2011-03-24T00:16:25 *** Accoun has quit IRC () 2011-03-24T00:31:16 *** Accoun has joined #aichallenge 2011-03-24T00:35:35 *** Ttech has quit IRC (Read error: Operation timed out) 2011-03-24T00:35:49 *** janzert has quit IRC (Read error: Connection reset by peer) 2011-03-24T00:37:35 *** janzert has joined #aichallenge 2011-03-24T00:41:58 *** Ttech has joined #aichallenge 2011-03-24T01:44:51 The version of Rhino that ships with the JRE, runs in interpreted mode. Parsing a replay took >20 seconds. After optimizing a subroutine it went down to 14 seconds. The separate release from Mozilla contains a JavaScript-to-Java compiler, that cannot compile to 100% clean Java because of the dynamic nature of JS, but does a fairly good job! The parsing takes only 3 seconds now. This is faster than Opera JS engine with 5 sec. 2011-03-24T01:48:32 So *technically* it works, but this compiler is a big blackbox that takes js files, adds a main method and compiles to a class with the same name. It also needs attributes and methods of objects defined before first use. I'm not sure yet if it is flexible enough. 2011-03-24T01:53:36 *** Anders__ has joined #aichallenge 2011-03-24T02:09:22 *** janzert has quit IRC (Read error: Connection reset by peer) 2011-03-24T02:11:08 *** janzert has joined #aichallenge 2011-03-24T02:19:35 ah, my fault. it doesn't need stuff defined before use; at least not more than any other implementation. 2011-03-24T02:20:25 *** delt0r___ has joined #aichallenge 2011-03-24T02:21:55 *** delt0r_ has quit IRC (Ping timeout: 264 seconds) 2011-03-24T02:35:09 *** janzert has quit IRC (Read error: Connection reset by peer) 2011-03-24T02:36:30 *** janzert has joined #aichallenge 2011-03-24T02:36:35 *** antimatroid has quit IRC (Ping timeout: 246 seconds) 2011-03-24T02:39:08 *** sigh has quit IRC (Ping timeout: 252 seconds) 2011-03-24T02:44:35 *** andy_ has joined #aichallenge 2011-03-24T02:45:57 *** amstan has quit IRC (Ping timeout: 250 seconds) 2011-03-24T03:21:36 *** Cyndre has quit IRC (Quit: Leaving) 2011-03-24T03:49:24 *** antimatroid has joined #aichallenge 2011-03-24T03:52:18 *** sigh has joined #aichallenge 2011-03-24T03:54:34 *** boegel has joined #aichallenge 2011-03-24T03:57:28 *** boegel has quit IRC (Client Quit) 2011-03-24T04:09:30 *** num1 is now known as num1_ 2011-03-24T04:13:27 *** Eruonen has joined #aichallenge 2011-03-24T04:16:30 *** boegel has joined #aichallenge 2011-03-24T04:27:37 *** espes has quit IRC (Ping timeout: 246 seconds) 2011-03-24T04:28:35 *** iFire has quit IRC (Ping timeout: 246 seconds) 2011-03-24T04:29:17 *** iFire has joined #aichallenge 2011-03-24T04:37:16 *** aerique has joined #aichallenge 2011-03-24T05:00:42 *** espes has joined #aichallenge 2011-03-24T05:00:58 McLeopold: you there? 2011-03-24T05:07:50 janzert: you there? 2011-03-24T05:08:23 i think i found a neat little solution to an issue i had with game trees 2011-03-24T05:15:14 sigh: you will do 2011-03-24T05:15:28 you know how in a game tree you don't link two nodes to the same child? 2011-03-24T05:15:45 what will i do now? 2011-03-24T05:15:58 do instead of janzert 2011-03-24T05:16:38 what do you mean, don't link two nodes to the same child? 2011-03-24T05:16:46 you mean, create a DAG? 2011-03-24T05:16:52 hang on, i'll do a paint drawing 2011-03-24T05:20:03 http://i.imgur.com/8X45j.png 2011-03-24T05:20:23 the first is the main way to do it, but if it lead to the same "state", then you're really calculating the same stuff twice 2011-03-24T05:20:38 you can still perform backward induction on the second tree somewhat like you would the first and always get the same result 2011-03-24T05:21:06 yup, so you have a DAG instead of a tree 2011-03-24T05:21:27 directed acyclic graph? 