Author: Ernst A. Heinz
Date: 07:43:28 05/15/00
Go up one level in this thread
Hi Brian,
>Tinker's q-search lazy eval (material - max_pos_score)> beta return (like
>standing pat?), and (material + queen+pawn) <= alpha return (quasi-futility?)
>was working fairly well, until a K vs KBN endgame. Since it was not getting to
>the eval function, the special mating code was never used and Tinker ended up
>with a 50 move draw...Now I check for opponent's pieces <= one minor and skip.
>
>This was discussed in ICCA Vol 21 # 2 (Extended Futility and Dark Thought). I
>also have tried various regular futility, extended futility and razoring (as
>outlined in the ICCA article), but they did not seem to help, at least given
>Tinker's mix of searching algorithims.
Most types of forward pruning which work well in the middlegame and
early endgame often backfire in situations with reduced material.
From my personal experience, the best remedies for this are:
1. less aggressive pruning in situations with reduced material,
2. more sophisticated material-balance scoring that adapts to the
peculiarities of endgames with reduced material, and
3. the integration of perfect knowledge (e.g. interior-node
recognizers, knowledgeable endgame databases) into the search
as described in Part II of my book "Scalable Search in Computer
Chess" (http://supertech.lcs.mit.edu/~heinz/node1.html).
=Ernst=
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.