Re: [computer-go] Re [Pasky]: Dynamic Komi at 9x9 ?
On Thu, Feb 18, 2010 at 07:39:21PM +0100, "Ingo Althöfer" wrote: > Petr Baudis asked: > > I'm confused. What do you mean by "This is evidence in the 95% > > region"? 3/4 has confidence interval from 19% to 99%, 0/4 has confidence > > interval from 0% to 60%. > > Assume for simplicity that both A and B have a 50 % winning chance > for each single game against C. But they don't! Isn't that the whole point? You are trying to find the winning chances. > The cases with "score difference 7 or higher in favour of A" are: I don't understand what do you mean, sorry. > 8 games mean, that each single result has probability (2 to the power -8). Ok. > So we have for the possible scores > > 3-1 and 0-4 -- prob= 4*(2 to the power -8) > 4-0 and 1-3 -- prob= 4*(2 to the power -8) > 4-0 and 4-0 -- prob= 1*(2 to the power -8) I don't understand the third score. Why are you considering that? -- Petr "Pasky" Baudis A great many people think they are thinking when they are merely rearranging their prejudices. -- William James ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic Komi at 9x9 ?
2010/2/18 > Ingo, > > I'm not a proper statistician, but I believe there's a crucial second step > that's missing in your analysis of significance. Even if this were the only > computer-go test that you personally had ever conducted, we would > nevertheless need to take into account all of the other tests being > conducted within the community. On any given day, some high number of > similar tests are carried out by members of this list. They are testing > different hypotheses to be sure, but that doesn't get us off the hook at > all. > > What it boils down to is this: how frequently does *somebody* get a 95% > confidence result about *something* that isn't going to hold up under > further testing? This issue comes up all the time in epidemiology (e.g. > cancer clusters near power lines), medical studies, bioinformatics, etc.. > When such results are reported, it is usually because the experimenter "felt" that he had a good result. When the same experimenter has a bad result and he is motivated to believe that what is trying will work, he will probably conclude that the bad result is a result of his running the experiment incorrectly and try something else. So what you are going to get is a random sampling of (mostly) good results. You can never be sure that the experimenter did not subconsciously accumulate good results either, tweaking the experiment as he goes (and throwing out the bad tweaks.) I'm not suggesting that bad results should be reported and factored in, but what should happen is that if someone believes they have found a good algorithm and have results to report, the experiment should be repeatable and needs to be verified by the entire community. This does not suggest any dishonesty, it just needs to be done that way.I have conducted experiments myself that returned results well ahead of statistical significance, only to discover that my setup was flawed. For instance I remember one case where the improved version "accidentally" corrected a bug which was not supposed to be part of the experiment. I'm not trying to "refute" any of what has been reported, but I don't see any science here yet.I would like to see a serious study based on a specific proposal or algorithm and I have yet to see that. Don't forget that when you report results with error bars, the test length has to determined in advance. You cannot just stop the test when you feel the confidence interval satisfies you, you have to have determined in advance that you are going to run N games and then interpret what you see based on exactly N games. Don > > - Dave Hillis > > > > -Original Message- > From: "Ingo Althöfer" <3-hirn-ver...@gmx.de> > To: computer-go@computer-go.org > Sent: Thu, Feb 18, 2010 7:28 am > Subject: [computer-go] Re: Dynamic Komi at 9x9 ? > > Hello Don, > several very good points by you! > > > > Does anyone have data based on several thousands games > > that attempts to measure the effect of dynamic komi? > > I would like to see results that are statistically meaningful. > > I had eight handplayed (4 + 4) games on 19x19 with very > high handicap, where the version with dynamic komi (rule 42) > gained a 3-1 score and the version with static komi > performed 0-4 versus the same opponent. This is evidence > in the 95% region that the version with dynamic komi is > not weaker than the static version. > > > We need to see a few thousand games played > > A few hundreds or even a few dozens may be sufficient when > the outcome is very clear. > > > against a fixed opponent WITH dynamic komi, and > > then the same program without dyanmic komi playing > > against the same opponent with the same number > > of games. The number of games must be decided before > > the test is run, or the error margin calculation is > > meaningless. > > I am willing to provide the statistical part, when programmers > run the experiments. > > > > As far as I can tell, nobody has yet to produce anything more > > than anecdotal evidence that this works. > > I have. See the 4 + 4 games mentioned above, > played with my "rule 42". > > > Having a person manually adjusting this after every game is > > completely non-sceientific, unless they are doing it in a fixed > > way with no decision making on their part > > Right. > > > and they are playing thousands of games (or at least > > enough to get statistically significant results.) > > Right, especially also the bracket part of your sentence. > > > I'm not trying to rain on anyone's parade, but I cannot > > understand why no one has produced a statistically meaningful > > result on this subject - > > I would have. Unfortunately I am not a programmer, and am also > not fit in modifying a program code to include dynamic komi. > > But, to repeat it, I am willing to do statistical home > work. > > > I am genuinely interested in this since I never was able to > > make it work when I spent about one intense week on it. > > (I did not do this with ha
[computer-go] Re [Pasky]: Dynamic Komi at 9x9 ?