2011-03-24T05:21:34 yes 2011-03-24T05:21:39 yep 2011-03-24T05:21:55 you need to make sure a node doesn't preceed yours before adding the link for that 2011-03-24T05:21:56 in games like chess you use a transposition table to detect repeated positions 2011-03-24T05:22:00 otherwise it wouldn't be acyclic 2011-03-24T05:22:09 sure 2011-03-24T05:22:28 alpha-beta becomes a tad more complex too, but not much 2011-03-24T05:22:37 although you could do such a game and it'd be like a game on a graph with terminal nodes indicating the end of the game 2011-03-24T05:23:18 depends what the rules are on the game if there is a repeating board position 2011-03-24T05:23:27 *** computer1iz_222 has quit IRC (Ping timeout: 252 seconds) 2011-03-24T05:23:34 yeah 2011-03-24T05:23:45 *** computerwiz_222 has joined #aichallenge 2011-03-24T05:23:52 but if i was going an implementation generally, i'd want to make sure 2011-03-24T05:23:56 but what is your question? 2011-03-24T05:24:08 if it's acyclic then you get a partial order from a preceeds relation on nodes 2011-03-24T05:24:12 i have no question 2011-03-24T05:24:16 i just thought it was neat :P 2011-03-24T05:24:27 lol 2011-03-24T05:24:54 :) 2011-03-24T05:25:34 antimatroid: just be aware that you can't apply alpha-beta naively, because you won't have the same window when coming form two different paths 2011-03-24T05:26:04 sigh: i don't use alpha beta 2011-03-24T05:26:11 it doesn't work for the way i do game trees in this contest 2011-03-24T05:26:23 ok, then you don't need to worry :) 2011-03-24T05:26:31 i shall play around with it anyway though 2011-03-24T05:26:40 if i do a library, it'd be nice to include such functions 2011-03-24T05:27:14 i could use alpha beta for my evaluation at the end of my turn actually, i should do that 2011-03-24T05:27:25 ideally one would like to use it while making the tree, but that just doesn't work 2011-03-24T05:27:42 heh 2011-03-24T05:28:01 *** iFire has quit IRC (Read error: Connection reset by peer) 2011-03-24T05:28:21 I'd be surprised if transposition tables helped in this game 2011-03-24T05:28:22 *** iFire has joined #aichallenge 2011-03-24T05:29:54 It looks like its going to be a challenging game to write an ai for 2011-03-24T05:30:25 but I've never written an ai before 2011-03-24T05:30:56 well, except for hunting the wumpus 2011-03-24T05:32:50 phirenz: i expect some pretty good bots in this 2011-03-24T05:32:57 and a tight contest in the top 100 2011-03-24T05:33:29 Yeah 2011-03-24T05:33:56 it's a much more penetrable game than planet wars was i think 2011-03-24T05:34:10 Right now I haven't decided if I should use python or c 2011-03-24T05:34:32 i suspect you may want the speed boost for this one 2011-03-24T05:34:36 it didn't matter so much with planet wars 2011-03-24T05:34:52 but that's just my opinion, and a very biased one on that :P 2011-03-24T05:34:56 Yeah, I'm getting the same feeling 2011-03-24T05:35:14 why not c++? 