[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] look at regress file 13x13b:6
From: |
kevin yong |
Subject: |
Re: [gnugo-devel] look at regress file 13x13b:6 |
Date: |
Thu, 3 Apr 2003 22:38:47 -0500 (EST) |
Hi, Evan:
I did some work on 13x13:74.
Problem: gnugo should move at G10, not G7.
From the show_dragons() report, I notice that
there is no attack/defend point produce either for
worm G11 at make_worms() stage, or for dragon J12
(including worm G11) at make_dragons() stage.
Especially, not identify G10 as such a point for worm
G11 or for dragon J12. I think its hard to say there
is something much wrong with these, because worm
G11/dragon J12 is NOT under threat.
Of cause, shapes() analysis recoganized G10 as
strategic effects on dragon M12, and some values
assigned to it. I think its hard to say there is
something much wrong with these either.
My point is: the issue for this case is not about
something wrong with the current code, but about need
to add another new section of analysis. This new
analysis is: gnugo needs think at least one more step
deep.
What I mean is: currently, gnugo only think whats the
best move for the current-color (of course, with some
multi-step reading regarding this move), without
thinking whats the next best move for the
opponent-color;
I did another test: after B move at G7 (step 33), I
let gnugo think for W move step 34. It choose F5. it
did not give value to G10.
As human player, why do we think G10 is move value
than G7? Because if B move at G7, W will move G10, and
have fairly good chance to break through. If B do move
at G10, then it secure the situation.
So, regarding this specific case, gnugo need
improvements on two fronts:
1. as if it think for W to move step-34, it should
choose G10 (thats originally why we think G10 is
important for B to move);
2. so, at the stage when gnugo think for B to move
step-33, it should look at one step deeper: whats the
next best move for W, if B move at G7. it should found
B would best move at G10, and may break through. Then
prune back to step-33 the best move for B is also G10.
kevin.
______________________________________________________________________
Post your free ad now! http://personals.yahoo.ca
- Re: [gnugo-devel] what's wrong with regress file 13x13:77?, (continued)
- Re: [gnugo-devel] look at regress file 13x13b:6, bump, 2003/04/01
- Re: [gnugo-devel] look at regress file 13x13b:6, kevin yong, 2003/04/01
- Re: [gnugo-devel] look at regress file 13x13b:6, Evan Berggren Daniel, 2003/04/01
- Re: [gnugo-devel] look at regress file 13x13b:6,
kevin yong <=
- Re: [gnugo-devel] look at regress file 13x13b:6, Evan Berggren Daniel, 2003/04/03
- Re: [gnugo-devel] how to patch regress 13x13:6, kevin yong, 2003/04/04
- Re: [gnugo-devel] how to patch regress 13x13:6, Evan Berggren Daniel, 2003/04/04
- Re: [gnugo-devel] how to patch regress 13x13:6, Paul Pogonyshev, 2003/04/04
Re: [gnugo-devel] look at regress file 13x13b:6, Arend Bayer, 2003/04/02