Re: [computer-go] Details of AnchorMan
This is happening everyday for me. My IP is not changing. I don't think it's a lag issue. But I could be wrong. Is it possible that there is a bug in the Windows TCL interpreter? How many other people out there are running TCL on Windows for cgos? On 2/6/07, Magnus Persson [EMAIL PROTECTED] wrote: I have the problem that my DSL provider disconnects me and give me a new IP-adress. When that happens my programs lose on time in a similar to your problem description. My solution is to disconnect/connect my Internet and CGOS connection manually often enough. I also had some problems with lag but this is unlikely to cause your problem. Quoting Chris Fant [EMAIL PROTECTED]: It seems that some of my games are being lost on time after only a single move (for instance http://cgos.boardspace.net/public/SGF/2007/02/06/465927.sgf) -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] cgos ggexp
To who it may concern: ggexp appears to be losing all of it's games on time. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Why not forums?
Eduardo Sabbatella wrote: No please. I use my email client, I sort them, I store them I'm happy with it. Personally, I will not be able to read the forum at work. It will be the difference between reading and not reading the list. I want to choose which info will push me, and forget. I don't want to log into a forum every time I remember about Go. (I have a very bad memory) Mailing lists exist on internet since 20+ yrs ago and continues to be used, they are not outdated! I don't care about having an AVATAR. My 2 cents. Eduardo I agree - please don't move to a forum... ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Monte Carlo (MC) vs Quasi-Monte Carlo (QMC)
Upon continuing to learn about the general Monte Carlo field, I've found it seems there is a general consensus in this community about a distinction between Monte Carlo (MC) and what appears to be commonly called Quasi Monte Carlo (QMC). MC is defined as using random/pseudo-random distributions and QMC using more deterministic or designed distributions that fit the problem better. Just do search on google web or scholar and you'll get a wealth of hits. But here are a few links to documents or pages that specifically address this terminology: http://www.arts.cornell.edu/econ/CAE/final.pdf http://www.mas.ncl.ac.uk/~ngl9/docs/MCQMC.pdf http://www.math.hkbu.edu.hk/~gwei/sci3510/ch1.pdf http://mathworld.wolfram.com/MonteCarloMethod.html http://mathworld.wolfram.com/Quasi-MonteCarloIntegration.html It also seems that today quite often Monte Carlo generally is used to describe any kind of statistical sampling using random or other distributions to approximate solutions to problems like David Doshay pointed out. So would it be helpful to distinguish between MC go and QMC go programs - maybe a little. Since I'm just learning about this I might be misunderstanding some concepts. But besides doing obvious things like minimizing memory usage and optimizing code so that you can increase the sample size, there are many well known strategies to decrease the variability of the simulation. These are called variance reduction techniques. Generally Monte Carlo standard error decreases based on the square root of the sample size (quadrupling the sample size cuts the the standard error in half). I would think in part this would depend on the problem, so not sure if this applies to MC go or how to measure. Variance reduction methods are used to improve the distribution improving the results (error) without increasing the simulation size as much. Here is a list of some of them without any explanation (searching on any of these terms with monte carlo should turn up lots of hits): -Common Random Numbers -Antithetic Variates -Control Variates -Importance Sampling -Stratified Sampling -Conditional Sampling -Systematic Sampling Most of the research using MC methods seems to be for numerical integration, finance applications, and physics applications; not applied to game theory. It may be be challenging to understand how to translate these ideas to MC go or whether they would be helpful. -Matt ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte Carlo (MC) vs Quasi-Monte Carlo (QMC)
It seems that there are at least three cases: 1: Choosing a random move from a uniform distribution 2: Choosing a random move from a nonuniform distribution (patterns etc.) 3: Choosing a move taking into account what has been chosen before The concensus seems to be that numbers 1 and 2 are MC and 3 is QMC. Mogo uses QMC within the tree in memory and MC for the leaves, so which should it be called? And about reducing variance: In games you only care about estimating the goodness of the best moves (in order to select the best one). You don't care how bad a move is, if you are fairly certain that it is not the best one. You should thus reduce the variance of the best moves, that is, study them more often. This is exactly what UCT is about, reducing the variance of variables of interest. I could see a case where it is possible to reduce a variance of a single variable even in the 0-1 case. Let us say that black has about 5% chances of winning. If we could (exactly) double the chances of black winning by changing the nonuniform sampling somehow (say, enforce bad moves by white), we could sample from that and divide the estimated black's winning chance in the end by 2. This would of course be very difficult in practice. (A binary random variable gives more information when the chances are closer to 50-50.) This could be useful in practice in handicap games, by for instance enforcing a black pass with 1% chance every move. Sampling would be distorted towards white win, which is realistic since white is assumed to be a stronger player, anyway. To summarise, I agree that there are links to other MC research, and they should be explored. -- Tapani Raiko, [EMAIL PROTECTED], +358 50 5225750 http://www.cis.hut.fi/praiko/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] cgos ggexp
I just checked this for January and here are the statics: When playing white ggexp played: 1087 games 295 losses 8 of these were time losses. When playing black ggexp played 1036 games 341 losses 17 losses So I don't see that it's losing all it's games on time. - Don On Tue, 2007-02-06 at 08:00 -0500, Chris Fant wrote: To who it may concern: ggexp appears to be losing all of it's games on time. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] cgos ggexp