2011-03-24T05:35:27 i wrote the c++ starter bot, although i need to clean it a bit 2011-03-24T05:36:08 sigh: it's not that much harder to do alpha beta (sorry didn't think about it) 2011-03-24T05:36:18 *** iFire has quit IRC (Read error: Connection reset by peer) 2011-03-24T05:36:20 just store the number of parents to a child 2011-03-24T05:36:29 or links 2011-03-24T05:36:32 *** iFire has joined #aichallenge 2011-03-24T05:37:01 antimatroid: not that simple 2011-03-24T05:37:09 yeah it is 2011-03-24T05:37:13 because the search window is different 2011-03-24T05:37:26 so you can't just use the eval that you got from a different branch 2011-03-24T05:37:50 you have to store if you got the exact eval, or if you hit the alpha or beta cutoff 2011-03-24T05:37:53 ahh yeah 2011-03-24T05:38:04 I like c better as a language 2011-03-24T05:38:05 wait nah 2011-03-24T05:38:08 it works fine 2011-03-24T05:38:12 if you don't have the exact, then you stil need to search a bit mow 2011-03-24T05:38:13 the tree below is the same tree 2011-03-24T05:38:36 same tree, but different search window 2011-03-24T05:38:51 if you mark a child to be deleted, when deleting if the number of parents goes to zero, delete that section of the tree 2011-03-24T05:39:13 otherwise just remove the child link 2011-03-24T05:39:14 ok, I don't know what deleting has got to do with it 2011-03-24T05:39:31 well, you could just skip over it 2011-03-24T05:39:44 no, you can't 2011-03-24T05:39:48 in a game like tron, you wanted to recycle your tree across turns 2011-03-24T05:39:59 just because you searched it, doesn't mean you got the exact eval out of it 2011-03-24T05:40:14 because you might have terminated early by hitting an alpha or beta cutoff 2011-03-24T05:40:18 if you have a tree that's not going to change for the decision you make 2011-03-24T05:40:26 yes, it will 2011-03-24T05:40:30 :P 2011-03-24T05:40:48 no, you reach the end of your turn, so you stop generating and evaluate your tree 2011-03-24T05:40:53 because if you use alpha-beta you did NOT search the entire subtree 2011-03-24T05:41:06 no, but you will search it later if necessary 2011-03-24T05:41:15 sure, but then you can't delete 2011-03-24T05:41:27 you don't need to 2011-03-24T05:41:34 you just don't go searching down there now 2011-03-24T05:42:16 you're still avoiding searching sections of the tree that are not going to change your decision 2011-03-24T05:43:55 the only parts of the tree you can avoid searching down are the ones where you got an exact eval out of, or the alpha-beta window was tighter than the current one (with some qualifications) 2011-03-24T05:44:33 yes, but if you have a large tree, that might be lots 2011-03-24T05:45:15 sure, but my point is that you still need to remember what the previous window was 2011-03-24T05:45:33 techinically, only one side of the window, the side that caused the early termination 2011-03-24T05:45:59 if you don't remember that, then you can't prune at all 2011-03-24T05:46:12 i don't need to prune the tre 2011-03-24T05:46:27 prune/skip/whatever 2011-03-24T05:46:57 take that example image i gave you, but suppose player 2's second node linked to the middle node 2011-03-24T05:47:10 ok 2011-03-24T05:47:13 it might need to be bigger :\ 2011-03-24T05:47:39 *** woudshoo has joined #aichallenge 2011-03-24T05:47:44 but yeah, suppose you got to that node, and from evaluating the first (old second) child you realised 1 wouldn't move there, then you don't need to search the common path 2011-03-24T05:47:56 which might not have linked to one you've already searched 2011-03-24T05:48:03 so you could just move on yeah? 2011-03-24T05:49:02 but that decision was a function of the alpha-beta window you were using 2011-03-24T05:49:25 if you come from a different path, you will have a different window, and might come to a different decision 2011-03-24T05:49:50 hmm yeah okay 2011-03-24T05:50:13 antimatroid: here's a page describing how to update alpha-beta to deal with this: http://web.archive.org/web/20070822204120/www.seanet.com/~brucemo/topics/hashing.htm 2011-03-24T05:50:41 as you can see, the changes are huge, but certainly not trivial 2011-03-24T05:50:48 *changes are NOT huge 2011-03-24T05:51:19 (scroll down to the part where there's code) 2011-03-24T05:51:41 So, I'm looking over the rules from the planet wars competition... 