03-27T01:07:55 then you try to update this list when the ant moves. 2011-03-27T01:08:30 Stocha: if you were to go with an incrementally update version it would be something like that yes 2011-03-27T01:08:39 but we're not really looking at that 2011-03-27T01:08:40 the obvious way being applying a translation to it 2011-03-27T01:08:49 but a simple translation wouldn't speed things 2011-03-27T01:09:01 that you have to build a clever way to do the translation ? 2011-03-27T01:09:25 so the question would be : what is the clever way of doint that translation then ? :p 2011-03-27T01:09:50 Stocha: http://pastebin.com/mRfSG8hi 2011-03-27T01:09:50 or maybe you don't update the points in the list 2011-03-27T01:09:55 janzert: yes, which is why I said that it was hacky :P 2011-03-27T01:09:59 maybe you update the list itself 2011-03-27T01:10:19 the numbers are squares that will go onto the queue before any number greater than it, but not necessarily in any order for the same number 2011-03-27T01:10:40 if you paint a grid like that with all ants, you can use the numbers from a "target square" to backtrack a shortest path as well 2011-03-27T01:11:13 I'm thinking it would be best to have a 2 dimensional array for the points, that in each dimension is like a circular buffer 2011-03-27T01:11:17 janzert: if vr2=96, then vision is 293 squares per ant, but symmetric difference between adjacent visions is only 38 2011-03-27T01:11:34 if you want concrete numbers :) 2011-03-27T01:11:36 so you want to build the distance to closest ant matrix really. 2011-03-27T01:11:37 instead of moving cells within the grid, you change the start/end coordinats 2011-03-27T01:11:57 and update the row and column that is old with new data 2011-03-27T01:13:11 i'm totally lost :p 2011-03-27T01:13:29 so are the ants 2011-03-27T01:13:34 :) 2011-03-27T01:14:08 don't worry half this conversation is just exploring what is possible, rather than what is practical :P 2011-03-27T01:14:11 But i least i understood that we are trying to incrementaly update the mindistance to ant matrix, as ants moves. 2011-03-27T01:14:38 wich is a good start i think. 2011-03-27T01:14:53 i even could use a matrix of this sort at some point, who knows. 2011-03-27T01:16:09 the easy way would be to recalculate the whole thing every turn of course. from scratch 2011-03-27T01:16:27 i don't even know how long that takes. 2011-03-27T01:16:51 spacebat: have you read the battle options and proposed scoring method yet? :P 2011-03-27T01:17:02 that's what i'm most interested in peoples opinion on 2011-03-27T01:17:03 not yet 2011-03-27T01:17:05 so maybe i should understand how to obtain the distance frome closest ant matrix, from scratch in the first place :p 2011-03-27T01:18:08 i could go with : for each ant X 2011-03-27T01:18:29 calculate the whole distance matrix for that ant 2011-03-27T01:18:41 apply the min of all matrix for each squares 2011-03-27T01:19:10 maybe i'll think about it. 2011-03-27T01:19:44 when will the forum be up ? 2011-03-27T01:19:56 who reads the forums? :P 2011-03-27T01:20:23 perhaps anyone who reads the forums should be disqualified from the competition 2011-03-27T01:21:33 i do. Although i read them incrementally. Wich is easy to do when noone is ever posting anything :p 2011-03-27T01:21:49 why would that be spacebat ? 2011-03-27T01:21:54 *** moongrass has quit IRC (Ping timeout: 250 seconds) 2011-03-27T01:22:01 *** moongrass has joined #aichallenge 2011-03-27T01:22:23 Stocha: for shits and giggles 2011-03-27T01:22:33 :) 2011-03-27T01:22:37 ? 2011-03-27T01:22:49 its a joke 2011-03-27T01:22:49 giggles is like laughing right ? 2011-03-27T01:22:56 it is 2011-03-27T01:22:57 but what does that expression stands for 2011-03-27T01:23:15 it means that disqualifying the forum readers was a joke ? 