Re: [computer-go] Computer go in 7x7
[EMAIL PROTECTED] wrote: Hello all, I am wondering about Go in 7x7. I know that the game in this size has no real interest in itself. However, I think that the level of computer go programs is much higher in 7x7 than in 9x9, and it could be interesting to see until where we can go in 7x7. There was a discussion on this list recently about the level of CrazyStone in 7x7, and also (perhaps not on this list, only on kgs I don't remember) of Magnus saying that the level of Valkyria was very good on 7x7. I am wondering if we can make a player that beats almost all humans in 7x7? I think Crazy Stone does, already. Look at its history on KGS. It won against strong dan players, and even a 5-dan professional. Crazy Stone works with wine in Linux, if you wish to try it by yourself. I have tried to put MoGo on kgs in 7x7, but it looses quite a lot of games. Ok, I have put only 5 minutes time, so it plays quite fast, and perhaps could be better with more time, but it is clearly far from a really good player. A lot of the strength of Crazy Stone comes from its hand-made opening book. I have implemented a meta-UCT (that means that instead of playing a random game after the tree, you make MoGo playing against itself) to generate an opening database. I have made this meta-UCT 100% parallel for a cluster (as the evaluation is now so costly, parallelisation is trivial). I can dedicate quite a lot of computers to this task as I have access to a cluster. So I wonder if you think that this meta-UCT can be effective to generate a good opening database, and this way manage to have a really good computer Go player in 7x7 ? Perhaps usual UCT exploits too much for this task, because my experiments show that the beginning of the tree is quite narrow. Here are the first moves: D4 D5 E5 C4 D3 C5. I have no idea if it is very stupid or normal. Remi was saying that the right komi for 7x7 is 9, but I have made my experiments with komi 7.5, and the meta-UCT predicts that black is almost always loosing. Is it because the blacks moves are very bad? Yes. I am very confident that 7.5 is too small. By the way, using an integer komi is interesting because the possible outcomes are not binary any more. The game can be a win, a draw, or a loss. I have the feeling that UCT works really very well when the outcome is binary. Introducing the possibility of a draw significantly changes the nature of the problem. My feeling is that UCT becomes too selective in that case. To conclude, I think this question of 7x7 is interesting because we can perhaps have a very good player. What do you think? If you have good ideas to make the meta-UCT generate a good database, I can dedicate a lot of CPU to this task, and would interested to see what happen. I don't believe it can beat hand-made book preparation, but I may be wrong. Rémi ___ computer-go mailing list [EMAIL PROTECTED] http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] CGOS server for 7x7
Don Dailey wrote:
I reported earlier that Lazarus wins as white with 8.5 komi. Out of 256
games the win percentage was 53.9 percent.
I tripled the level and ran off 179 games. Now BLACK wins by 54.2
percent. 179 is not enough games, but it appears to be a turnaround.
Of course black SHOULD be winning 100 percent of the games at that komi.
- Don
Hi Don,
Thanks to you and Thomas for setting this up.
Last night Crazy Stone lost one game on time. Probably my ISP cut the
connection for a while in the middle of the night (it has not been very
reliable lately). The error message that I saw this morning is shown
below. I wonder if it would be possible to have your script check for
re-connection instead of terminating, so that the program continues
playing when the connection comes back.
Rémi
error reading "sock5": connection timed out
while executing
"gets $sock s"
("while" body line 17)
invoked from within
"while {1} {
set err [catch {set sock [socket $server $port]} msg]
if { $err } {
puts "server startup return code: $err msg: $msg"
puts ..."
(file "./cgos77.tcl" line 103)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] UCT
Peter Drake wrote: I've decided to bite the bullet and implement UCT in Orego, since (a) everyone else appears to be using it and (b) not being a statistician, improving on this is probably not where I'm likely to make a contribution. Here's my understanding of UCT. It is an algorithm for choosing the best move from a given parent node. Let p = # of runs through parent node r = # of runs through move in question w = # of wins through move in question UCT proposes that we choose the node with the highest sum average + uncertainty where average = w/r and uncertainty = sqrt(2 * logn(p) / r) Is this correct? Are there any things to watch out for? You can try multiplying uncertainty by a well-chosen constant value. This way, you can tune how selective your search is. I found that using a constant < 1 improves the search on 9x9 for Crazy Stone (I use 1/sqrt(10), if I recall correctly). I wonder what is the experience of other UCT programmers, by the way. Also, are people pruning any move for which (average + uncertainty) is less than the (average - uncertainty) of some other move? I am not sure what you mean. In UCT, you only search the move that has the highest average + uncertainty, and you don't search any other at all. -- Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] UCT
Peter Drake wrote: On Nov 3, 2006, at 8:10 AM, Rémi Coulom wrote: You can try multiplying uncertainty by a well-chosen constant value. This way, you can tune how selective your search is. I found that using a constant < 1 improves the search on 9x9 for Crazy Stone (I use 1/sqrt(10), if I recall correctly). I wonder what is the experience of other UCT programmers, by the way. Whee, parameter tweaking! :-) Do you have some reason for choosing this value, or did it just work well in practice? I tried plenty, and measured how they performed against GNU. Also, are people pruning any move for which (average + uncertainty) is less than the (average - uncertainty) of some other move? I am not sure what you mean. In UCT, you only search the move that has the highest average + uncertainty, and you don't search any other at all. By "prune" I mean "remove from the stored tree", making the memory used on that branch available for something else. I have never had memory problems with Crazy Stone. Crazy Stone uses 32 bytes of RAM per node of the tree. It is not really a tree, but a hash table, to handle transpositions. For nodes closer to the root, I "extend" each node with more data (pointers to children). This is only needed for a small proportion of the total number of nodes, so it does not eat up a huge amount of RAM. That is because I don't do UCT in a node before 81 random simulations have been run there. Since I use transpositions, it is not really possible to remove a subtree, anyways. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Positions illustrative of computer stupidity ?
Hi, I am in search of Go positions that are easy to understand for humans, and difficult for computers. I will have to give a talk soon about computer Go to an audience of computer-science researchers who do not know the game. My objective is to try to make them understand why writing a Go-playing program is so difficult and exciting. I'll start by explaining the rules. Then I'll show some examples of tactical thinking in Go, such as ladders, semeai, and a very long connection through bamboo joints. These would illustrate that a naive alpha-beta search approach cannot work. But most Go programs have specific knowledge that allows them to solve these stereotypes. So, in the next step of my talk, I would need examples of positions that are still easy to understand for beginners, but that GNU Go (or your favorite reasonably-strong Go-playing program) would fail to solve because it is not a stereotype. I am going to try to make up artificial positions of this kind for my talk. If you have some to suggest, that would be very welcome. Thanks, Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Positions illustrative of computer stupidity ?
steve uurtamo wrote:
i think that these won't be terribly easy
for your audience to parse. part of the
problem is that gnugo is actually better than a
beginner, for instance.
A beginner cannot beat GNU Go, but it should be easy to make up
positions where a beginner can find the right move and GNU Go cannot.
i'd say anything that isn't tactical in nature
is a good example.
moves that don't directly make any territory, but
which threaten to, for instance. look at the
first 8-10 moves of most any KGS game played by
robert jasiek ("sum") that ended in an early
resignation.
s.
These are much too complicated to understand by beginners.
I have come up with a simple position where GNU Go fails: it is a
position with two ladders. The best move is a double-purpose ladder
breaker at the intersection of the two ladders. GNU Go plays terribly
bad when ladder-breakers are involved. I wonder what other Go programs
play in the attached position.
Also, I would appreciate if some strong players could check this
position for correctness. I tried to make the position balanced so that
if White does not play in the center, then it loses the game, and if it
plays in the center, then it should win. But I am really not sure at all
whether it is balanced or not.
Thanks, by the way, to all who replied to my original question.
Rémi
StupidComputer-0001.sgf
Description: application/go-sgf
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Positions illustrative of computer stupidity ?
John Tromp wrote: On 11/23/06, *Rémi Coulom* <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>> wrote: I have come up with a simple position where GNU Go fails: it is a position with two ladders. The best move is a double-purpose ladder breaker at the intersection of the two ladders. GNU Go plays terribly bad when ladder-breakers are involved. I wonder what other Go programs play in the attached position. But note that white should play K9 to break the ladders. If white plays K10, then black can reply at K9 to secure both ladders. Thanks John. I had not noticed that. Also, I would appreciate if some strong players could check this position for correctness. I tried to make the position balanced so that if White does not play in the center, then it loses the game, and if it plays in the center, then it should win. But I am really not sure at all whether it is balanced or not. I agree W loses if he loses both ladders, but the situation if white saves one ladder is not so clear. Black still holds 4 corners and can reduce w's territory around the ladder w saved. I'd say the odds are still in black's favor. What would you change to make the position more balanced ? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Positions illustrative of computer stupidity ?
