Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: adaptive search

Author: Vasik Rajlich

Date: 05:30:49 01/20/06

Go up one level in this thread


On January 20, 2006 at 07:18:21, Joseph Ciarrochi wrote:

>I like the idea of adaptive search as used in rybka (neutral setting in worse
>position, optimisitc search in better position (search terms defined at bottom
>of email)...
>
>
>I wonder if it is possible to expand adaptiveness into situations where the
>position is even, but one side has many more attacking chances wheras the other
>side has more material. (typical gambit situation). It seems like you should
>engage in pessimistic search if you have taken the gambit pawn and optimtimistic
>search if you have given the gambit pawn.
>
>To take this a step further, there are somewhat unsound gambits, where you sack
>a bit too much, but you do have more attacking chances than your oponent. The
>Cochrane Gambit is probably an example of this (C42: 1.e4 e5 2.Nf3 Nf6 3.Nxe5 d6
>4.Nxff, kxf7,5 d4)
>
>in these instances, adaptive would be working against you, because the computer
>would see that you are losing and stay in neutral instead of shifting to
>optimistic in order to find some winning chances.
>
>I'm pretty sure humans shift to a more optimistic style of search when they are
>playing the gambit, even if their position is negatively evaluationed overall.
>
>So...it would be fun if a computer quickly evaluated attacking chances,
>independent of material
>
>maybe you could just calculate : overal evalution - a pure material count. big
>numbers would indicate a position that has compensation for loss of material .
>the side with more material may need to engage in pessimistic search , wheras
>the side with more attacking chances may need to engage in optimistic search
>
>what do you think?
>
>
>best
>joseph
>

I am thinking about expanding the possibilities for adaptive searching in the
next Beta (12). There is currently some interesting data which suggests that:

1) The weaker the opponent, the better the optimistic settings perform
2) The shorter the time control, the better the pessimistic settings perform

Both of these hypotheses do also make some sense.

Unfortunately the adaptive setting which exists in Betas 10 & 11 seems to be
nothing special, although I still want to get a bit more data.

Vas

>
>
>
>terms defined (taken from rybka help file)
>
> Pessimistic values spend more time looking for ideas for the opponent - or, in
>other words, verifying that the intended move does not run into an unpleasant
>surprise. Optimistic values spend more time looking for one's own ideas - or, in
>other words, looking at alternative moves in search of an improvement. The
>adaptive setting uses a neutral search in worse positions and an optimistic
>search in better positions - you could say that it is the most human of the
>settings



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.