2011-03-24T05:52:06 And it says that the usage of multiple threads and processes is prohibited 2011-03-24T05:52:17 yes 2011-03-24T05:52:26 Is that just kernel threads? 2011-03-24T05:52:42 or userspace threads too? 2011-03-24T05:53:44 phirenz: you realise that is the old contest yes? 2011-03-24T05:53:51 yes 2011-03-24T05:54:01 if i recall correctly, they aren't actually checking whether people are using multiple threads 2011-03-24T05:54:13 antimatroid: this contest will likely have the same rules :P 2011-03-24T05:54:13 and "green threads" (whatever they are) are allowed 2011-03-24T05:54:17 i don't really know much about that stuff 2011-03-24T05:54:30 sigh: yeah, was just making sure :) 2011-03-24T05:55:37 although you would be up for disqualification if they realised 2011-03-24T05:55:43 so one would advise against it :P 2011-03-24T05:55:54 plus, it's more fun to do well fairly 2011-03-24T05:56:26 yes 2011-03-24T05:56:30 yeah, userthreads should be alright 2011-03-24T05:57:02 but either way that should be clarified in the rules 2011-03-24T05:57:16 yep 2011-03-24T05:57:54 sigh: want to do that for me at some point? 2011-03-24T05:58:03 i don't really have half a clue of what i'm talking about with processes 2011-03-24T06:00:02 not me, go tell amstan 2011-03-24T06:01:25 quoting jmcarthur: "You can't use multiple operating system threads, but you can use as many green threads as you want." - http://www.ai-contest.com/forum/viewtopic.php?f=18&t=662&p=3861&hilit=threads+green#p3861 2011-03-24T06:07:26 I'm tempted to write my own treading library 2011-03-24T06:07:36 *** iFire has quit IRC (Read error: Connection reset by peer) 2011-03-24T06:07:50 just so I have control over the scheduler 2011-03-24T06:08:12 *** iFire has joined #aichallenge 2011-03-24T06:08:18 *** iFire has joined #aichallenge 2011-03-24T06:12:52 antimatroid: did you figure out how alpha-beta works with repeated positions yet? 2011-03-24T06:13:13 didn't really look at the moment, i will at some point :P 2011-03-24T06:13:21 i have a quick browse at what he was doing 2011-03-24T06:13:24 had* 2011-03-24T06:18:30 *** iFire has quit IRC (Ping timeout: 246 seconds) 2011-03-24T06:19:09 *** iFire has joined #aichallenge 2011-03-24T07:03:34 *** Accoun has quit IRC () 2011-03-24T07:27:19 *** Accoun has joined #aichallenge 2011-03-24T07:34:49 *** iFire has quit IRC (Read error: Connection reset by peer) 2011-03-24T07:35:08 *** iFire has joined #aichallenge 2011-03-24T07:37:18 wtf, how did I not know about the -S option for git log 2011-03-24T07:40:04 *** boegel has quit IRC (Ping timeout: 248 seconds) 2011-03-24T08:00:26 hmm, do i need a threading library 2011-03-24T08:01:22 phirenz: any particular reason you want threads? 2011-03-24T08:01:33 i don't think people have ever really needed them 2011-03-24T08:01:47 Hmm, can't remember 2011-03-24T08:01:54 lets just backtrack my thoughts 2011-03-24T08:02:52 sigh: i just realised i've become a lazy evaluater of advice for things :P 2011-03-24T08:02:56 oh right, I was thinking about making use of the spare time between turns to do long term strategy 2011-03-24T08:03:11 i used to go look stuff up when pointed to it and go through, but then i went through lots of stuff i never actually use 2011-03-24T08:03:24 now i make a mental note