It lost several games in a row on time at the time that I sent that message. Obviously, it can't have lost ALL of it's games and still attained an 1800 rating. On 2/6/07, Don Dailey [EMAIL PROTECTED] wrote: I just checked this for January and here are the statics: When playing white ggexp played: 1087 games 295 losses 8 of these were time losses. When playing black ggexp played 1036 games 341 losses 17 losses So I don't see that it's losing all it's games on time. - Don On Tue, 2007-02-06 at 08:00 -0500, Chris Fant wrote: To who it may concern: ggexp appears to be losing all of it's games on time. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] cgos ggexp
On Tue, 2007-02-06 at 14:16 -0500, Chris Fant wrote: It lost several games in a row on time at the time that I sent that message. Obviously, it can't have lost ALL of it's games and still attained an 1800 rating. I assumed that you meant that of all the games it lost, they were mostly due to time. I knew you didn't mean it loses every game. This happens sometimes when you run a program on CGOS with a machine that is doing other things and is heavily loaded. Or it can happen if for some reason the internet connection is not stable. - Don On 2/6/07, Don Dailey [EMAIL PROTECTED] wrote: I just checked this for January and here are the statics: When playing white ggexp played: 1087 games 295 losses 8 of these were time losses. When playing black ggexp played 1036 games 341 losses 17 losses So I don't see that it's losing all it's games on time. - Don On Tue, 2007-02-06 at 08:00 -0500, Chris Fant wrote: To who it may concern: ggexp appears to be losing all of it's games on time. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte Carlo (MC) vs Quasi-Monte Carlo (QMC)
ivan dubois wrote: I dont understand how you can reduce the variance of monte-carlo sampling, given a simulation can return either 0(loss) or 1(win). Maybe it means trying to have mean values that are closer to 0 or 1 ? Well strictly speaking I agree the standard models don't fit that well - the application of monte carlo to go is much different than traditional applications. However, imagine the whole path of a simulation to the leaf as a meaningful set of points. We are only measuring the end, but the path is very important too. Also, as you mentioned one could target the larger scoped variance of the set of simulations mean value correctly classifying the point as a win or loss. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte Carlo (MC) vs Quasi-Monte Carlo (QMC)
It seems that there are at least three cases: 1: Choosing a random move from a uniform distribution 2: Choosing a random move from a nonuniform distribution (patterns etc.) 3: Choosing a move taking into account what has been chosen before The concensus seems to be that numbers 1 and 2 are MC and 3 is QMC. I don't think 3 is an accurate description of MC. Generally, MC is a process where a number of paths (aka sequences of random numbers) are used to sample some function. In go, a path would be a single playout, and the function is the score. QMC is when the paths are constructed using variance reduction techniques, meaning that they are more representative of the sample space. AFAIK no one has used any QMC techniques in go; I really doubt they would be much help because the function (the score) is not smooth in the inputs (that is, small changes in the path are not small changes in the score). I think what is confusing the matter is the sample space--i.e. what games we are evaluating. The standard MC engine's sample space is all games that don't fill an eye. Better might be all games that don't fill an eye and don't play self-atari. Mogo has an even more restrictive sample space, designed to be a much better evaluation function. Finally, UCT is not MC. MC is an evaluation function, UCT is a tree search technique. You could just as easily use UCT with any other (stochastic) evaluation fuction, or MC with any other tree search. It turns out that UCT has proven to be very effective using MC evaluation. So, one could say Mogo is UCT, with a MC evaluation function, with heuristics to improve the MC games. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] MC Go Effectiveness
It seems to me, the fundamental reason MC go (regardless of details) works as it does is because it is the only search method (at least that I am aware of) that has found a way to manage the evaluation problem. Evaluation is not as problematic because MC goes to the bitter end where the status is known with certainty. With random distributions it probably tends to find robust moves that leave a lot favorable options open. With MoGo, Sylvain has shown that better simulation policies can achieve much better results. But what are some of the reasons MC is not even better? -Since MC engines don't deal with tactics directly, they're not likely going to play tactical sequences well for low liberty strings, securing eye space, cutting and connecting, ko fights, or ladders, etc. -Also because most of the play-outs are usually nonsense, they may have trouble dealing with meaningful nuances because the positions that will lead to these distinctions just don't arise with enough statistical frequency in the play-outs to affect the result. Yet when very selective moves are used in the play-outs, too many possibilities can be missed. -Finally, with 19x19 anyway, the size of the board and game tree probably limits the practical effectiveness of the sampling and move ordering. I don't try to address this last point any further in this message. So here is an idea for MC research: Incorporate multiple types of distributions in one MC player. Available time resources would be divided between the different distribution methods. Then the results of these could be combined in some kind of sum/rank/vote/etc. For UCT this could be used to direct the search at those most interesting nodes. As an example, distributions such as these could be used: 1. A random or near random distribution 2. A more selective pattern based distribution 3. A simple tactical reader based distribution - this might not be obvious how to implement, but perhaps it could play tactical sequences if such conditions (based on heuristics) existed on the board, otherwise switch to one of the others. With regard to variance reduction techniques, #2 and #3 might be examples of importance sampling and conditional sampling. And the above overall method might fall under the category of stratified sampling. Thoughts? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