Petr Baudis asked: > I'm confused. What do you mean by "This is evidence in the 95% > region"? 3/4 has confidence interval from 19% to 99%, 0/4 has confidence > interval from 0% to 60%. Assume for simplicity that both A and B have a 50 % winning chance for each single game against C. The cases with "score difference 7 or higher in favour of A" are: 8 games mean, that each single result has probability (2 to the power -8). So we have for the possible scores 3-1 and 0-4 -- prob= 4*(2 to the power -8) 4-0 and 1-3 -- prob= 4*(2 to the power -8) 4-0 and 4-0 -- prob= 1*(2 to the power -8) in total prob= 9/256 = 3.5 % Other elementary winning probability instead of 50 % give similar results. Ingo. Hint for reading: the Vieweg book of Ernst A. Heinz on computer chess (testing). -- Sicherer, schneller und einfacher. Die aktuellen Internet-Browser - jetzt kostenlos herunterladen! http://portal.gmx.net/de/go/chbrowser ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic Komi at 9x9 ?
Hi! On Thu, Feb 18, 2010 at 01:28:51PM +0100, "Ingo Althöfer" wrote: > > Does anyone have data based on several thousands games > > that attempts to measure the effect of dynamic komi? > > I would like to see results that are statistically meaningful. > > I had eight handplayed (4 + 4) games on 19x19 with very > high handicap, where the version with dynamic komi (rule 42) > gained a 3-1 score and the version with static komi > performed 0-4 versus the same opponent. This is evidence > in the 95% region that the version with dynamic komi is > not weaker than the static version. I'm confused. What do you mean by "This is evidence in the 95% region"? 3/4 has confidence interval from 19% to 99%, 0/4 has confidence interval from 0% to 60%. -- Petr "Pasky" Baudis A great many people think they are thinking when they are merely rearranging their prejudices. -- William James ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re [Dave]: Dynamic Komi at 9x9 ?
Hello Dave, > I'm not a proper statistician, but I believe there's a > crucial second step that's missing in your analysis of > significance. You are right in the sense that I was not precise enough in my statement. Here comes a new attempt. I have three players A, B, C. A plays four times against C, score 3-1 (for A). B plays four times against C, score 0-4 (from B's point of view). If A and B would have the same strength against C then a result like above (score difference 7 or higher) would happen with pobability around 5 %. If A were weaker than B, than the probability would be even smaller. So, I make a statement only for A=MFoG with MCTS and rule 42 B=MFoG with MCTS with static komi C=MFoG old version (without MC, strength around 12k) for some specific high value of handicap. So, no general statement like "dynamic komi helps always in computer go". Ingo. -- GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT! Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Strong programs on cgos 19x19?