David Fotland wrote: Many Faces plays L10, which looks like it also breaks both ladders. -David Thanks for testing. What if Black replies with K9 ? It looks like K9 restores both ladders (to my naive eye). What about the first position I posted, where more tempting moves are available elsewhere ? Do you know whether your program chooses that move for the right reason ? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Miai
Sanghyeon Seo wrote: Recently Remi Coulom asked for Go positions that are easy to understand for humans, and difficult for computers. I missed the thread at the time, so here's my late two cents. GNU Go passes as black. White plays C1. Now A1 and E1 are miai. White wins 1 point. (There are same number of black and white stones on the board.) GNU Go does understand double purpose moves, but in this case, one move creates an eye and the other move threatens to capture, and GNU Go can't combine them. I am interested in other similar positions and results for other programs. Thanks for the position. Crazy Stone has no difficulty with it. I suppose all Monte Carlo programs would find the right move in no time. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Positions illustrative of computer stupidity ?
alain Baeckeroot wrote: Le mercredi 22 novembre 2006 20:44, Rémi Coulom a écrit : Hi, Hi Rémi I am in search of Go positions that are easy to understand for humans, and difficult for computers. One incredibly simple example for human, where GNU Go horribly fails. The only move is tengen (center of the board). I don't know if its a simple bug, or a more difficult evaluation problem. It happens even if all databases are disabled (fuseki and joseki). Thanks to Aloril, who tell it to Min-u Kang (my go teacher :) Cheers. Alain. Thanks Alain. Again, this position is very easy for Monte-Carlo programs. Even on 19x19, Crazy-Stone finds tengen rapidly (in about 30s with a single thread on a 3.4GHz PIV, but this varies a lot depending on randomness). Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte-Carlo is the future of 19x19
Chrilly wrote: I believe that MC will be the only way to write a GO program in the near future leaving the other stuff in the dust (like Mogo has with 9x9 Monte Carlo Go.)This happened in computer chess several times, someone came up with some breakthrough idea, proved it with actual results and everyone else had to play follow the leader to catch up. Do you think its also the future of 19x19 or only of 9x9 (maybe 13x13)? Chrilly ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ I am certain it is for 19x19. Just look at the KGS games of Mogo on 19x19. I played one game against it, and won. I got the feeling it was slightly easier to beat than GNU Go, but that may be because I am used to the way Monte-Carlo programs play. I predict that in one year or two, classical programs will be far behind MC programs on 19x19. Maybe it will take less than one year. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Don Dailey wrote: Hi Steve, What you fail to take into considerations is that a monte/carlo player may ruin it's chances before the weaker player has a chance to play a bad move. The monte carlo player sees all moves as losing and will play almost randomly. I don't agree. Here is the winning percentage I get with Crazy Stone at various handicaps, with a komi of 0.5, over 1 random simulations: 9 Stones: 0.74 8 Stones: 0.73 7 Stones: 0.69 6 Stones: 0.67 5 Stones: 0.63 4 Stones: 0.61 3 Stones: 0.57 2 Stones: 0.54 My program still plays reasonable moves at these winning rates. I tend to believe MC programs would handle handicap better than pure territory-based programs, because they know how to play safe when they are ahead, and risky when behind. If my program is much stronger than its opponent, then it will not play blunders that the opponent can easily take advantage of, whatever the handicap. This being said, I don't believe my program can give handicap to any other on 19x19 ;-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Fw: Compensation for handicap plays?
Don Dailey wrote: I'll take a final poll - speak now or forever hold your peace! Should we: 1. Give white N-1 stones at end of game. (where N = handicap) 2. Give white N stones at end of game. (N = handicap) 3. Give white N stones except handicap 1 case. 4. Not worry about giving white anything but the appropriate handicap stones. Option 4 seems a lot cleaner and is WYSIWYG at end of game along with komi of course. I vote for 2 because that is what KGS does, and that is how I have implemented handicap in my program. It has already been said already, but I insist: what we need is a GTP command to tell the program what are the rules of the game. I would be glad to implement such a command, and would not care about the compensation method, then. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] f(score) instead of sign(score)
Hi Jonas, welcome to the list. The idea of using f(score) instead of sign(score) is interesting. Long ago, I tried tanh(K*score) on 9x9 (that was before the 2006 Olympiad, so it may be worth trying again), and I found that the higher K, the stronger the program. Still, I believe that other f may be worth trying. By the way, Olivier and Sylvain mentioned earlier on this list that they were using floating point in their tree data structure. So MoGo may be using a floating point function to estimate the score of a playout, otherwise there would be no reason to use floating point. But I may be guessing wrong. Maybe they can tell us ? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Lockless hash table and other parallel search ideas
Hi, I have got a lockless hash table to work, and I thought I'd share the results. A lockless hash table is very important, because the usual approach that consists in using a global lock during tree search and update does not scale well, especially on 9x9. But it is possible to create a completely lockless hash table data structure that works with multiple threads. Here are some links that give some indications of how such a thing can be done: http://video.google.com/videoplay?docid=2139967204534450862 http://blogs.azulsystems.com/cliff/2007/03/a_nonblocking_h.html http://www.cs.rug.nl/~wim/mechver/hashtable/index.html Some of those links may look intimidating, but that is because the resize part of the algorithm is complicated. In my implementation, I do not resize the table, so it is very simple. Also, I update counter in each node with atomic increments and decrements (no need to lock). Here is some preliminary experimental data for 9x9 up to 8 cores, running 840,000 playouts, from a tactical middle-game position: (Cores / Playouts per second with spinlock / Playouts per second with lockless hash table) 1 22,477.9 22,447.9 2 37,769.8 43,076.9 3 55,888.2 60,825.5 4 61,448.4 79,470.2 5 64,665.1 95,346.2 6 62,407.1 110,092.0 7 66,508.3 130,638.0 8 59,196.6 146,341.0 BTW, using a pthread mutex is a lot worse than a spinlock, in my experience. I used the fair spinlock from the Linux kernel. But any implementation should work. So, it is pretty cool. This was measured on only one run. Since it is not deterministic, performance may vary from one run to the other (especially since it does not always select the same best move). But it still clearly shows the superiority of the lockless hash table, and seems to indicate that it would still scale beyond 8 cores. I believe I could improve further by reducing the number of atomic operations. Also, thinking about how to reduce atomic operations led me to imagine a scheme that may works as a distributed hash table over a network of PCs. A simple scheme that would work on a single PC would consist in storing one counter per thread in the table. This way, it would not be necessary to use atomic operations to increment counters, and the cache coherency mechanism of the CPUs would handle sending data from core to core. The cost would be that in order to get the node counters, it would be necessary to add N values. Also, some information may arrive a little late to some threads (but I believe it is better to go run a playout rather than wait for data). This scheme is a bit equivalent to using a separate hash table for each thread, and could be generalized to a distributed hash table over a network: each PC would have its own hash table, and each node of the tree would contain two counters: my_counter and other_counter. Every now and then, for instance when my_counter reaches a threshold, this PC would broadcast my_counter to the whole network, so that everybody updates other_counter. I have not implemented this yet, but I will probably try it. Right now, I will test the lockless hash table more, and will probably connect to 9x9 CGOS with that machine sometime during the week-end. If you wish to implement your own lockless hash table, you should read Intel's documentation about its memory architecture. It can be found there: http://www.intel.com/products/processor/manuals/index.htm In particular, it is important to read "Architecture Memory Ordering White Paper", and about the lock prefix, the cmpxchg operation, sfence, lfence, and mfence. The primitives I used in my algorithm are a store fence, atomic increment, atomic decrement, atomic compare and swap. If you understand what they do, you should be able to make your own lockless hash table. Have fun, Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Lockless hash table and other parallel search ideas
Don Dailey wrote: These are used in parallel chess programs, and it's very common. A pretty good article on this written by Hyatt (Crafty programmer and author of former world computer chess champion Cray Blitz) and it's called "A lock-less transposition table implementation for parallel search chess engines", I see an on-line version of a similar article here: http://www.cis.uab.edu/hyatt/hashing.html - Don Hi Don, Yes, I knew Bob's paper. In his approach, an entry will be lost in case of a collision. In my Go program, I never replace hash entries of the current search, because I have enough memory to store them all. I only have to be careful when allocating a node for the first time, so that two threads do not allocate the same slot. This happens rarely enough that the cost of a Test-And-Swap is negligible, so I prefer to do it that way. What I do is essentially the beginning of the Google talk I indicated yesterday, without resizing. I believe it is a lot cleaner than Bob's idea, although atomic increments are costly. In fact, now that I think a little more about it, Bob's scheme would probably not work at all, because updating counters would mean updating the hash code, and any collision would cause a loss of the hash entry. It does not matter for alpha-beta, but losing an entry near the root in MC search would be very bad. Really ugly stuff would be necessary to repair the consequences of such a collision. So, I believe Bob's idea would not work. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Lockless hash table and other parallel search ideas
