Author: Brian Richardson
Date: 15:05:20 10/18/00
Go up one level in this thread
On October 18, 2000 at 03:17:48, Pham Minh Tri wrote: >Could anyone explain how to detect and play endgame? What is different between >playing middle and end game? > >Thank in advance for any help. > >Pham This is an area that every program seems to do a little differently. I am still experimenting with the transition to endgame phase in Tinker. Right now I am using piece (excluding pawn) material greater than 4 minors, or 2 rooks and one minor or queen and one minor to say not endgame. My own chess is not that good, but I think a 4 minor mate (in a reasonably even game) is unlikely. I also have numerous evaluation terms that are scaled down by the opponents piece material (to become less important--like king safety). It is relatively cheap to incrementally update piece subtotals by color. I am also experimenting with a "signature" method in Tinker. It works fine for all "normal" combinations (up to 3 queens, and 2 of everything else, ignoring pawns). It breaks with underpromotions (duplicate signatures) and does not seem any faster than incremental material updates (since the signature must also be incrementally updated). The signature may have other uses (limiting null moves, fast test of EGTBs present, etc). I was quite surprised at how little overhead "extra" incremental updates added (much less than 5% in most cases). It certainly appears as though move ordering and the search (hashing, nulls, killers, etc) are far more important than raw move generation efficiency.
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.