Is this 23 cores SMP working on the same tree, or four by 6-cores? I'm running a cluster of four 4-core 2.3 GHz machines, using MPI to share the core of the trees a few times a second. The results between zen-1c, mfgo-16c and pachi-23c are interesting. Zen wins about 60% against many faces and pachi, but pachi wins 70% against many faces. There aren't very many games so this could be just a statistical anomaly. david From: computer-go-boun...@computer-go.org [mailto:computer-go-boun...@computer-go.org] On Behalf Of Jean-loup Gailly Sent: Thursday, February 18, 2010 1:31 AM To: computer-go Subject: Re: [computer-go] Strong programs on cgos 19x19? > The strong pachi is really strong! What hardware is it running on? > Can you say how it differs from the vanilla pachi? It's exactly the same software. The only difference is that is running on 23 cores. I am amazed at how well MCTS scales on 19x19. Looking forward to desktop machines with thousands of cores in a few years... Jean-loup ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic Komi at 9x9 ?
Ingo, I'm not a proper statistician, but I believe there's a crucial second step that's missing in your analysis of significance. Even if this were the only computer-go test that you personally had ever conducted, we would nevertheless need to take into account all of the other tests being conducted within the community. On any given day, some high number of similar tests are carried out by members of this list. They are testing different hypotheses to be sure, but that doesn't get us off the hook at all. What it boils down to is this: how frequently does *somebody* get a 95% confidence result about *something* that isn't going to hold up under further testing? This issue comes up all the time in epidemiology (e.g. cancer clusters near power lines), medical studies, bioinformatics, etc.. - Dave Hillis -Original Message- From: "Ingo Althöfer" <3-hirn-ver...@gmx.de> To: computer-go@computer-go.org Sent: Thu, Feb 18, 2010 7:28 am Subject: [computer-go] Re: Dynamic Komi at 9x9 ? Hello Don, everal very good points by you! Does anyone have data based on several thousands games that attempts to measure the effect of dynamic komi? I would like to see results that are statistically meaningful. I had eight handplayed (4 + 4) games on 19x19 with very igh handicap, where the version with dynamic komi (rule 42) ained a 3-1 score and the version with static komi erformed 0-4 versus the same opponent. This is evidence n the 95% region that the version with dynamic komi is ot weaker than the static version. > We need to see a few thousand games played A few hundreds or even a few dozens may be sufficient when he outcome is very clear. > against a fixed opponent WITH dynamic komi, and then the same program without dyanmic komi playing against the same opponent with the same number of games. The number of games must be decided before the test is run, or the error margin calculation is meaningless. I am willing to provide the statistical part, when programmers un the experiments. As far as I can tell, nobody has yet to produce anything more than anecdotal evidence that this works. I have. See the 4 + 4 games mentioned above, layed with my "rule 42". > Having a person manually adjusting this after every game is completely non-sceientific, unless they are doing it in a fixed way with no decision making on their part Right. > and they are playing thousands of games (or at least enough to get statistically significant results.) Right, especially also the bracket part of your sentence. > I'm not trying to rain on anyone's parade, but I cannot understand why no one has produced a statistically meaningful result on this subject - I would have. Unfortunately I am not a programmer, and am also ot fit in modifying a program code to include dynamic komi. But, to repeat it, I am willing to do statistical home ork. > I am genuinely interested in this since I never was able to make it work when I spent about one intense week on it. (I did not do this with handicap games, but with normal games.) Your sentence in brackets is crucial. I only proposed to use ynamic komi in games with high handicap. Especially I had in ind the situation where the stronger side (giving high handicap) s MC-based. Perhaps, 9x9 instead of 19x19 makes it easier for some programmer o start test series with dynamic komi. Ingo. -- icherer, schneller und einfacher. Die aktuellen Internet-Browser - etzt kostenlos herunterladen! http://portal.gmx.net/de/go/atbrowser __ omputer-go mailing list omputer...@computer-go.org ttp://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] Re [on pasky]: Dynamic Komi at 9x9 ?
Hi!
On Thu, Feb 18, 2010 at 05:43:58PM +0100, "Ingo Althöfer" wrote:
> thx for the diagram.
>
> A few question for clarification:
> Does RAVE mean pachi-with-RAVE?
Yes.
> Does RAVE-linkomi mean pachi-with-RAVE-linkomi?
Yes.
> Did all three bots (GNUGO, RAVE, RAVE-linkomi) have the same (weak)
> opponent in these experiments?