Olivier Teytaud wrote: Hi, I have got a lockless hash table to work, and I thought I'd share the results. [...] Great! For networks of 4-cores, it is not very useful, but for highly smp machines it could be great - with your grid5000 account, you might run crazystone on a 16-core machine and have a very impressive crazyStone. Olivier Thanks for the tip. Here are the results on 9x9: cores playouts/s 118,025 233,749 348,165 462,268 575,539 687,136 7 100,840 8 115,862 10 134,185 12 158,491 14 174,274 16 205,379 (speedup = 11.4) and on 19x19: 13,780 27,427 3 11,009 4 13,905 5 17,684 6 21,265 7 24,418 8 27,907 10 34,710 12 40,975 14 48,000 16 54,193 (speedup = 14.3) Hardware is 8x Dual Core AMD Opteron 875, 2.2 GHz I'll run tests to try to figure out how much strength is lost by parallelization (ie, what is the winning rate of 10,000 sequential playouts vs 1,000 playouts over 10 processors). Hideki ran similar tests against GNU Go, and found 25 Elo loss with 4 CPUs. So 54,193 playouts per second over 16 CPUs will certainly not perform as well as 54,193 sequential playouts. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] CG'2008 paper: Whole-History Ratings
Hi, This is my CG2008 paper, for statisticians: Whole-History Rating: A Bayesian Rating System for Players of Time-Varying Strength Abstract: Whole-History Rating (WHR) is a new method to estimate the time-varying strengths of players involved in paired comparisons. Like many variations of the Elo rating system, the whole-history approach is based on the dynamic Bradley-Terry model. But, instead of using incremental approximations, WHR directly computes the exact maximum a posteriori over the whole rating history of all players. This additional accuracy comes at a higher computational cost than traditional methods, but computation is still fast enough to be easily applied in real time to large-scale game servers (a new game is added in less than 0.001 second). Experiments demonstrate that, in comparison to Elo, Glicko, TrueSkill, and decayed-history algorithms, WHR produces better predictions. http://remi.coulom.free.fr/WHR/ Feedback is welcome. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] CG'2008 paper: Whole-History Ratings
Andy wrote: Remi, you mentioned how the other algorithms predicted well and guessed that it's because the great majority of games are between experienced players whose strength is not changing much. I also feel that the existing KGS ratings work well for those players already. So how about focusing on how the various algorithms perform in the case of improving players. I think it would be interesting to simulate game results of various improving players and show how the different rating algorithms work. For example: Suppose a player's true strength is 1500 for some time, and then he suddenly improves to 2000. Both before and after he plays a fixed number of games per day (say 10). Show a graph of what each rating algorithm would think his rating is over time. Many people complain that the KGS algorithm does not move fast enough for a case like this. I believe that the main weakness of KGS (and all decayed-history algorithms) is that rating uncertainty grows like the exponential of time. It should grow like the square root of time, which has a completely different shape. So, in the case of players who play frequently (10 games per day is a lot !) the ratings get completely stuck. On the other hand a player who stops playing for a while and comes back to the server will experience huge rating jumps. The WHR algorithm can handle this correctly, but the KGS algorithm cannot, whatever parameter is used. Your suggestion to illustrate the difference on artificial scenarios is good. In fact, you are not the first one to make it. I will probably use artificial scenarios in my presentation at the conference. Also the last paragraph of section 4 talks about how the model does not account for the different ability of new players to change (improve) their ratings compared to older players. Could you vary the parameter 'w' based on the player's current rating? (Assume players with low ratings are capable of improving more quickly than strong players). I don't know enough about the math to know if this would blow up the computation time or if that's simply impossible. Yes, it is my next direction for improving the system. In this paper, I focused on trying to compare different approaches: incremental, decayed history, whr, etc. In order to be fair, I used the same simple but wrong model for every algorithm. Now, I am convinced that WHR is significantly better than alternative approaches, and the next step is to improve the model. Your idea is good, and it would not blow up the computation time. I have already started to work in this direction. I split the game database by sorting the games based on average player strength into different levels, and tune optimal parameters for each level. Not surprisingly, I found that the optimal w² is higher for weaker players than for the stronger players. I still have to find a nice way to handle the fact that with a variable w², ratings don't have a relative value anymore, but an absolute value. It is then important to avoid drift, and some other subtle problems. Thanks for your comments, Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] CG'2008 paper: Whole-History Ratings
Don Dailey wrote: Hi Rémi, For a while I have considered overhauling the rating system for CGOS. My system is ad-hoc and based on gradually increasing K factor based on your opponents K in the standard ELO formula. I don't know if your idea here is feasible for a computer server, because presumably the players are fixed in strength, but in practice I think some bots change. Anyway, I'm no expert on this but want to find something better than what I'm doing and I have considered using some kind of whole history approach (such as running bayeselo after every round on every game, which of course is not very scalable :-) - Don Hi Don, Maybe you could consider implementing Glicko. Glicko is described there: http://math.bu.edu/people/mg/glicko/glicko.doc/glicko.html It should be better than any intuitive hand-made formula you could come up with. Bayeselo would probably produce better ratings than Glicko. Running Bayeselo from scratch after every round may be too costly. But it is possible to make very efficient incremental updates: adding a few games, and running a couple of iterations of MM should be extremely fast. This would require keeping bayeselo in memory all the time, with current game results. Since it cannot be done with the current program you'd have to use my C++ code and somehow incorporate it into the server software. This would be complicated, and may use a significant amount of memory on the server. But computation time would be very short (less than 0.001 second). The algorithm I describe in my paper may be overkill for rating programs. If you look at table 1, you'll see that even when rating humans, Bayeselo outperforms Glicko. Since most programs on CGOS are constant, I believe that Bayeselo would be very difficult to beat. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Programmers representative (ICGA)
Hi, I thought I should let you know that I have accepted to serve as programmers representative on the board of the ICGA. That position had been vacant for a while, and the ICGA offered me to be an interim representative until the election of the new board (which will take place in Beijing). I also plan to be a candidate at the upcoming election. So I am your representative, and any question or suggestion is welcome. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Programmers representative (ICGA)
Jason House wrote: On May 19, 2008, at 10:09 AM, Rémi Coulom <[EMAIL PROTECTED]> wrote: Hi, I thought I should let you know that I have accepted to serve as programmers representative on the board of the ICGA. That position had been vacant for a while, and the ICGA offered me to be an interim representative until the election of the new board (which will take place in Beijing). I also plan to be a candidate at the upcoming election. So I am your representative, and any question or suggestion is welcome. Rémi I don't know what that means... The ICGA is the International Computer Games Association: http://www.cs.unimaas.nl/icga/ The ICGA organises the Computer Olympiad. It also publishes a quarterly journal, and organises conferences. The programmers representative position was created in 1997, because programmers were concerned that the association was run by old university professors who had stopped writing programs for a very long time. So the position was created to let active programmers have more influence in the association. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] learning algorithms such as TD(lambda)- state of the art
Carter Cheng wrote: Hi, I have been leafing through a book "Reinforcement Learning" by Sutton and Barto. The book seems to serve as a decent introductory text to some of the issues which are mentioned in parts of Sylvain Gelly's thesis. The book however, is a decade old and I curious if there are resources which discuss updates and more recent results in this and related fields where you are exploring a state space through sampling methods. I am not sure if any readers of the mailing list have looked at this- but other than the journal Machine Learning are there other good journals which explore this material in a probablistic/statistical framework? Thank again, Carter. This does not really answer your question about journals, but Richard Sutton and Martin Müller currently supervise the PhD thesis of David Silver. Here are their recent publications: http://www.cs.ualberta.ca/~silver/research/publications/publications.html Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Computer Olympiad registration reminder: 11 days left
