Re: [gnugo-devel] Opponent follow-up value and group connection

[SP LEE]

I have revised the patch to include shape factor for larger groups. Now it's
changed to have maximum 15 from the territorial_value to be weighted by the
shape_factor, either bonus or punishment. So (territorial_value-15) will be
contributed to the total value without being multiplied by the shape factor.

>From the regression test 3 unexpected passes and 3 unexpected fails are
generated. If add my game, there is one more pass.

make all_batches
    .
    .

./regress.sh . strategy2.tst
53 unexpected FAIL: Correct 'S15', got 'E10'
./regress.sh . trevor.tst
1110 unexpected PASS!
./regress.sh . 13x13.tst
52 unexpected FAIL: Correct 'H10', got 'N9'
./regress.sh . nngs2.tst
600 unexpected PASS!
./regress.sh . nngs3.tst
260 unexpected PASS!
./regress.sh . 13x13b.tst
40 unexpected FAIL: Correct 'F8', got 'H10'

It can be concluded that the shape factor could generate both good and bad
results. For the fail of strategy2 53, GNUGO didn't play local letting the
corner group dead. However E10 seems to be very big too. The 13x13.tst 52 is
also a life/death problem. Anyway I will check the unexpected fails. I think
it's rather risky to use shape factor to solve life/death problem.

SP Lee

----- Original Message -----
From: "Heikki Levanto" <address@hidden>
To: "GNU Go development" <address@hidden>
Sent: Wednesday, May 14, 2003 12:31 AM
Subject: Re: [gnugo-devel] Opponent follow-up value and group connection


> On Tue, May 13, 2003 at 01:41:12PM -0700, SP LEE wrote:
> > The fix here is like "forget the shape factor if either attacking or
> > defending a dragon of more than 20 points is involved." This kind of
play
> > can be seen even in professional games.
>
> I have mixed feelings about this. I agree that we should play the best
> attack/defense move, even when it is ugly. But when there are two possible
> attacks, I still believe we should prefer the one that attacks in good
> shape.
>
> I would recommend reducing the shape bonus for big groups, perhaps
> proportionally to their size, but never removing it completely.
>
>
> --
> Heikki Levanto  LSD - Levanto Software Development   <address@hidden>
>
>
>
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gnugo-devel
>

patch
Description: Binary data