> Who was this "common" opponent?
As I said, they all play against GNUGO Level 10.
> Was this on 19x19?
Yes.
> IF your answers to the first three questions are all "yes",
> then linkomi helps ("blue" never below "green"), and its
> advantage grows with the size of handicap.
Yes. But GNUGo is still better at dealing with handicap, at least when
playing against the stronger version of itself. ;-)
--
Petr "Pasky" Baudis
A great many people think they are thinking when they are merely
rearranging their prejudices. -- William James
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re [on pasky]: Dynamic Komi at 9x9 ?
Hello Petr,
thx for the diagram.
A few question for clarification:
Does RAVE mean pachi-with-RAVE?
Does RAVE-linkomi mean pachi-with-RAVE-linkomi?
Did all three bots (GNUGO, RAVE, RAVE-linkomi) have the same (weak)
opponent in these experiments?
Who was this "common" opponent?
Was this on 19x19?
IF your answers to the first three questions are all "yes",
then linkomi helps ("blue" never below "green"), and its
advantage grows with the size of handicap.
Ingo.
--
Sicherer, schneller und einfacher. Die aktuellen Internet-Browser -
jetzt kostenlos herunterladen! http://portal.gmx.net/de/go/atbrowser
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Dynamic Komi at 9x9 ?
I'm not being critical of anything that has already been presented, I just have not seen it myself and I've been pretty busy working on chess so my focus is not currently on this. But I look forward to reading the paper if it's public. (I'm not going to buy the paper.) Don 2010/2/18 Petr Baudis > On Wed, Feb 17, 2010 at 05:29:36PM -0500, Don Dailey wrote: > > Does anyone have data based on several thousands games that attempts to > > measure the effect of dynamic komi?I would like to see results that > are > > statistically meaningful. We need to see a few thousand games played > > against a fixed opponent WITH dynamic komi, and then the same program > > without dyanmic komi playing against the same opponent with the same > number > > of games.The number of games must be decided before the test is run, > or > > As far as I can tell, nobody has yet to produce anything more than > anecdotal > > evidence that this works. > > In pretty much all of my original dynamic komi mails, I have pointed at > the presentation I've sent here earlier - if you have reservations > against that, I'll be happy to hear about them; the paper I'm preparing > should be ready in couple of weeks. > > For extra convenience, here is the graph from the presentation, > including 95% error bars; I'm sorry, I don't have data about exact > numbers of games anymore. The y-axis is winrate against GNUGo Level10. > > -- >Petr "Pasky" Baudis > A great many people think they are thinking when they are merely > rearranging their prejudices. -- William James > > ___ > 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] Dynamic Komi at 9x9 ?
On Wed, Feb 17, 2010 at 05:29:36PM -0500, Don Dailey wrote: > Does anyone have data based on several thousands games that attempts to > measure the effect of dynamic komi?I would like to see results that are > statistically meaningful. We need to see a few thousand games played > against a fixed opponent WITH dynamic komi, and then the same program > without dyanmic komi playing against the same opponent with the same number > of games.The number of games must be decided before the test is run, or > As far as I can tell, nobody has yet to produce anything more than anecdotal > evidence that this works. In pretty much all of my original dynamic komi mails, I have pointed at the presentation I've sent here earlier - if you have reservations against that, I'll be happy to hear about them; the paper I'm preparing should be ready in couple of weeks. For extra convenience, here is the graph from the presentation, including 95% error bars; I'm sorry, I don't have data about exact numbers of games anymore. The y-axis is winrate against GNUGo Level10. -- Petr "Pasky" Baudis A great many people think they are thinking when they are merely rearranging their prejudices. -- William James <>___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Dynamic Komi at 9x9 ?