2011-03-27T01:23:25 giggling - little school girls in a group giggling 2011-03-27T01:23:34 yeah i though it was that 2011-03-27T01:23:45 so it's a kind of laughing. 2011-03-27T01:23:57 shits and giggles is a common english expression for taking the piss kind of 2011-03-27T01:24:12 taking the piss ?? 2011-03-27T01:24:13 antimatroid: I prefer the damage option 2011-03-27T01:24:16 what's that. 2011-03-27T01:24:21 just out of intuition 2011-03-27T01:24:21 spacebat: correct answer :) 2011-03-27T01:24:38 damage is nice. But there are parameters to choose. 2011-03-27T01:24:45 there aren't 2011-03-27T01:24:47 1 is fine 2011-03-27T01:25:10 it doesn't make sense to have say 2 ants kill 3 in a single battle 2011-03-27T01:25:19 i dont know what you are talking about, but i like dealing damage to ants in step 1 and checking which and is full off in step 2 2011-03-27T01:25:26 *ant 2011-03-27T01:25:40 Frontier: Stocha wants to vary the amount of damage each ant has to dish out 2011-03-27T01:25:45 geez, my spelling adopts to the internet Oo 2011-03-27T01:25:53 Frontier: also, your vision algo is going to be broken too, sorry :P 2011-03-27T01:26:03 if two ants kill 3 then you have a reason for being carefull about ant positionning. Especially if 3 can kill 2 too 2011-03-27T01:26:32 but the point of damage is that you get more ants in to get more kills 2011-03-27T01:26:33 antimatroid: my vision algorithm? it is brute force and simple, how can it be broken? 2011-03-27T01:26:51 adding a counter balance to that would really be complicating things 2011-03-27T01:26:54 i though the point was that you had to plan for ant position. 2011-03-27T01:27:05 Frontier: you didn't use my vision algorithm? then you're ok 2011-03-27T01:27:08 my one was broken 2011-03-27T01:27:38 should i include one in my starter bot? 2011-03-27T01:27:48 i'm referring to your visualiser? 2011-03-27T01:27:50 Stocha: all of the resolutions have implcations for ant placement 2011-03-27T01:27:52 one vision algorithm ? 2011-03-27T01:27:57 i think you should yes 2011-03-27T01:28:02 it's pretty usefull. 2011-03-27T01:28:14 antimatroid: ok, my visualizer is fine. 2011-03-27T01:28:22 otherwise you don't know wich squares are solid ground. 2011-03-27T01:28:30 i mean: I SEE the results, that is a benefit 2011-03-27T01:28:34 wich is kinda embarassing 2011-03-27T01:28:45 Frontier: http://pastebin.com/gscdvw3u 2011-03-27T01:28:55 that outlines a case where the function we had in the engine didn't work quite right 2011-03-27T01:28:58 sigh made that 2011-03-27T01:29:03 i wondered if the engine should not send first seen ground squares also. As it does for water 2011-03-27T01:29:52 why? 2011-03-27T01:30:03 people can work out what's unseen for themselves 2011-03-27T01:30:13 why don't you stick with straight forward code into the engine ? until you have it fully tested and used. Then you can make some optimization with an heavy regression test policy (cross checking the result of the well testing simple engine with that of the new) 2011-03-27T01:30:33 Maybe simple bots won't want to do that. 2011-03-27T01:30:58 ants shouldn't know the map, that seems wrong 2011-03-27T01:31:00 The starter bot probably have to provide a mean to at least know what are the solid ground squares :p 2011-03-27T01:31:04 use unit tests as my granny used to say 2011-03-27T01:31:35 Just program, motherfucker! is what my granny used to say 2011-03-27T01:31:41 i'm not a unit tests fanatic; They are usefull. But tricky to build. And trikier to maintain :p 2011-03-27T01:32:17 My granny used to say : You jumped on that biscuit like misery on the poor wolrd. 