Dear Go programmers, I remind you that the deadline for early registration to the ICGA Computer Olympiad is June 15. After that date, registration fees will be doubled. You'll find all information on the web site of the tournament: http://www.grappa.univ-lille3.fr/icga/event.php?id=37 I hope to meet you there. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Olympiad registration reminder: 11 days left
David Fotland wrote: Thanks! I just registered. Who else is going? David Great that you registered. The list of registered participants will be published on the web site after the deadline for early registration (June 15). I expect it will be an exceptional tournament. According to a paper in a previous ICGA Journal, ZhiXing Chen and Handtalk should be there. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Olympiad registration reminder: 11 days left
John Fan wrote: What is the protocol? GTP? Usually, games are played with whichever protocol both programs support. In 2006 and 2007, some games were played with gogui-twogtp, some were played on KGS, and some were played by hand. Note that the Go rules of the Computer Olympiad for superko are not the same as those of KGS, so using KGS may be a problem. You can check the rules there: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=20#Go Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Olympiad registration reminder: 11 days left
John Fan wrote: For uct/mc bots rely on the time_left command to dynamically adjust the number of simulations, seems kgs would the preferred protocol. I do not know well enough whether gogui-twogtp tracks time or not. But I guess it does not. Obviously playing by hand is hard to tell the bot how much time left. gogui-twogtp tracks time, but I am not sure it sends "time_left" to the bots, maybe only "time_settings" at the beginning. If both programs are playing locally, then gogui-twogtp looks like the best solution to me Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Olympiad registration reminder: 11 days left
Carter Cheng wrote: I am actually relatively nearby (Hong Kong) and would like to attend but I may not have a competitive program at this time. Is it possible to attend without a program and machine? Regards, Carter. Usually, access to the tournament is free for the audience. I am not sure if accomodation at the Golf club is open to non-participants, though. I bcc this message to the organizers. They should probably add contact information on the web site for such questions. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Presentation of Combinatorial Game Theory by Elwyn Berlekamp
Hi, I just saw this: http://www.usgo.org/resources/downloads/berlekamp.wmv I thought some of you could be interested. References are there: http://math.berkeley.edu/~berlek/cgt/go.html Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[Fwd: Re: [Fwd: Re: [computer-go] Computer Olympiad registration reminder: 11 days left]]
Some answers by the organizers. Rémi --- Begin Message --- Hi Remi, 1) Entrance Audience The ICGA will not charge entrance fee for spectators. We have no information /indication that there are restrictions / entrance fee (as in Italy) by the local organization. 2) Cluster Computing Is allowed. However, we don't have confirmation regarding the internet access. The Chinese are busy with it. Best regards, Mark Winands -Original Message- From: Rémi Coulom [mailto:[EMAIL PROTECTED] Sent: Monday, June 09, 2008 10:36 AM To: Winands M (MICC) Subject: [Fwd: Re: [computer-go] Computer Olympiad registration reminder: 11 days left] Hi Mark, another question on the computer-go mailing list, in case you don't read there. I suppose the answer is yes, but it is better to have the answer by the TD. Rémi --- End Message --- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [Fwd: Re: [Fwd: Re: [computer-go] Computer Olympiad registration reminder: 11 days left]]
Carter Cheng wrote: Thanks Remi. Are standard accommodations being provided by the site organizers or will we have to for the most part make our own arrangements? I guess I have a few more days to decide whether or not to submit a basic program which most likely will not win any games given how stiff the competition is. Regards, Carter. For accomodation, this is the information on the web site: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=19 I vaguely remember that the local organization will also provide meals for a very reasonable price. Regarding the strength of your program, it does not matter if it is not strong. The spirit of this event is mainly to let programmers meet, chat, and have fun together. Some very weak programs participated in previous editions. The more participants, the better. And registering will give you a strong motivation to improve. As the creed of the Olympics says, the most important thing is not to win, but to take part. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Participants in the Computer Olympiad
Rémi Coulom wrote: David Fotland wrote: Thanks! I just registered. Who else is going? David Great that you registered. The list of registered participants will be published on the web site after the deadline for early registration (June 15). I expect it will be an exceptional tournament. According to a paper in a previous ICGA Journal, ZhiXing Chen and Handtalk should be there. Rémi Hi, The list of programs currently registered has just been published. No Handtalk, unfortunately. 9x9: (14 participants) http://www.grappa.univ-lille3.fr/icga/tournament.php?id=180 19x19: (15 participants) http://www.grappa.univ-lille3.fr/icga/tournament.php?id=181 a schedule was posted too: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=22 I may participate in 9x9 too, if I can arrange my teaching schedule in order to miss more than one week. The detailed schedule of individual tournaments is not available yet. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Participants in the Computer Olympiad
Erik van der Werf wrote: On 6/23/08, Rémi Coulom <[EMAIL PROTECTED]> wrote: a schedule was posted too: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=22 Not much details yet in this schedule... I'm interested to participate, but I'd like to know some more details first. I find it strange that the (penalty-free) registration already closed before any schedule was published. The registration page also talks about "events in Amsterdam"; maybe I shouldn't take that date seriously either? Erik Mark told me he would put the detailed schedule online soon. I expect it should be easy to negociate a deadline extension for the penalty. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] UCB/UCT and moving targets
Peter Drake wrote: Can anyone point me to a thread, or at least some buzzwords? I'm having little luck googling for words like "recent" and "forget". Thanks, Peter Drake http://www.lclark.edu/~drake/ Try "discounted UCB": http://computer-go.org/pipermail/computer-go/2007-March/009033.html http://www.lri.fr/~sebag/Slides/Venice/Kocsis.pdf Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] KCC Paduk
Hi, KCC just registered to the 19x19 tournament of the Computer Olympiad: http://www.grappa.univ-lille3.fr/icga/tournament.php?id=181 I remember reading about some plagiarism allegations, for instance there: http://www.msoworld.com/mindzine/news/scandals/scandal0400.html http://web.archive.org/web/2523171809/www.wulu.com/plagiary.htm Anybody has some more recent info about the originality of KCC? Why was it allowed to participate in Gifu? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] 2008 World 9x9 Computer Go Championship in Taiwan
Hi, This was just announced on the ICGA Tournaments web site: http://go.nutn.edu.tw/eng/main_eng.htm It is right before the Computer Olympiad, and registration is free for participants in the Olympiad. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 2008 World 9x9 Computer Go Championship in Taiwan
Erik van der Werf wrote: That's a pretty good deal!!! http://64.68.157.89/forum/viewtopic.php?topic_view=threads&p=193819&t=21591 Why isn't there any sponsoring like this for the other tournaments? Erik I considered entering with my chess program for this reason. But since the tournaments are taking place at the same time, it would be difficult to actually participate in both. Every Go programmer should now quickly write a chess program ;-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] EGC 2008 computer go event
[EMAIL PROTECTED] wrote: http://www.computer-go.info/egc2008/ Hi, I see myself listed as possible participant on that page, so I would like to say that I will not participate, sorry. I won't travel to Sweden, and if I am not going to be there in person, then I prefer to participate in KGS tournaments. Also, I am not very motivated to prepare a Windows version, and explain everything to the operator. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Fuego
Hi, I believe they did not announce it to the list, so in case you don't know, you can find the source code of Fuego there: http://fuego.sourceforge.net/ It seems to be performing very well on CGOS. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Some recent papers
Hi, I recently found those papers. The unimaas website seems to be down right now, but I expect it will be back online soon. THE PARALLELIZATION OF MONTE-CARLO PLANNING: http://hal.archives-ouvertes.fr/docs/00/28/78/67/PDF/icin08.pdf Parameter Tuning by the Cross-Entropy Method: http://www.cs.unimaas.nl/g.chaslot/papers%5Cewrl.pdf Parallel Monte-Carlo Tree Search: http://www.cs.unimaas.nl/g.chaslot/papers%5CparallelMCTS.pdf I could not find the link for suggesting updates to the computer go bibliography, so if the people from the University of Alberta read this, they might like to add those papers to their bibliography. The ICML paper of David is not listed there either: http://computer-go.org/pipermail/computer-go/2008-April/014820.html Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Some recent papers
Markus Enzenberger wrote: Rémi Coulom wrote: I could not find the link for suggesting updates to the computer go bibliography, so if the people from the University of Alberta read this, they might like to add those papers to their bibliography. I probably won't continue maintaining the bibliography (from the beginning of next year on at the latest, when I leave UofA). If someone is interested in maintaining an updated version and knows a stable place on the web for it, I would be happy to send him the files. The bibliography is simple a BibTeX file, which is converted to HTML using bibtex2html and a manually updated HTML page with the recent additions. - Markus If nobody at UofA wants to continue, then I can do it. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Some recent papers