Don Dailey wrote: > Ingo Althofer: >> I had eight handplayed (4 + 4) games on 19x19 with very >> high handicap, where the version with dynamic komi (rule 42) >> gained a 3-1 score and the version with static komi >> performed 0-4 versus the same opponent. This is evidence >> in the 95% region that the version with dynamic komi is >> not weaker than the static version. > > Were these games against humans or other computer players? Good questions, but only this first one applies. The strong program was Many Faces of Go with (modified) MCTS (from the official version MFoG 12.013). The weak opponent was Many Faces of Go on a weaker level (if I remember correctly it was level 12k or so). These weaker versions are (still) included in MFoG 12. I ran the two programs on different PCs of comparable strength. Ingo. > If the games were against a human player, were they blind? Did > the player know he was participating in an experiment?Did he > know what results you hoped to see?And were the games alternated > so that the result was not skewed too much by his experience with the > program? -- NEU: Mit GMX DSL über 1000,- ¿ sparen! http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic Komi at 9x9 ?
On Thu, Feb 18, 2010 at 7:28 AM, "Ingo Althöfer" <3-hirn-ver...@gmx.de>wrote: > Hello Don, > several very good points by you! > > > > Does anyone have data based on several thousands games > > that attempts to measure the effect of dynamic komi? > > I would like to see results that are statistically meaningful. > > I had eight handplayed (4 + 4) games on 19x19 with very > high handicap, where the version with dynamic komi (rule 42) > gained a 3-1 score and the version with static komi > performed 0-4 versus the same opponent. This is evidence > in the 95% region that the version with dynamic komi is > not weaker than the static version. > Were these games against humans or other computer players?If the games were against a human player, were they blind? Did the player know he was participating in an experiment?Did he know what results you hoped to see?And were the games alternated so that the result was not skewed too much by his experience with the program? Don > > We need to see a few thousand games played > > A few hundreds or even a few dozens may be sufficient when > the outcome is very clear. > > > against a fixed opponent WITH dynamic komi, and > > then the same program without dyanmic komi playing > > against the same opponent with the same number > > of games. The number of games must be decided before > > the test is run, or the error margin calculation is > > meaningless. > > I am willing to provide the statistical part, when programmers > run the experiments. > > > > As far as I can tell, nobody has yet to produce anything more > > than anecdotal evidence that this works. > > I have. See the 4 + 4 games mentioned above, > played with my "rule 42". > > > Having a person manually adjusting this after every game is > > completely non-sceientific, unless they are doing it in a fixed > > way with no decision making on their part > > Right. > > > and they are playing thousands of games (or at least > > enough to get statistically significant results.) > > Right, especially also the bracket part of your sentence. > > > I'm not trying to rain on anyone's parade, but I cannot > > understand why no one has produced a statistically meaningful > > result on this subject - > > I would have. Unfortunately I am not a programmer, and am also > not fit in modifying a program code to include dynamic komi. > > But, to repeat it, I am willing to do statistical home > work. > > > I am genuinely interested in this since I never was able to > > make it work when I spent about one intense week on it. > > (I did not do this with handicap games, but with normal games.) > > Your sentence in brackets is crucial. I only proposed to use > dynamic komi in games with high handicap. Especially I had in > mind the situation where the stronger side (giving high handicap) > is MC-based. > > Perhaps, 9x9 instead of 19x19 makes it easier for some programmer > to start test series with dynamic komi. > > Ingo. > > -- > Sicherer, schneller und einfacher. Die aktuellen Internet-Browser - > jetzt kostenlos herunterladen! http://portal.gmx.net/de/go/atbrowser > ___ > 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] Strong programs on cgos 19x19?
