Re: [computer-go] How to improve my minimax speed?
- Original Message - From: "Chris Fant" <[EMAIL PROTECTED]> To: "computer-go" Sent: Thursday, November 16, 2006 11:00 PM Subject: Re: [computer-go] How to improve my minimax speed? I doubt any serious Go programs base their move selection on proximity to the last move. Cool and how this gonna help me to improve my program? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] How to improve my minimax speed?
- Original Message - From: "Darren Cook" <[EMAIL PROTECTED]> To: "computer-go" Sent: Thursday, November 16, 2006 10:48 PM Subject: Re: [computer-go] How to improve my minimax speed? My simple idea is to analyze first moves near the last move on board, ... Ofcourse we can extend this idea by gathering not only one last move but few of them and on this basis create priority queue. Sounds reasonable. But it will still require you to go through far too many moves. The more sophisticated way would be to use a pattern library, either hand-crafted or automatically harvested. Did I understand You correctly and You saying that some moves shouldn't be analyzed at all? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] How to improve my minimax speed?
>>> My simple idea is to analyze first moves near the last move on board, >>> ... >>> Ofcourse we can extend this idea by gathering not only one last move but >>> few of them and on this basis create priority queue. >> >> Sounds reasonable. But it will still require you to go through far too >> many moves. The more sophisticated way would be to use a pattern >> library, either hand-crafted or automatically harvested. > > Did I understand You correctly and You saying that some moves shouldn't > be analyzed at all? I intended the meaning of when using alpha-beta you want to consider the better moves first. I.e. move ordering is very important. But, yes, unless you've managed to discover a high accuracy, quick evaluation function, full-width search is doomed. You need to prune aggressively. For the hard part of a 9x9 game (moves 8 to 16) you need to be searching 20-40 ply to reach a quiet position that can be "easily" evaluated. For the hard part of a 19x19 game it is more like 100-200 ply. Darren ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Proximity to last move
- Original Message - From: "David Fotland" <[EMAIL PROTECTED]> To: "'computer-go'" Sent: Friday, November 17, 2006 12:37 AM Subject: [computer-go] Proximity to last move Most of the strong programs use pattern matching to select moves to try. Usually there are some patterns that are centered on the last move, so they will do move selection based on the last move. Since the list of moves must be highly pruned, it makes sense to sort it so moves near the last moves are searched first (since they are more likely to be good). If an enemy move threatens a group, I give priority to moves that save that group over moves that save other groups that became threatened on earlier moves. I don't use pure proximity, but I do look at local responses first. David Could You give an example of such pattern or maybe is there somwhere a list of common patterns? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] How to improve my minimax speed?
Shouldn't base the entire game play on the last move. But looking at the last move could be an excellent search optimisation. Indeed, I think any serious Go program "should" look closer at the last move. ;-) The thing is that any local move is as important as any other move in the board. --- Chris Fant <[EMAIL PROTECTED]> escribió: > I doubt any serious Go programs base their move > selection on proximity > to the last move. > __ Correo Yahoo! Espacio para todos tus mensajes, antivirus y antispam ¡gratis! ¡Abrí tu cuenta ya! - http://correo.yahoo.com.ar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] How to improve my minimax speed?
> > I doubt any serious Go programs base their move > > selection on proximity > > to the last move. i think that some fairly serious go players base their selection first on proximity to the last move. if you assume tenuki, you are ignoring the importance of the last move played. it's of some fundamental importance that you answer a peep, for instance, or deal with a bamboo cut or a cross-cut. all of these will likely be dealt with in a 1-2 intersection proximity to the last move played. sure, you may *think* that your full-board evaluation code is really sexy, but in a local fight, you're just plain going to lose the game unless you spend a good fraction of your thinking time making absolutely sure that a local answer isn't needed first. s. The all-new Yahoo! Mail beta Fire up a more powerful email and get things done faster. http://new.mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Testing against gnugo
Orego speaks GTP, as does gnugo. I'd like to run a bunch of games (say, 50) between them to see how many Orego wins. Does anyone have a handy script (ideally bash or Python) for this? Thanks, Peter Drake Assistant Professor of Computer Science Lewis & Clark College http://www.lclark.edu/~drake/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Proximity to last move
On 17, Nov 2006, at 12:49 AM, Wodzu wrote:
Could You give an example of such pattern or maybe is there
somwhere a list of
common patterns?
I just got this last night from a Go-playing computer-programming
friend:
I found new English translations of very popular Japanese Go books
that I am really enjoying:
Dictionary of Basic Tesuji, by Fujisawa Shuko, translated by Steve
Betherick, published by Slate and Shell:
Vol. 1 Tesuji for Attacking
Vol. 2 Tesuji for Defending
Vol. 3 Tesuji for Opening, for Capturing Races, and for Life and
Death problems, Part 1.
In the first section of Vol. 1, on Separating Moves, there is a
discussion
of the "normal" (non-Tesuji) moves, as well as the "clever" moves.
This book gives a clear explanation of the many
different separating moves I had seen in other places:
Jump descent (4th and 2nd lines)
Solid descent ("iron post")
Sideways bump
Diagonal move
Diagonal attachment
Jump attachment
Jump move
Cut
Push and Cut
Double Hane
Knight's Move
Attach and Block
Attach and Hane
Attach and Cut
Attach and Pull back
Attach and Hane
After studying the material, I actually think I can choose the
"appropriate"
move sequence to use in different situations...
I believe this is the type of information that can be coded for use by
a computer, and combined with look-ahead for accurate evaluation of
specific situations.
I agree that "proximity to last move" is an OK place to look for
moves to search,
and I also think that I would *not* use it as part of an evaluation
function, that is,
no extra value for that fact.
Cheers,
David
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Proximity to last move
> I agree that "proximity to last move" is an OK place to look > for moves to search, and I also think that I would *not* use > it as part of an evaluation function, that is, no extra value > for that fact. I second that. Some statistics on the distance of a move from the previous move to back this up: 0-1: 21% 1-1: 12% 1-2: 9% 0-2: 6% 1-3: 4% x-y: Everything else is less than 2% each. This is based on thousands of professional 19x19 games. Together, this adds up to 52%, so yes, this is a good place to start looking, but it also means that for the other half of the moves proximity to the last move is not a good criteria. Anders Kierulf www.smartgo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Proximity to last move
On Fri, Nov 17, 2006 at 12:00:04PM -0700, Anders Kierulf wrote: > > I second that. Some statistics on the distance of a move from > the previous move to back this up: > > 0-1: 21% > 1-1: 12% > 1-2: 9% > 0-2: 6% > 1-3: 4% > x-y: Everything else is less than 2% each. > > This is based on thousands of professional 19x19 games. Together, > this adds up to 52%, so yes, this is a good place to start looking, > but it also means that for the other half of the moves proximity > to the last move is not a good criteria. Interesting. I presume that those moves not close to the previous move, many were close to the move before that, or even before that... Strong players are said to be shifty, squeezing in a forcing move before answering an obvious thing, and so on. -H -- Heikki Levanto "In Murphy We Turst" heikki (at) lsd (dot) dk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Proximity to last move
> Strong > players are said to be shifty, squeezing in a > forcing move before > answering an obvious thing, and so on. it's pretty fun to watch an extremely strong player. they'll make sure to use every ounce of their sente before they give it up. s. Sponsored Link $420k for $1,399/mo. Think You Pay Too Much For Your Mortgage? Find Out! www.LowerMyBills.com/lre ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