Urban Hafner wrote: Rémi Coulom wrote: | If nobody at UofA wants to continue, then I can do it. How about using citeulike.org? Someone could create a group for computer-go and people could add new papers. That way it wouldn't be the responsibility of just one person to keep the list up to date. And the FAQ says that bibtex files can be imported. Urban This seems to be a good solution. I have just applied to your computer-go group. We just have to try to import the bibtex file. Markus, can you do it ? Or tell us where to download the bibtex file ? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Some recent papers
Markus Enzenberger wrote: Rémi Coulom wrote: This seems to be a good solution. I have just applied to your computer-go group. We just have to try to import the bibtex file. Markus, can you do it ? Or tell us where to download the bibtex file ? I added a link to the bib-file on the website: http://www.cs.ualberta.ca/~games/go/compgo_biblio/ Most of the entries also contain the abstract. A while ago, I found a BibTeX to HTML converter written in Java that could produce separate pages for the entries including the abstract. I thought, it would be nice to combine that with a Google site search box that would allow you to search the abstracts. Unfortunately, the converter was not stable at that time, but I haven't tried a recent version. I forgot the name of the converter, but it should be easy to google for it. Anyway, once you decide on a new place, tell me the URL so that I can add a link from the UofA page, there are a number of inbound links to that page. - Markus Thanks, Markus. I imported it to citeulike: http://www.citeulike.org/group/5884/library I think that citeulike is really very nice. It is possible to get the collection back into bibtex format if we wish to move to another solution. The only problem is that it is a commercial site, and it displays ads. I personaly don't mind since adblock filters them out. From a quick look to the list of tags and authors, it is clear that the current database would need a lot of cleanup. Names of several authors are in the wrong order. For instance Wechsler Harry and Harry Wechsler are probably the same person: http://www.citeulike.org/group/5884/author/Abramson If I understand correctly, every member of the group should be able to edit the entries. So, maybe we can try to test this solution for a while. I miss a way to sort papers by publication date (instead of addition date). Maybe I should re-do the import by first sorting the entries by date in the bibtex file ? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] gnugo3.7.10_10bis on 19x19 CGOS
Martin Mueller wrote: This is a message for whoever is running gnugo3.7.10_10bis on 19x19 CGOS. This program does not seem to finish the games under CGOS rules correctly. I believe you need to set options such as --mode gtp --chinese-rules --capture-all-dead For example, look at game 37209.sgf against "ordinary" gnugo3.7.10_10. The program "lost" by 1.5, but if it captured the dead stones it would win. Thank you Martin Also, now that this mistake is fixed, it would be a good idea to let it play with another name. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Go tournament at EGC, Leksand, Sweden
Nick Wedd wrote: CrazyStonepossible possible This is "yes" from my point of view. It all depends on the availability of an operator. What is the komi for the 9x9 tournament ? I would prefer 7.5 because it is also the komi of the Computer Olympiad. Thanks, Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Go tournament at EGC, Leksand, Sweden
Don Dailey wrote: I also feel for the Mac people and also people that have built programs that run on networks of workstations or other potential supercomputer programs that would not be able to participate. - Don Although I am one of the participants with access to non-conventional computational power, I must say I like the idea of uniform-platform tournaments. Uniform platform allows to avoid comments such as "that program won because it had better hardware", or the frustration of the poor participants that don't have access to big hardware. I like tournaments such as the Computer Olympiad that allow anything, too. It is particularly cool to meet participants such as Hideki who uses a network of playstations. But it does not mean that uniform-platform is evil. It is a different kind of tournament format, that also has its qualities. Nick is in a better position to comment about this, but I also suppose that when a sponsor such as Toshiba provides hardware and prizes, it may not be very happy to see a program win with non-Toshiba hardware. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Go tournament at EGC, Leksand, Sweden
David Doshay wrote: Someday computer Go will evolve enough to have enough trust for remote computing. But not today, unfortunately. Cheers, David The Computer Olympiad has allowed it, at least in 2006, 2007, and 2008. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Computer Go tournament at EGC, Leksand, Sweden
David Doshay wrote: Had I known that I might have participated. I thought I would have to ship my cluster, and with my previous traveling cluster I thought it would never get past the US airport security ... is was such a mass of wires and parts that it hardly even looked like a computer. Cheers, David Note however that you would have probably not been allowed to participate together with GNU Go (I am not completely sure). Since nobody registered GNU Go this year, it may not be too late to enter. But the availability of internet connection in Beijing has not been confirmed yet. Also, you should be able to play on a local computer as a backup in case of connection problem. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] linux and windows
Don Dailey wrote: But all of this presents another issue - most of us don't like to be forced to work in another environment. It really is far more logical to accommodate the programmers when it is not so difficult to do so and especially when it requires much less total effort than making several of them accommodate you because you don't want to be bothered. This is surely much more likely to produce a smoothly running tournament than to force some of the players out of their element and requiring them to cope the best they can. Your post seems to suggest that Nick is not making enough efforts for that tournament, and I think that would be very unfair to him. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Olympiad Schedule
Hi, A schedule for the 9x9 and 19x19 tournaments at the Computer Olympiad has been posted: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=22 Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Monte Carlo evaluation in chess
Rybka 3 has Monte-Carlo evaluation: http://www.chessbase.com/newsdetail.asp?newsid=4772 Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Terminology: MC-UCT vs MCTS vs MC-BFS
Don Dailey wrote: I hate to keep calling this MC-UCT because many programs do not use UCT so until we come up with better terminology I'm going to call it MC-BFS for Monte Carlo with Best First Search. I think "Monte Carlo Tree Search" is a good name. The University of Maastricht people consistently call it MCTS in their papers. I like it much better than MC-UCT, or UCT-like. MC-BFS has the merit of being a little more precise, and maybe would exclude alpha-beta with MC evaluation. But it sounds a little complicated to me. So I'd vote for MCTS. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] August KGS bot tournament
Nick Wedd wrote: In message <[EMAIL PROTECTED]>, Jason House <[EMAIL PROTECTED]> writes Option C sounds best to me too. (C.) Hold it on August 17th. Unless people here persuade me otherwise, I tend to prefer (C). Nick Ok, C is carried unanimously. August 17 it will be. Another problem is the October KGS bot tournament. I would say October 5th, except that this is the last day of the "13th Computer Olympiad" in Beijing, and I don't yet know when its Go events are scheduled for; AND it may be the date of the "World Computer Go Championship" (formerly the "Gifu Challenge"), which has been in early October in the past. I wonder if anyone here has any knowledge of the schedule of the Computer Olympiad, or of the date of the Gifu challenge? Nick This is the schedule: http://www.grappa.univ-lille3.fr/icga/event_info.php?id=22 So, October 5th won't be possible. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Reports on UEC Cup and Japanese Computer-Games Workshop
Nick Wedd wrote: My impression is that in Japan, there are conferences like that. Nick Your message reminds me that I had not told the list yet about those reports I wrote for the ICGA Journal: http://remi.coulom.free.fr/reports/ICGAJ-GPW.pdf http://remi.coulom.free.fr/reports/ICGAJ-UECCup.pdf Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] What Do You Need Most?
Hideki Kato wrote: Mark Boon: <[EMAIL PROTECTED]>: Opposed to removing 9x9. In favor of adding 13x13 wthout removing 9x9. Me too. If, however, limited two 9x9 and 13x13 might be better now as 19x19 is not so utilized, IMHO. It's just early this year many programs started being running on 9x9. I will donate too but no so much until I'll have a job :). -Hideki I second that suggestion of replacing 19x19 cgos by 13x13 cgos. For 19x19, I prefer KGS, it is a lot more interesting. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Gui,GTP, and exploration of internal nodes states.
Denis fidaali wrote: Hi there. To my best knowledge, most people do use Gogui and gtp. This provides interesting ways to see analysis results. But only in a "flat" way. You'll find a tool for visualizing MC trees in gogui, on Guillaume's page: http://www.cs.unimaas.nl/g.chaslot/ Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strange problem while connecting to CGOS
Thomas Lavergne wrote: 15:02:10E->C =c5 15:02:10Apparently, engine crashed But there is no sign of crash in my engine, all goes like if the cgos client have interrupt my engine. Maybe you should send "= c5" (with a space) instead of "=c5". Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] mogo beats pro!
Well done, Mogo team ! terry mcintyre wrote: moves,” like those in the lower right-hand corner, where Moyogo took Typo :-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] What's happening at the European Go Congress?