of "if i get frustrated with this, go do that" 2011-03-24T08:04:02 what did you do this time? 2011-03-24T08:04:19 it reminded me of you sending me that link before 2011-03-24T08:04:34 *** iFire has quit IRC (Ping timeout: 246 seconds) 2011-03-24T08:04:34 i don't really care for implementing it now, but i know where to go later when i do 2011-03-24T08:04:46 a similar thing happened with real work haskell advice 2011-03-24T08:05:01 heh 2011-03-24T08:05:13 I just read stuff if it seems interesting at the time, don't if it doesn't 2011-03-24T08:05:45 *** fdsgfdsg has joined #aichallenge 2011-03-24T08:05:54 yeah, if it's super interesting there and then i'll read it, but if i know it might be useful later, i'll try to make some kind of mental note 2011-03-24T08:06:20 *** iFire has joined #aichallenge 2011-03-24T08:07:02 *** FireFly has joined #aichallenge 2011-03-24T08:07:20 *** amstan has joined #aichallenge 2011-03-24T08:07:20 *** ChanServ sets mode: +o amstan 2011-03-24T08:08:13 hi amstan 2011-03-24T08:08:18 hi 2011-03-24T08:08:26 i have time for a chat about wiki/packaging stuff etc. if you do 2011-03-24T08:08:36 nah, i have to leave for school 2011-03-24T08:08:42 that's cool 2011-03-24T08:08:49 i'll keep programming 2011-03-24T08:26:57 *** robbs has joined #aichallenge 2011-03-24T08:28:13 *** FireFly has quit IRC (Remote host closed the connection) 2011-03-24T08:28:41 *** FireFly has joined #aichallenge 2011-03-24T08:45:54 *** p4p4p4 has joined #aichallenge 2011-03-24T08:57:33 *** fdsgfdsg has quit IRC (Quit: Page closed) 2011-03-24T09:04:52 *** sigh has quit IRC (Remote host closed the connection) 2011-03-24T09:22:16 *** UncleVasya has joined #aichallenge 2011-03-24T09:23:09 *** boegel has joined #aichallenge 2011-03-24T09:26:11 *** phirenz has quit IRC (Ping timeout: 264 seconds) 2011-03-24T09:37:53 *** Accoun has quit IRC () 2011-03-24T09:45:31 *** Accoun has joined #aichallenge 2011-03-24T09:57:35 *** needsch has joined #aichallenge 2011-03-24T10:05:06 *** moongrass has quit IRC (Read error: No route to host) 2011-03-24T10:08:31 *** boegel has quit IRC (Ping timeout: 250 seconds) 2011-03-24T10:09:30 *** Accoun has quit IRC () 2011-03-24T10:20:33 *** delt0r_ has joined #aichallenge 2011-03-24T10:21:58 *** delt0r___ has quit IRC (Ping timeout: 252 seconds) 2011-03-24T10:29:21 *** Stocha has joined #aichallenge 2011-03-24T10:47:54 *** needsch has quit IRC (Ping timeout: 252 seconds) 2011-03-24T10:50:23 *** needsch has joined #aichallenge 2011-03-24T10:58:37 *** andy___ has joined #aichallenge 2011-03-24T11:06:50 *** Accoun has joined #aichallenge 2011-03-24T11:06:53 *** needsch1 has joined #aichallenge 2011-03-24T11:07:42 *** needsch has quit IRC (Ping timeout: 252 seconds) 2011-03-24T11:10:27 *** McLeopold has left #aichallenge 2011-03-24T11:17:36 *** |UncleVasya| <|UncleVasya|!kvirc@178-133-68-153.dialup.umc.net.ua> has joined #aichallenge 2011-03-24T11:18:32 *** UncleVasya has quit IRC (Ping timeout: 260 seconds) 2011-03-24T11:33:42 *** McLeopold has joined #aichallenge 2011-03-24T11:47:36 jmcarthur: ping 2011-03-24T11:47:53 *** janzert has quit IRC (Read error: Connection reset by peer) 2011-03-24T11:49:24 *** janzert has joined #aichallenge 2011-03-24T11:49:38 McLeopold: pong 2011-03-24T11:50:07 update for everybody on the haskell lib... i haven't been able to work on it in a while :\ 2011-03-24T11:50:26 so, can you talk trueskill? 