On Thu, Feb 18, 2010 at 02:57:32PM +0100, Jean-loup Gailly wrote: >Yes it would be interesting but it's a bit difficult to run this >experiment. >The software and its parameters is constantly changing. We can't >create a new kgs bot for every new version or parameter change, >it wouldn't have enough time to get a rating. Also people escape >more often against bots, it takes time to get these escaped games >counted in favor of the bot. I'd argue that measuring performance against humans is the main challenge now. Beating up other monte carlo bots with bigger hardware isn't very interesting. The way to get good measurements is to put a stake in the sand and let it run, preferably with somewhere between 15 and 30 seconds byo-yomi. However, I don't really have the hardware or inclination to run the test myself, so I won't harp on the issue. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
> It would be interesting to know how well Pachi scales on KGS against > ranked humans vs a single core version. Yes it would be interesting but it's a bit difficult to run this experiment. The software and its parameters is constantly changing. We can't create a new kgs bot for every new version or parameter change, it wouldn't have enough time to get a rating. Also people escape more often against bots, it takes time to get these escaped games counted in favor of the bot. Here is one comparison, but it's far from a good scientific experiment. pachiIV and pachi2 on kgs are running exactly the same software or may be differ by a few days. pachiV runs on 4 cores and is 3k. pachi2 runs on different machines, from 15 to 23 cores, and is 2k. This would indicate that scalability against humans is not as good as against bots. But again this is not a scientific experiment. Jean-loup ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
> What is your 23 core hardware? It's actually 24 cores but I'm leaving 1 core free for the OS and other background tasks. The chips are commercially available, search for "6 cores" or "24 cores" on your favorite search engine. Jean-loup ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
On Thu, Feb 18, 2010 at 10:30:56AM +0100, Jean-loup Gailly wrote: >It's exactly the same software. The only difference is that is >running on 23 cores. I am amazed at how well MCTS scales on 19x19. It would be interesting to know how well Pachi scales on KGS against ranked humans vs a single core version. It's well established that monte carlo scales well against itself, but there's very little data against humans, especially in the near dan range. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
Jean-loup Gailly wrote: >> The strong pachi is really strong! What hardware is it running on? >> Can you say how it differs from the vanilla pachi? > >It's exactly the same software. The only difference is that is >running on 23 cores. I am amazed at how well MCTS scales on 19x19. >Looking forward to desktop machines with thousands of cores >in a few years... What is your 23 core hardware? How much is it? -- Yamato ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Dynamic Komi at 9x9 ?
Hello Don, several very good points by you! > Does anyone have data based on several thousands games > that attempts to measure the effect of dynamic komi? > I would like to see results that are statistically meaningful. I had eight handplayed (4 + 4) games on 19x19 with very high handicap, where the version with dynamic komi (rule 42) gained a 3-1 score and the version with static komi performed 0-4 versus the same opponent. This is evidence in the 95% region that the version with dynamic komi is not weaker than the static version. > We need to see a few thousand games played A few hundreds or even a few dozens may be sufficient when the outcome is very clear. > against a fixed opponent WITH dynamic komi, and > then the same program without dyanmic komi playing > against the same opponent with the same number > of games. The number of games must be decided before > the test is run, or the error margin calculation is > meaningless. I am willing to provide the statistical part, when programmers run the experiments. > As far as I can tell, nobody has yet to produce anything more > than anecdotal evidence that this works. I have. See the 4 + 4 games mentioned above, played with my "rule 42". > Having a person manually adjusting this after every game is > completely non-sceientific, unless they are doing it in a fixed > way with no decision making on their part Right. > and they are playing thousands of games (or at least > enough to get statistically significant results.) Right, especially also the bracket part of your sentence. > I'm not trying to rain on anyone's parade, but I cannot > understand why no one has produced a statistically meaningful > result on this subject - I would have. Unfortunately I am not a programmer, and am also not fit in modifying a program code to include dynamic komi. But, to repeat it, I am willing to do statistical home work. > I am genuinely interested in this since I never was able to > make it work when I spent about one intense week on it. > (I did not do this with handicap games, but with normal games.) Your sentence in brackets is crucial. I only proposed to use dynamic komi in games with high handicap. Especially I had in mind the situation where the stronger side (giving high handicap) is MC-based. Perhaps, 9x9 instead of 19x19 makes it easier for some programmer to start test series with dynamic komi. Ingo. -- Sicherer, schneller und einfacher. Die aktuellen Internet-Browser - jetzt kostenlos herunterladen! http://portal.gmx.net/de/go/atbrowser ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
> The strong pachi is really strong! What hardware is it running on? > Can you say how it differs from the vanilla pachi? It's exactly the same software. The only difference is that is running on 23 cores. I am amazed at how well MCTS scales on 19x19. Looking forward to desktop machines with thousands of cores in a few years... Jean-loup ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