Gian-Carlo Pascutto wrote: Hi all, there doesn't seem to be any news from the European Go Congress. Nevertheless, I see that partial results were posted: 19 x 19 Results 1stCrazy Stone6/6 2nd Leela 5/6 3rdMany Faces of Go 4/6 9 x 9 Results 1stLeela4/5, SoDOS=13 2nd Crazy Stone 4/5, SoDOS=12 Also I see: Thursday August 7th about 19:00 (17:00 GMT)Demonstration 9×9 game between winning 9x9 program (Leela) and professional. This game should be played via KGS. What happened in this game?? Hi Gian-Carlo, and congratulation for winning 9x9. Leela played a 9x9 game with black and no komi. Leela lost. You can look at the game of Leela in her KGS history: http://files.gokgs.com/games/2008/8/7/egc1p-LeelaBot.sgf Maybe Nick will make a more detailed report. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Nick Wedd wrote: Looking at LeelaBot's games on KGS since the tournament, I see only two: the one I posted, against sestir, and one against egc1p with 0.5 komi, which I cannot open, as it was not finished by the players and KGS is treating it as escaped. Nick The link I sent yesterday works for me: http://files.gokgs.com/games/2008/8/7/egc1p-LeelaBot.sgf Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Gian-Carlo Pascutto wrote: In message <[EMAIL PROTECTED]>, This was foolish of me because I had resumed the game, and was allowing LeelaBot's time to pass. I have carelessly destroyed the evidence of LeelaBot's remaining time. There is now only my word (and perhaps the operator's) for my claim that LeelaBot had more than a minute left. No worries :) I saved the game earlier today after Remi posted the link and before you resumed it. It is included in attachement. Leela scores this at about 30% winning chances for itself after a long think. I have no idea whether that's a reasonable assesement. If I am not mistaken, bottom left is seki. This is probably what Leela misunderstood. And it may also be what you don't understand. The game look like an obvious win for W, starting from move 62. So it looks very fair that Leela did not win on time. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Seki in playouts
Martin Mueller wrote: I did not realize that his program, even with a large tree, would not be able to recognize the seki. I knew of course that the original Mogo playouts had this problem, but I thought all strong programs had solved it by now... Hello Erik, seki in playouts is still an unsolved problem for Fuego as well. I have tried many times to fix it by disabling selfatari moves. But anything I tried made the program measurably weaker overall. Martin Crazy Stone now evaluates most usual cases of seki correctly. The way I do it is extremely complicated. I did not measure an improvement in playing strength from that work, but the program does not play weaker either. I am happy that it now scores every game correctly, and avoids some horrible moves caused by misunderstanding of seki. Also, seki knowledge is necessary to play games with Japanese rules (tournaments in Japan are played with Japanese rules). The sgf file in attachment contains a variety of interesting seki and non-seki situations that the playouts of Crazy Stone evaluate correctly. I believe it contains all the usual kinds. Some sekis are so complicated that it is not reasonable to evaluate them in the playouts. That page is fun: http://senseis.xmp.net/?StrangeSekis My warning: if you'd like seki knowledge in your playouts, then get ready for headaches. Rémi Seki-0004.sgf Description: application/go-sgf ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Depth-first UCT
Gian-Carlo Pascutto wrote: The *paper* about MTD(f) is extremely interesting because it shows that many best-first algorithms can be rewritten as depth-first algorithms. It happened for SSS, it happened for proof-number search. Who will make it happen for UCT? Actually, there was a paper presented at the 2007 Computer-Game Workshop in Japan entitled "Depth-First UCT and its Application to Go", by Haruhiro Yoshimoto, Akihiro Kishimoto, Tomoyuki Kaneko, and Kazuki Yoshizoe. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] rz-74 on CGOS ?
Hi, Anybody knows what is rz-74? It is playing very strong at 19x19, and it looks like nothing I have seen before. Monte-Carlo with a lot of domain knowledge, maybe. A very strong classical program converted to Monte-Carlo would be my guess. Probably not MonteGNU. Handtalk? Go++? Very interesting mystery program, anyways. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] rz-74 on CGOS ?
Don Dailey wrote: I was curious about that too, who is rz-74? The name is perhaps a clue. Is it at version 74?I haven't been watching the games, but are you saying it behaves like a Monte Carlo program? - Don After watching more games, I am less impressed by its strength. It is definitely a Monte-Carlo program (constant thinking time per move, weird moves in the opening, hopeless tricks when losing). Maybe Monte GNU, after all. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] rz-74 on CGOS ?
Magnus Persson wrote: I looked at the last games played by rz-74, and it looks like a MC-program given how how it plays in the opening (odd moves in the center). I also doubt there are any traditional programs who can get 90% against gnugo on 19x19. Are there? I agree that the winning rate against GNU Go is an indication. It is extremely difficult for a MC program to score 94% against GNU. Crazy Stone does it, but it is much stronger than rz-74. Many Faces scores only 73% or so. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] http://www.grappa.univ-lille3.fr/icga/ site down?
Ian Osgood wrote: I haven't been able to access the ICGA tournament site for over a week. Anyone know anything about it? Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Sorry for the downtime. I hope it will be back online very soon. A maintenance operation at the university is taking longer than expected. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] http://www.grappa.univ-lille3.fr/icga/ site down?
Ian Osgood wrote: I haven't been able to access the ICGA tournament site for over a week. Anyone know anything about it? Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ The website is back now: http://www.grappa.univ-lille3.fr/icga/ Sorry for the interruption. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 2008 Olympiad
Ian Osgood wrote: Thanks! I see that KCC Paduk is no longer on the list of participants for 2008. Have they withdrawn? Ian Their registration was rejected because of "past problems with [this] program in other computer Go tournaments" (these are the words of David Levy). The ICGA will make an official statement later. I hope it will give to Chen Zhixing some incentive to participate. He has registered to the TaiZhou tournament, but not to the Olympiad. So have Gostar and some other Chinese programs. That clash between TaiZhou and the Olympiad is very unfortunate. I have noticed that the Olympiad was renamed "13th International Computer Games Championship". I suppose they cannot use "Olympiad" for legal reasons. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Counting the final score with seki
Jacques Basaldúa wrote: But when self atari is avoided in the playouts, groups live in seki and counting dame points really fast is required. You should not worry about that kind of speedup. Looking at neighbors has a very small cost. Negligible in comparison to the playout. Especially if you plan to score seki correctly. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] reminder, there is another possible computer go contest in Taizhou
David Fotland wrote: In late september there is a computer go contest in Taizhou, with cash prizes. They might cancel this contest due to lack of participation, so if you are thinking of going, please let them know today or tomorrow. David How many participants do they have ? In the last message they sent, there were 12 registered participants, including Hand Talk, Go Star, Go Intellect, Many Faces and MoGo. This is not what I would call a lack of participation !!! Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] reminder, there is another possible computer go contest in Taizhou
Erik van der Werf wrote: I already booked my flights; Steenvreter will only participate in the official ICGA events. Personally I think it is bad to have 3 competing tournaments so close to each other. I must say that I'm quite puzzled by the behavior of the local organizers in China. In my opinion, instead of running yet another tournament, professor Liu should have tried to supported the ICGA events with sponsorship for the Go competitions just as was done for the Chess tournament. Erik It is all the more amazing as Professor Liu is also an official member of the organizing committee of the Computer Olympiad. And when I talked to Mark Winands and David Levy about TaiZhou, they replied to me that they were not aware of that tournament! The TaiZhou organizers simply collected e-mail addresses of participants registered to the Olympiad to announce their tournament. Maybe you were not registered yet when they did it. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] reminder, there is another possible computer go contest in Taizhou
No need of translation: http://219.142.86.87/English/index.asp This is the location, as indicated in the invitation e-mail: http://maps.google.com/maps?q=%E5%8F%B0%E5%B7%9E&ie=UTF8&oe=utf-8&client=firefox-a&ll=24.846565,120.058594&spn=67.50517,125.683594&z=3 Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] yet a mogo vs human game
Don Dailey wrote: On Wed, 2008-08-27 at 14:56 -0700, Bob Hearn wrote: The MoGo team has said that MoGo wins 62% of its games against a baseline version when the processing power doubles. That's about half a stone (if you assume you can generalize to human opponents). Yes, I believe it does generalize on average. This data matches my 13x13 study pretty closely, about 62% give or take for each doubling. That is about 90 ELO or so. I have heard that 100 ELO is 1 stone which is what I was basing this on. But it's not clear to me at all if that is true. So I can only guess that 4x in Mogo is worth something like 1 or 2 stones or something between. - Don According to my experience with Go data, it is not possible to give the value of one stone in terms of Elo ratings. For weak players, one stone is a lot less than 100 Elo. For stronger players, it may be more. Also, it is very important to understand that the Elo model is very wrong, and Elo against humans has nothing to do with Elo against computers (and even less with Elo against the previous version). In games against GNU Go, Crazy Stone improved 200-300 Elo points in one year. On KGS, this translated into an improvement from 2k to 1k. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] yet a mogo vs human game