2011-03-24T11:50:42 in fact, i might not be able to even participate at all this time. looks like i might be moving around the time i expect the contest to start :( 2011-03-24T11:50:49 yeah sure 2011-03-24T11:51:04 so, the new york interview went well? 2011-03-24T11:51:37 yes it did. i have an offer. all the other interviews i have planned are for out of town places, too 2011-03-24T11:52:04 We've now created an ants game, that may or may not need skill to win. How do we pick a value for beta? 2011-03-24T11:52:24 *** KP13 has joined #aichallenge 2011-03-24T11:52:48 I'd also like to double the xbox live default values to mu=50 and sigma=16.6666 2011-03-24T11:52:48 *** aerique has quit IRC (Quit: ...) 2011-03-24T11:53:33 beta probably needs to be determined dynamically 2011-03-24T11:53:43 not automatically, but by our observations of how things are going 2011-03-24T11:54:03 best i can say for now is to guess :\ 2011-03-24T11:54:40 If you had to guess what it was for planet wars, what would that guess be? 2011-03-24T11:55:04 whether ants is high-skill or low-skill might depend on the combat resolution we choose 2011-03-24T11:55:09 hmm 2011-03-24T11:55:18 janzert provided a list of how far up or down the rankings you were likely not to win when expected. 2011-03-24T11:55:34 *** Kingpin13 has quit IRC (Ping timeout: 250 seconds) 2011-03-24T11:56:13 mine was, I could beat 10 rankings up and lose to 20 rankings down 2011-03-24T11:56:35 by what probability? 2011-03-24T11:56:46 and what was the point distance for these? 2011-03-24T11:56:52 I don't remember 2011-03-24T11:56:57 I need to find the list 2011-03-24T11:57:22 The list should have had the bayeselo scores on them 2011-03-24T11:57:25 was this probability of superiority? 2011-03-24T11:57:37 quote unquote 2011-03-24T11:57:43 I think it was something like that 2011-03-24T11:58:06 likelihood of* 2011-03-24T11:58:29 http://ai-contest.com/forums/viewtopic.php?f=3&t=1208 2011-03-24T11:58:31 *** mceier has joined #aichallenge 2011-03-24T12:00:53 yeah, there it is 2011-03-24T12:00:58 i'm trying to parse this likelihood-of-superiority output. looks like the values range from 0 to 1000. i guess i'll check the bayeselo docs 2011-03-24T12:01:20 trueskill wants an 80/20 value 2011-03-24T12:01:32 i'm not sure these results are exactly applicable 2011-03-24T12:01:54 this is likelihood of superiority, but for coming up with the beta value we want to have a ratio 2011-03-24T12:02:02 wins:losses 2011-03-24T12:02:05 no, but if we can figure out a way to get a beta out of this, we can run initial results through beyeselo and get one for ants 2011-03-24T12:02:30 we might be able to estimate one 2011-03-24T12:02:56 I was thinking we wanted to try to get the average bayeselo score different between each players high and low mark 2011-03-24T12:03:08 *** p4p4p4 has quit IRC (Remote host closed the connection) 2011-03-24T12:03:31 and that would be the beta if we had trueskill ranking on whatever scale this list is on 2011-03-24T12:05:50 LOS is calculated using some given draw ratio, normally 2011-03-24T12:06:15 so, for my bot, if I go up 8, the rating difference is 30, if I go down 18, the rating difference is 33 2011-03-24T12:06:30 so, beta could be about 30 for this scale 2011-03-24T12:06:52 *** andy___ has quit IRC (Ping timeout: 252 seconds) 2011-03-24T12:07:35 ah i see what you mean 2011-03-24T12:07:59 *** Eruonen has quit IRC () 2011-03-24T12:08:05 but this is still for a given draw ratio. maybe janzert used the statistics from the tournament to come up with that though? 