[Top][All Lists]

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

[gnugo-devel] look at regress file

From: kevin yong
Subject: [gnugo-devel] look at regress file
Date: Wed, 26 Mar 2003 13:46:59 -0500 (EST)

Hi, Arend and Evan:

  First, thank you to advice me on using options.

  Following your suggestion, I start to look at some
regress files. I like to share what I found/think, and
like to here some more comments/suggestions. Here are
some comments after I exam the regress/13*13,

The discription says gnugo SHOULD (correctly) move at
M2, but it actually move at J2, which is judged as

There are a couple of points I'd like to make:

1. let's assume this judgement is right. then i went
to trace file, found that: gnugo give high valuation
points to move-reason 'change in territory':
    J2: 31.91 - change in territory
    Move generation values J2 to 39.26
    M2: 19.14 - change in territory
    Move generation values M2 to 24.82
If our human player make judgement say should move at
M2, we need to consider based on what move-reason.
Then, we could improve gnugo by modify its way to
assign valuation points to different move-reasons.
Regarding this specific case, gnugo should lower down
the valuation points to 'change in territory', and
give higher valuation points to the move-reason we
think is more important in our judgement.
If the move-reason we use to judege M2 is better move
is also 'change in territory', then we can look at and
improve the way gnugo assign points to different
positions within the catogary of 'change in

I'd like to here some comments/suggestions from you,
then we can identify where/why gnugo failed in this
specific case, and how could we improve it.

I have other comments about this case, but to avoid
any confusion, i will talk about them in my next

best regards.


Post your free ad now!

reply via email to

[Prev in Thread] Current Thread [Next in Thread]