Don Dailey wrote: I don't really believe the ELO model is "very wrong." I only believe it is a mathematical model that is "somewhat" flawed for chess and presumable also for other games. Do you have an alternative that might be more accurate? - Don I don't have very precise data about computer vs human. But in computer vs computer, I believe some of the data of your scalability study clearly demonstrated that strength of MoGo against Leela scaled differently from the strength of MoGo against GNU. But my memory of that data may not be very good. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] yet a mogo vs human game
This was my post about multi-dimensional Elo: http://www.mail-archive.com/computer-go@computer-go.org/msg06267.html I have not tried it since that time. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Kaori-Crazystone
[EMAIL PROTECTED] wrote: Wasn't it today that Crazystone had a match against a professional player? During the FIT2008 conference at Keio University? Does anyone know the result and if the game is available somewhere? Jonas ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Crazy Stone won, with a 8-stone handicap. It was running on an 8-core PC. The game record is attached. I will set up a web page with more details soon. Rémi FIT2008_aoba_kaori_VS_CS.sgf Description: application/go-sgf ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Kaori-Crazystone
terry mcintyre wrote: Congratulations! Thanks. I'm dying for details! What was the time limit? The organizers asked that the program should play at a constant time (30 second) per move. The sgf file contains time stamps (you can see the time with gogui, for instance). I don't know what was her time control, but she apparently played at the same pace as the program. Did the game end on time or by resignation at move 179? She resigned. The pro was Aoba Kaori, yes? Yes. The only other information I have about the match are these pages in Japanese: https://secure1.gakkai-web.net/gakkai/fit/program/html/event/event.html#6 http://www.ipsj.or.jp/10jigyo/fit/fit2008/events.html#1-4-1 I hope the organizers can send me some photos tomorrow. Then I will set up a web page and tell the list. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Kaori-Crazystone
Don Dailey wrote: I'm thinking that if we estimate Aoba at 10d amateur and CrazyStone wins with 8 stone handicap, it is roughly equivalent to beating a 2d player without handicap and that we can subtract 2 stones to say that with pretty high confidence CrazyStone is playing at least 1 kyu (but that's it's much more likely Crazy Stone is stronger than this - after all it performed in this one game at least as well as 2d player.) Hi Don, Crazy Stone played a few games on KGS, and has a strong 1k rank there. When watching the games, I feel that its level is sometimes very high, sometimes very low. If there is a semeai on the board that the playouts evaluate wrongly, then its playing style is absolutely horrible (much lower than 1k). Otherwise, it plays very strong (stronger than 1k). Well, I am only 5k, so I cannot really tell. But when I see the horrors it plays in some games, I suppose it must play much stronger than 1k in some other games in order to get a rating of 1k. Look for instance at these two games: a win: http://files.gokgs.com/games/2008/8/23/CrazyStone-mandelbrot.sgf a loss: http://files.gokgs.com/games/2008/8/23/CrazyStone-beoren.sgf (with comments of the opponents at the end) I have the feeling that MoGo has the same problem. It seems to be typical. Against Kim Myung Wan, it was very clear. In the games it lost, there were big semeais on the board. And MoGo played horribly. In my opinion, this is what explains the apparent difference of level between the blitz and the slow game. Not the time control. Of course, this is just a guess. Maybe members of the MoGo team can confirm/infirm this. When the playouts evaluate a critical semeai the wrong way, then no supercomputer can help, even at long time control. Semeais require a better algorithm, because no computing power can search them out with a tree, and playouts have to be extremely intelligent in order to evaluate them correctly. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Kaori-Crazystone
Andy wrote: I'm excited to see a computer reach 1d as well. For me I'm waiting to see a bot hold a 1d rating consistently on kgs. Right now CrazyStone has been rated 1d briefly, but hasn't been able to maintain it. It's currently 1k. I put a small table of the progress of a few bot's ratings on kgs at http://senseis.xmp.net/?KGSBotRatings I would like to see MogoTiTan play many rated games on KGS and see how it does there. Anyone have a few million dollars lying around to sponsor this? :) Leela is becoming strong. It has reached 1k now. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Kaori-Crazystone
Nick Wedd wrote: When mandelbrot resigns, saying "I was pwned", it appears to me that he is ahead. If he plays at q11 instead of resigning, I think he can kill Crazy Stone's s12 group - but it's difficult, and I'm not sure. Bots are strong at psychological wins :-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Time controls in bots vs human matches
Gian-Carlo Pascutto wrote: Andy wrote: Same thing for bots on KGS. As far as I understand, GTP only supports canadian byo-yomi which is not so popular. Leela should support it but I have never tried. The problem is not GTP, but the KGS client. The time_left of kgsgtp does not give any indication of the number of byo-yomi periods left, if I remember correctly. Crazy Stone plays on KGS with a fast byo-yomi (5x10s). It does not use the byo-yomi. So Crazy Stone plays on KGS with a disadvantage. But I don't mind, and I find it is more friendly to human opponents. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 to 19x19 scaling strangeness
David Fotland wrote: I made a change over the weekend, which looks like it makes 9x9 150 ELO weaker and 19x19 over 200 ELO stronger. Very strange. David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Now you must tell us what the change was :-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 to 19x19 scaling strangeness
David Fotland wrote: Can you put crazystone back on 19x19 so I can see if it is just a fluke against fuego? I added locality to the light playouts - play near last or second to last move, and some code to handle long ladders in playouts. I don’t think this is anything unusual. I think you have a problem with ladders. Crazy Stone played three games, and all started by a huge ladder-like tactical blunder by Many Faces. Maybe you should try keeping the locality change, and dropping ladders in the playout. Crazy Stone does almost nothing special for ladders. The only use of ladder knowledge is as a pattern feature, as explained in my Amsterdam paper. It is not used in the playouts, only for progressive widening. I never managed to obtain any significant strength improvement by incorporating ladder knowledge in the playouts. I can't leave Crazy Stone on CGOS longer, sorry. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: books recommend
Sihai Zhang wrote: > Are there some good books for computer Go course? Thanks! Hi, I am not aware of a book that would cover modern go-programming techniques (ie Monte-Carlo). You'll find plenty of ressources in the Computer-go bibliography: http://www.citeulike.org/group/5884/library Maybe you came to this list from there, but in case you do not know, there is a web page I put up for Chinese programmers: http://www.grappa.univ-lille3.fr/~coulom/ChineseGo/ I copied my files there, because my site at http://remi.coulom.free.fr/ seems to be inaccessible from China. I could mirror some more ressources that are not accessible from China, if there is some interest. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] computer-go textbook ?
Sihai Zhang wrote: Hi all, I am now seeking some good textbooks for computer-go course in my university for undergraduates. I hope I can get some advice from this mailing list, thank you all! Best regards, Jacky Zhang USTC, China ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Hi, In case you missed my (late) reply to your previous message, there it was: " I am not aware of a book that would cover modern go-programming techniques (ie Monte-Carlo). You'll find plenty of ressources in the Computer-go bibliography: http://www.citeulike.org/group/5884/library Maybe you came to this list from there, but in case you do not know, there is a web page I put up for Chinese programmers: http://www.grappa.univ-lille3.fr/~coulom/ChineseGo/ I copied my files there, because my site at http://remi.coulom.free.fr/ seems to be inaccessible from China. I could mirror some more ressources that are not accessible from China, if there is some interest. " There is also a book in Japanese, ISBN4-320-12150-3: http://www.hiroshima-pu.ac.jp/~sasaki/cgf/book/index.html But it is probably out of date regarding current MC techniques. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte-Carlo and Japanese rules
Ingo Althöfer wrote: Hello all, two questions. (i) Do there exist strong 9x9-go programs on Monte-Carlo base for Japanese rules? (ii) Having available only programs for Chinese rules, but playing in a tournament with Japanese rules, which special tricks and settings should be used to maximise winning chances? (This is meant especially in the light of MC's tendency to win games by 0.5 points according to the rules implemented.) Ingo. Hi Ingo, The standard trick is to pass as soon as your opponent passes. For this to work, you need to take a one-point security margin with the komi. You should also score seki the Japanese way. That is how Crazy Stone played the UEC Cup and the matches at FIT2008. In fact, for the UEC cup, Crazy Stone did not understand seki, so I took a bigger security margin with the komi. But a bigger security margin is not good for 9x9. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: hex robot
Dave Dyer wrote: At 01:52 AM 11/27/2008, Denis fidaali wrote: ... But what really lacks (or i wasn't able to find anyway) is a strong community like there is for go. A CGOS equivalent. A GTP equivalent. A Gogui equivalent. A Kgs equivalent. I don't think there's a match between your goals and what boardspace wants, but you ought to discuss cloning and adapting CGOS/GTP for Hex with Don. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ This may be what you want: http://www.cs.ualberta.ca/~mburo/ggsa/ You may also be interested in a few messages in the game-programming forum: http://www.grappa.univ-lille3.fr/icga/phpBB3/viewforum.php?f=8 Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: WMSG - Scoring
Ingo Althöfer wrote:
Concerning the next Computer Olympiad and having in mind
the discussion on the last one ("how fair is 7.5 komi for
9x9 computer games?") the WMSG scoring should be worth
to be discussed for 9x9.