2011-03-24T12:08:32 the list shows draw percent 2011-03-24T12:08:36 if it was given a high percentage for the draw ratio then the numbers on the table would generally be lower, i think 2011-03-24T12:08:39 oh? /me checks 2011-03-24T12:09:38 i see draws % for each bot, but not overall 2011-03-24T12:09:46 i guess it could be inferred too 2011-03-24T12:10:18 that 2011-03-24T12:10:35 that's probably high due to being a 2 player game with some bad maps 2011-03-24T12:10:59 I would think ants would be a 0 draw chance 2011-03-24T12:11:37 very small chance, at least 2011-03-24T12:12:16 If I scale this list to 0-100, that makes the 30 beta scale down to 0.075 2011-03-24T12:12:21 which seems a bit small 2011-03-24T12:12:46 maybe we could do the scale as 0-1000 2011-03-24T12:13:27 if we make the scale too large then people might mistake it for elo points 2011-03-24T12:13:39 does that matter? 2011-03-24T12:13:44 i guess not much 2011-03-24T12:14:16 *** FireFly has quit IRC (Quit: swatted to death) 2011-03-24T12:14:23 you get to "climb the ladder" rungs faster :) 2011-03-24T12:14:24 but elo has more absolute meaning than trueskill, and people in the know might try to infer information that isn't really there. i guess it's not a big deal though 2011-03-24T12:14:53 absolute as in your skill value doesn't change if you keep the same ability? 2011-03-24T12:14:56 *** Stocha has quit IRC (Ping timeout: 252 seconds) 2011-03-24T12:15:33 absolute as in if you are X points above another player then that means you beat that player Y% of the time (which of course is maybe an overly liberal assumption on elo's part) 2011-03-24T12:15:36 I though elo and glicko high water marks depended on the number of players in the pool 2011-03-24T12:15:47 oh 2011-03-24T12:15:59 that's what the beta should be 2011-03-24T12:16:02 right 2011-03-24T12:16:09 it's basically fixed in elo, not in trueskill 2011-03-24T12:16:21 beta is fixed, isn't it? 2011-03-24T12:16:31 i mean it's not even configurable in elo 2011-03-24T12:16:36 because the range is free 2011-03-24T12:17:01 but they always saw 200 elo is a 75% chance? 2011-03-24T12:17:04 say 2011-03-24T12:17:11 yeah something like that 2011-03-24T12:17:15 ello 2011-03-24T12:17:17 whereas in trueskill it depends on the parameters 2011-03-24T12:17:39 okay, yes, we have an epsilon as well 2011-03-24T12:17:51 that is the draw percentage 2011-03-24T12:17:52 antimatroid: ping 2011-03-24T12:17:57 antimatroid: i'm available now if you are 2011-03-24T12:17:59 but like you said earlier, i think it's not a huge deal if somebody mistakes it for elo anyway 2011-03-24T12:18:06 ok 2011-03-24T12:18:38 so, in planet wars, the likely hood of superiority is a very small piece of the rankings 2011-03-24T12:18:54 i'm concerned, though, that if we add more precision to the scale then people might assume that it means the scores are more accurate than they actually are 2011-03-24T12:20:22 the implementation stores a higher precision anyway. i think it's fine if we just scale to 100 and round to ones 2011-03-24T12:20:32 *set the scale to 2011-03-24T12:20:44 round to ones for display, i mean 2011-03-24T12:20:50 ok, I'll stick with 100 2011-03-24T12:21:10 actually, the los is really small at the top and bottom of the pw rankings 2011-03-24T12:21:23 beta for that would probably be much higher 2011-03-24T12:21:37 yeah, i was noticing that as well 2011-03-24T12:21:47 seems to be a downside for having a global beta 2011-03-24T12:22:20 upside is that if you know the parameters to trueskill and have a relative score difference then it's a meaningful value, though 2011-03-24T12:23:33 have you guys seen this? http://ai-contest.com/forum/viewtopic.php?f=21&t=1361 2011-03-24T12:23:41 emphasis on 1st and 3rd post 2011-03-24T12:23:42 hmm, in the center, the range is still about 30-40 elo 2011-03-24T12:24:02 *** jmpespxo1eax has joined #aichallenge 2011-03-24T12:24:22 *** jmpespxoreax has quit IRC (Ping timeout: 248 seconds) 2011-03-24T12:24:34 amstan: nice 2011-03-24T12:24:48 another game with it's creator support 2011-03-24T12:24:57 cool! 2011-03-24T12:25:12 how about warlight+communication for next contest? 