Author: Pham Minh Tri
Date: 05:11:33 08/03/01
Go up one level in this thread
On August 03, 2001 at 03:35:42, Tony Werten wrote: >On August 03, 2001 at 03:27:33, Tony Werten wrote: > >>On August 03, 2001 at 01:08:22, Pham Minh Tri wrote: >> >>>Hi all, >>> >>>I saw that my weak engine quickly lost some games or missed some other victories >>>by not recognising many common combinations. I have written codes for detecting >>>them and tried many extensions. Unfortunately, they made the search tree grew >>>exponentially (2-6 times slower) but did not help much, many combinations were >>>still 4 or 5 plies deeper than the full search deep (8-9, some branches are >>>extended to 13-15). The situation was even worse when a combination did not make >>>any check or capture before mate. >>> >>>Could someone give me some suggestions/comments? >> >>Well, at least a useless one: You have to search deeper. >> I could search deeper in normal games, but 8-9 are usually in blitz games. However, some my friends have good knowledge to against computers, they could prepare some traps from 12-14 plies before, when my engine fall into their traps, almost no hope (or serious lose material). >>Be carefull with extensions though. You have to limit them. The idea behind most >>extensions is to find something that otherwise would be hidden behind the >>horizon. >> >>A threat extension is a clear example of this (after heavy fail low on nullmove >>). If you do nothing you'll be checkmated, so extend. But do you really want to >>do that when there are still 8 plies of search left ? Those 8 ply should find >>the correct continuation, not your extension. Yes, I have that extension (I learn and implement all kind of extensions from Crafty, and invent some more), but still lose. In some cases, I need only 1 or 2 plies extensions. I know that buying a stronger computer could help to solve some, but that is not systematic. >> >>Tony > >BTW, you do have a quiescence search do you ? > Yes, I have. My engine has most functions as a normal one. > >> >> >>>Thanks in advance. >>>Pham
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.