[Top][All Lists]

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

Re: [Gnushogi-devel] Comparing GNUShogi 1.2 and later versions

From: h . g . muller
Subject: Re: [Gnushogi-devel] Comparing GNUShogi 1.2 and later versions
Date: Tue, 20 Feb 2018 21:26:54 +0100
User-agent: SquirrelMail

I solved the the pondering: it was a fault of the adapter. Due to the way
the easy & hard command were treated in USI, converting it first to an
'option' command, and then treating that, the hacked adapter treated these
commands twice. And in contrast to CECP, 'easy' for GNU Shogi is a toggle
command. So sending it twice had no effect.

I fixed that, and now CrazyWa and GNU Shogi use the same amount of CPU. (I
attached the fixed XS2WB source.) It doesn't seem to have much effect on
the result, though: at 40moves/min CrazyWa is now leading by 57-6 (plus 2
repetition draws). It looks like CrazyWa loses mainly by its evaluation
making it pursue a wrong goal, which in the end turns out to be losing,
and not so much because GNU Shogi surprises it. Most of the time that is
awarding the destruction of the opponent's King safety more than it is
worth, sacrificing huge amounts of material to draw the King to the middle
of the board, and then failing to checkmate it there.

CrazyWa has a rather special approach to King safety, other than any other
engine I know: the King safety is asymmetric, only the side to move being
awarded a bonus for the 'exposure' of the opponent King. The latter being
calculated as the number of attacked empty squares next to the King times
the value of the pieces in hand. This is rather course, but it seems to
work well in all variants with drops. I suppose a more precise evaluation
of the King exposure could be made using variant-specific weights, which
would reduce the umber of self-inflicted losses.

Attachment: XS2WB.c
Description: Text Data

reply via email to

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