Author: blass uri
Date: 07:32:05 11/27/99
Go up one level in this thread
On November 27, 1999 at 09:43:10, Heiner Marxen wrote: >On November 27, 1999 at 09:22:23, blass uri wrote: > >>On November 27, 1999 at 08:51:43, Heiner Marxen wrote: >> >>>>How many nodes(legal moves) does your program search per second? >>> >>>Not even 20K NPS on a P/166, so CHEST is not a fast searcher. >>>But at the same time it can do 1500 "mate in 2" per second. >> >>Simple calculation shows that it needs only 20000/1500=13.333 nodes to find if >>there is a mate in 2. >> >>I do not understand it. >>only generating all the legal moves in the first ply is more than it and I do >>not understand how can you avoid it in cases when there is no mate in 2. >> >>Even if there is a mate in 2 and you are lucky to guess the right move by first >>try if you check all the legal moves of the opponent you usually have more than >>13.333 legal moves. >> >>I can understand saving nodes in the last ply when you need only to generate >>moves that threat the king but how do you do it before the last ply? > >For the last ply, where a mate move is needed, I have a special move >generator, which not only restricts to checking moves, but also estimates >the attacks around the king, and omits moves which leave obvious escapes >for the king. > >Then, generally the defender is eager to say check, before there is left >a "mate in 2". Therefore the attacker himself is more often in check, >than not. This obviously reduces the number of legal moves to consider. > >Then, for the defenders move inside the mate in 2 (after which we search >a direct mate move), there is a tricky logic to detect a "fatal anti check". >I.e. if the defender here can check, in most cases we can easily detect >that no mate move can follow, since the attacker must go out of check, >which severely restricts his chances to deliver a mate. As a consequence, >the defender move need not be executed. > >I´ll stop here (there is even more). Such considerations make CHEST fast. >Most of them are not exactly trivial. > I understand that the general idea is that Chest can see by evaluation that there is no mate next move. 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.