[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] tactical move ordering tuning
From: |
Gunnar Farneback |
Subject: |
Re: [gnugo-devel] tactical move ordering tuning |
Date: |
Tue, 19 Nov 2002 19:42:08 +0100 |
User-agent: |
EMH/1.14.1 SEMI/1.14.3 (Ushinoya) FLIM/1.14.2 (Yagi-Nishiguchi) APEL/10.3 Emacs/20.7 (sparc-sun-solaris2.7) (with unibyte mode) |
Evan wrote:
> reading.tst from 106624 to 93970 (11% reduction)
> owl.tst from 31198038 to 30678644 (1.7% reduction)
>
> I tuned it against reading.tst because that was fast to run, but more
> performance gains can probably be achieved with the selection of a better
> benchmark.
It's important to evaluate it on as large a test set as possible. The
patch looks reasonable in any case.
> This gets a FAIL in owl:170. I believe this is because attack() returns a
> different move, with the same code, that is then used as a vital owl
> attack, and is less successful in that role. I believe this should be
> counted as purely accidental.
Probably, but it would be good to know what kind of change in attack
point was involved.
> However, I think it means it might be worth while to write a function that
> gives an array of all attack moves (and a corresponding defense function)
> with the same highest code, and use all such attack moves in owl_attack.
Such functions would be useful in a number of places, but I don't
think we can afford to try all tactical attack moves in owl_attack().
/Gunnar