2011-03-27T01:32:37 nothing to do with programming. She wasn't what you would call a geek or hacker. 2011-03-27T01:32:51 she didn't have an iphone to say the truth. 2011-03-27T01:33:07 i don't think either of my grandmothers knew/know what the hell programming is let alone a unit test 2011-03-27T01:33:25 they could knit, couldn't they? 2011-03-27T01:33:32 ha :P 2011-03-27T01:33:36 what's knitting ? 2011-03-27T01:33:39 touche 2011-03-27T01:33:52 i search google for incremental ways of updating distance matrixes 2011-03-27T01:33:53 Stocha: programming 2011-03-27T01:34:02 and found nothing. I never find anything on google :'( 2011-03-27T01:34:06 well actually, it's running a program :P 2011-03-27T01:34:33 except when i query for britney spears. 2011-03-27T01:34:42 Stocha: it's more fun to solve problems yourself 2011-03-27T01:34:59 sometimes it is. But i 'm growing tired of that 2011-03-27T01:35:19 aichallenge: sigh epsilon * r62fee81 / ants/util/vision_compare.py : Added 2 more vision methods - http://bit.ly/fWw64M 2011-03-27T01:35:21 janzert: you there? 2011-03-27T01:35:31 janzert: check out vision_by_ants_faster ^^ :D 2011-03-27T01:35:38 although i always have to. Because most of the time i don't understand articles until i'm able to recreate the solution. 2011-03-27T01:36:11 Still you can find nice ideas in existing algorithms. Sometimes. 2011-03-27T01:37:16 when will the forum be up again ? 2011-03-27T01:38:39 antimatroid: you too, try running the vision_compare script :P 2011-03-27T01:38:45 My new bot is significantly stronger than the reference one on 90x90 maps. But weaker on 200x200 maps. how is that : / 2011-03-27T01:38:57 I dont get it. 2011-03-27T01:39:14 when i looked at the 200x200 games, the new bot really look stronger. 2011-03-27T01:39:39 but on 200x200 map i got Score inWin for P0 -> 304 Score inWin for P1 -> 464 2011-03-27T01:39:54 with P0 being the new bot (which i believe is the strongest one) 2011-03-27T01:40:34 At least it supports my advocating for a fixed map size for the contest :p 2011-03-27T01:41:14 People will not want to test their bot for every possible map size, not knowing what the setting will be. 2011-03-27T01:42:57 the squaredistance=6 is the standard for fight, with Damage(x=1) being the prefered candidate so far ? 2011-03-27T01:43:58 http://ns368394.ovh.net/stocha/Sz80_Fh02G_ExplorAggr_rad6_x1_00/ 2011-03-27T01:44:19 here is a game with those settings. 2011-03-27T01:47:41 I guess i should start diggin a bit into that X=1 thing. 2011-03-27T01:48:01 and that everybody has left the irc channel. 2011-03-27T01:49:38 The two bots are of same strength with those settings. 2011-03-27T01:50:01 *** Stocha has quit IRC (Quit: Page closed) 2011-03-27T01:53:55 *** closedbracket has joined #aichallenge 2011-03-27T02:00:03 *** fmeyer has joined #aichallenge 2011-03-27T02:23:05 everyone's gone :( 2011-03-27T02:28:49 janzert? antimatroid? 2011-03-27T02:29:06 sigh:? 2011-03-27T02:29:07 contestbot? 2011-03-27T02:29:30 antimatroid: I've been cleaning up the vision methods 2011-03-27T02:29:39 *** Stocha has joined #aichallenge 2011-03-27T02:29:55 turns out that most of the time is taken up by the function calls 2011-03-27T02:30:05 haha typical 2011-03-27T02:30:14 you can't inline that with python can you? 2011-03-27T02:30:23 would need to be compiled 2011-03-27T02:30:48 aichallenge: sigh epsilon * r9306fc5 / ants/util/vision_compare.py : Refactored some of the vision candidates - http://bit.ly/egSESb 2011-03-27T02:30:49 aichallenge: sigh epsilon * rab18586 / ants/util/vision_compare.py : Refactored vision_by_distance_2* methods - http://bit.ly/hOgoAC 2011-03-27T02:30:51 yeah, you can't 2011-03-27T02:30:59 don't call functions then :p 2011-03-27T02:31:45