Nobody will want such a scoring in computer go. It is not a matter of
adding one line of code, like Michael Williams suggested. This rule
means that passing first gives a one-point advantage to White. Since
playouts don't pass early, MC programs cannot cope well with this rule.
Rémi
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] UEC cup
Darren Cook wrote: I was reading a report on the UEC Cup [1] on a (Japanese) blog, and if I understood correctly the results were: 1. Crazy stone 2. Fudogo (?) (Hideki Kato's program) 3. Many Faces 4. Katsunari (Mogo had time trouble and pulled out?) Crazy stone then played against Kaori Aoba, 4p, at a 7-stone handicap and won by resignation. Making crazy stone 4 or 5 dan, by Japanese standards. Maybe 2-3 dan European? Can anyone post more information? What hardware was Crazy Stone running on? Darren [1]: The home page is here, but no information here yet: http://jsb.cs.uec.ac.jp/~igo/2008/eng/ Crazy Stone was running on the same 8-core PC that won the H8 game of FIT2008. I am waiting impatiently for more information. At least 3 persons who attended sent an email telling that the game was very beautiful, but I don't have a game record yet. MoGo lost because of problems with remote play. This is the info I have: "MoGo lost the first game of the final round today by time-out due to the overhead of network delay (via KGS) and the manual input of the moves". It seems that Aya took 5th place. Aya was not lucky with the pairings. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: Results of the 2nd UEC Cup (Re: [computer-go] UEC cup)
David Fotland wrote: Congratulations to Remi for Crazystone's second UEC cup victory, and solid win over a professional. David Thank you David. For some reason, games between Crazy Stone and MFG are always complicated and exciting. I watched a few when we were playing on CGOS, and they were always spectacular. I would like also to thank the organizers for their invitation: they arranged everything for Crazy Stone, provided good hardware, and I had to do no work at all. What was your hardware, by the way ? I did not travel to Japan for the tournament, but I will visit UEC in January. I am looking forward to meeting some participants then. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: Results of the 2nd UEC Cup (Re: [computer-go] UEC cup)
jonas.k...@math.u-psud.fr wrote: Although Tei and Aoba Professionals explained the match at the front stage with a projection, the game was so complicated that I couldn't see which is winning until near the end. Another semi-final match, my Fudo Go vs Katsunari, also was shown on the screen but in a small picture at upper right corner and had explained very shortly. Yes, all the people, including me :), were concentrated on your game and exciting. Is there any record of this game between CS and MFG ? Jonas It is attached to this message. Rémi mfg-crazystone-.sgf Description: application/go-sgf ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Format of the UEC Cup
Shunsuke SOEDA wrote: Another reason we chose the single elimination tournament, is that the finals in the single elimination is the "finals", while in other matching systems, the final game might become a dull game. We know that we are sacrificing accuracy, and do want to know what the participants (and might-be-participants) think about how the tournament should be organized. #This includes opinions on how should remote attendance be allowed At first, I was also surprised by the format of the tournament. From a statistical point of view, it certainly does not rank participants in the most accurate possible order. But the ranking of a 9-round tournament always depends a lot on luck, anyways. The advantage of direct elimination is that there is no risk of tie, and it makes the final rounds more exciting. I think the format is very good. Still, I have a couple suggestions. First, it might be better to let everybody play in the qualification tournament, even the previous top 4. The quality of pairings in the second day depends a lot on the quality of ranking of the qualification tournament. Strong programs would have no difficulty to qualify, anyways. And a program that loses a game because of technical problems can try to fix the technical problem for the next game (or the next day). Also, some participants may feel frustrated that they were eliminated in the first day. Maybe they could play swiss between them in the second day. Regarding remote play, I don't like it, but that is probably because I don't profit from it, whereas my strongest opponents do :-) It seems that the UEC Cup is very well organized, and I thank you very much for your work. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Rumors on next Computer-Olympiad
Ingo Althöfer wrote: PS: Are there any gentlewomen programmers around in computer go ? Yumiko Suzuki participated in the First UEC Cup with Bell. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Nullmoves in MCTS and UCT?
Ingo Althöfer wrote: Question: Have nullmove-concepts been tried or analysed in MCTS or UCT-settings? Background of the question: Using alpha-beta tree search, the (asymptotic) percentage of nullmove cutoffs may help as an indicator for the "naturality" or "interestingness" of a (newly invented) game. Unfortunately, it is nontrivial to design "traditional" evaluation functions for newly invented games... Ingo. Null-move pruning only make sense in alpha-beta. MCTS/UCT are more like min-max. They do no alpha-beta pruning, so cannot do null-move pruning. Still, some programmers tried null move in 9x9 alpha-beta go programs. As far as I remember, they got no strength improvement from it. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Nullmoves in MCTS and UCT?
Gian-Carlo Pascutto wrote: Rémi Coulom wrote: Null-move pruning only make sense in alpha-beta. MCTS/UCT are more like min-max. They do no alpha-beta pruning, so cannot do null-move pruning. Null move works like this: if after passing and a small search the position still looks good, do not do a bigger search. There is absolutely no reason why this can't be used in MCTS/UCT. You will have some implementation differences, of course. The problem I see is that null move gives a lower bound on the value of a node in the tree. What value would you back-up to the parent ? If that node is given the value of the null move, then it means that the value of the move that leads to this node will be over-evaluated by the parent. So, it will be explored more with a MCTS algorithm. It would seem more effective to me to play a better real move, get a better value for this node, which will make it more effectively pruned at the parent level. Or do you mean that you would prune the node completely after a null-move search ? That sounds too dangerous. I can't see how it could be done. Pruning in MCTS works by estimating how better on move is than another. With null move, you lose much of that information. I may be wrong, but this makes me believe that null-move pruning is not compatible with the spirit of MCTS. Did you manage to get something to work with null move ? Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Nullmoves in MCTS and UCT?
Gian-Carlo Pascutto wrote: Rémi Coulom wrote: Did you manage to get something to work with null move ? When Leela runs the first simulation in a node, it plays 2 moves for the same side, then does a playout. If the playout loses for the not-passing side, I add x lost games to the RAVE values. I got maybe 10 ELO or so from this. If someone can improve upon it, I'm all ears. Thanks Gian-Carlo, that's an interesting idea. Now I see how it can be done. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 3-4-5 rule
Don Dailey wrote: You are right, the d3p version rallied to come from behind and staged an exciting and dramatic comeback: Rank Name Elo+- games score oppo. draws 1 d3p 2016 77 7521 52% 20000% 2 base 2000 75 7721 48% 20160% - Don If you'd like to try a simple pruning scheme that improves playing strength on 19x19, then I'd suggest progressive widening. It only works in the tree, not in the playouts. You don't need complex patterns for progressive widening to work. You can simply use distance to the previous move. Search moves at distance 1 from the previous move for N playouts. Then add moves at distance 2 for N*x playouts. Moves at distance 3 for N*x*x. N and x are the two parameters that need tuning. I expect you can get tremendous strength improvement on 19x19 with this simple scheme. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 3-4-5 rule
Don Dailey wrote: I'm not sure I understand - when you say N playouts, do you mean N visits of that node? Because once you visit a node, you expand it, no longer doing playouts from that point. Yes, I mean N visits. In my view, every playout starts at the root. For instance if e5 is played (from the root position) are you saying we would only look at the moves touching e5 the first few times e5 was visited, then start looking at distance 2 for a while, and so on stopping after 3? Yes, except that you would not stop after 3. Continue at distance 4, then 5, etc. This guarantees that in the limit of infinite thinking time, this algorithm would be optimal. The reference bot of course does not build a tree, what I'm actually looking for is a way to produce a medium strength but really simple bot that does not build a tree and just has a lot of playout magic. I wonder if this behavior can be emulated in the playouts somehow? It's not so simple because we are not expanding a tree, and when we look at some moves more often than others we get more statistical noise in the moves we don't look at, and this can make them look artificially good or bad. I can definitely see how such a scheme would work well in the tree. - Don Oh, sorry, I thought your program was an UCT bot. If you build no tree, then progressive widening probably does not help much. I am not sure. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
