Author: Nobuhiro Yoshimura
Date: 17:07:53 04/06/00
Go up one level in this thread
On April 06, 2000 at 02:21:58, Will Singleton wrote: >On April 06, 2000 at 00:45:28, Nobuhiro Yoshimura wrote: > >>On April 05, 2000 at 20:57:30, Robert Hyatt wrote: >> >>>On April 05, 2000 at 20:54:52, Nobuhiro Yoshimura wrote: >>> >>>> I am writing a Japanese Chess program using PVS with iterative deeping and >>>>hash tables. I often get a fail-low at the root node at the end stage of >>>>the game. I would like to know how offten do you get fail-lows in >>>>searching the chess game tree and also what kind of bugs are possible to >>>>cause many fai-lows. >>> >>> >>>I don't quite understand why you think it is a bug. IE sometimes when you >>>go a ply deeper, you discover some threat your opponent has that you simply >>>can't defend against, and the score drops. It is not abnormal... >> >>I just thought that since PVS is a good searching algorithm and is >>being used by many chess programmers that it would not produce >>many fail-lows that are very bad for search efficincy. > >I'm not sure if you understand this or not, but a fail-low or -high is simply a >returned score that's outside your current search window. The benefit from >using a narrow search window outwieghs the cost of researching after the >failure. > >Having said that, let me say that I use PVS, but I don't use a narrow window at >the root. I always use +- inf, so no fails or researching is necessary. My >testing shows that PVS doesn't benefit that much, if at all, from the narrow >window. > >Will Thanks. I have tested with wider window range in the root node in my program. The results are that it differs +-10% in searching time( sometime good and sometime bad ). Do you think thse results are reasonale?
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.