2011-03-24T12:25:13 I'm emailing him now... 2011-03-24T12:25:20 i like basing our games on existing ones, especially with support :) 2011-03-24T12:25:21 nah.. let's keep it on the forums 2011-03-24T12:25:30 email him with a ping 2011-03-24T12:25:50 or.. use the mailing list! 2011-03-24T12:25:55 that would be appropirate too 2011-03-24T12:25:55 that looks cool 2011-03-24T12:26:07 amstan: i'm here 2011-03-24T12:26:08 *** woudshoo has quit IRC (Remote host closed the connection) 2011-03-24T12:26:18 antimatroid: so.. how's it going? 2011-03-24T12:26:38 alright... i wanted to know what the plans are for packaging? 2011-03-24T12:26:46 a lot of the wiki is hard to fix up without that 2011-03-24T12:26:59 like quick start guides, how to improve your bot etc. 2011-03-24T12:27:24 antimatroid: as a contestant, you'll have to download 2 zips 2011-03-24T12:27:28 - your starter package 2011-03-24T12:27:37 - the tools(visualizer, playgame.py, ...) 2011-03-24T12:27:48 tools will be os specific? 2011-03-24T12:28:02 probably not, since they're in python, html and so on 2011-03-24T12:28:35 we're including the shell/bat scripts for running a game? 2011-03-24T12:28:45 we can just include both, just it needs to be mentioned etc. 2011-03-24T12:28:50 i think we'll make the python part better 2011-03-24T12:28:56 that way there's no need for bat or sh 2011-03-24T12:29:13 ah okay, i'll leave the "getting a game running" stuff for the moment 2011-03-24T12:29:41 McLeopold, jmcarthur: btw, if you're looking at the actual LoS output from bayeselo the likelihood's are given as percentages 2011-03-24T12:29:41 can you put something on the wiki about not using new os processes but green ones are fine too? 2011-03-24T12:29:46 i don't really know what i'm saying there 2011-03-24T12:30:02 antimatroid: the README.md file from tools.zip should be the same as Using the Tools page on the site 2011-03-24T12:30:03 but bayeselo never includes the decimal point (for formatting reasons I believe) 2011-03-24T12:30:31 antimatroid: oh... right.. what page? i can do it now 2011-03-24T12:30:39 specs would be best 2011-03-24T12:30:42 janzert: yeah that's the assumption i was running with. thanks 2011-03-24T12:30:46 but not really sure where to put it either :P 2011-03-24T12:30:57 antimatroid: specs is game specific, this is more about rules 2011-03-24T12:31:09 it's a rule regarding the sandbox 2011-03-24T12:31:10 janzert: did you see my conversation with sigh about doing extensive form games with a DAG? 2011-03-24T12:31:18 also as far as draw percentage, yes I was having bayeselo calculate it from the actual game data 2011-03-24T12:31:19 amstan: perhaps there needs to be a rules page? 2011-03-24T12:31:25 there is a rules page 2011-03-24T12:31:35 :P 2011-03-24T12:31:36 antimatroid: yes, I skimmed over it at least 2011-03-24T12:31:37 idk if on the wiki or not 2011-03-24T12:31:39 lol 2011-03-24T12:32:02 janzert: any though on what beta should be for ants, with a scale of 0-100? 2011-03-24T12:32:02 one thing, generally when doing alpha-beta the tree isn't actually built at any time 2011-03-24T12:32:10 McLeopold: no idea 2011-03-24T12:32:20 blizzard.com? hmm 2011-03-24T12:32:24