Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Cutting down my qsearch

Author: Ulrich Tuerke

Date: 09:17:24 04/23/01

Go up one level in this thread


>
>I still do not understand how you can miss something.
>
>I assume that the sequence of captures is not sequence of forced moves and the
>side that captured could avoid capturing.
>I assume also that you do not search in the qsearch to find if there are quiet
>moves to avoid captures and the Qsearch includes only captures.
>
>In this case the sequence of captures can win more than a queen but the sequence
>of a capture and after it a quiet move may lead to the end of the captures.
>
>Assuming that the sides must capture in the qsearch seems risky to me because
>you can miss things because of these assumption.
>It seems to me better to assume that the sides can also avoid moving and prevent
>the opponent to play captures.

Yes, usually the q-search ends when the side on move stands very good. There is
no force to capture. Technically, this means, when the current score for the
side to move is larger than beta, the line will be terminated. Side to move is
not forced to execute insane captures. This is possible, because the q-search
does not consider all moves. There may be reasonable quiet moves which are not
taken into consideration in the q-search.

However in the q-search fututilty pruning above, you terminate the line even
earlier, when the score is below alpha. The idea is that the next ply will
probably produce a score > beta for the other side and terminate anyway. But
that's only an assumption and thus not safe.

IMHO, problems can arise with large positional terms. Remember that above we
were talking about material balances only. Consider the case where the capture
series results in an advanced passed pawn, if continued. We would miss the large
positional bonus in case the last (cutted) capture would generate the advanced
passed pawn.

I must admit that my knowledge about this risk comes basically from empirical
experimenting with the search algos, not from modelling of such move sequences.
But I think, my example should be okay.

Uli

>
>Uri



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.