[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [computer-go] 9x9 search, tsume-go

Of course you can't translate Chess programming techniques straight into
Go programming and get immediate results. However, if a good Go programmer
would spend a few man-years on making a good evaluation and move-selection
for 9x9 Go and combine it with Chess search techniques (why reinvent the
wheel), I think you'd get a program much much stronger than any of the
existing Go programs (for 9x9 that is) which was the point of my previous
mail. I don't know if you'd reach 1-dan level but I think for 9x9 it would
probably be the best approach.
I'm glad about the attention 9x9 is getting: I made a conscious decision 2-3 years ago to not make a 19x19 (or even 13x13) version of my program [1], and not enter into any tournaments, until it was playing 9x9 at the strong professional player level.

My overall approach has been to build a 9x9 endgame solver using a very large pattern database [2], and then gradually extend back into the middle game and then try and join up with an opening library. Then (assuming I don't die of old age first) I'll move to 13x13 and repeat.

Why? I suspect a 9 dan 9x9 go program will be dan level at 19x19 with just the addition of a reasonable fuseki/joseki library. 9x9 go is a deep game. The fact that a 6 dan will consistently beat a 1 dan at both 9x9 and 19x19 tells me that the 9x9 game contains everything that is hard about go.

Darren

[1]: A lie - I compile all my unit tests on both 9x9 and 19x19 boards.

[2]: Using patterns with lots of information, not just black/white/empty, which is why I've spent most of this year working on tactical search. Or, rather, not working on it - as often happens, the need to pay the bills has got in the way again :-(.

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go