Author: Dieter Buerssner
Date: 08:11:36 08/08/03
Go up one level in this thread
On August 07, 2003 at 23:25:45, Russell Reagan wrote: >I am a little unclear about what you mean here. Could you explain what you mean >by "older searches"? Do you mean transposition table entries, or something to do >with the results of the aspiration search, or something completely different? In this context, I meant the researches after zero-window fail high compared to the zero-window search. When you research with -alpha as upper window bound, you come to the following situation: Your first search proved, that score >= alpha+1, while you next search proves, that score <= alpha. To contradictiong things. Which should one trust? Similar situations can happen at other points. Say a fail high at root, followed by an immediate fail low in the research. Which search was right? This is connected to the way, transposition tables are used in detail. For example, when you adjust both alpha and beta when suitable, it will happen less likeley. Whether it is better is not totally clear. I have seen now and then a fail high and fail low in research at the same depth, that then really was best at the next depth. When I watched Shredder, I never have seen, that it ever discards a fail high of zero window search. Shredder as an UCI engine will report the "fail high score" as lower bound first, with "++" in the display. Later it will come up with a real score (without the "++"). Sometimes, it is just equal to the fail high score (which is typically one centipawn more than the last best at the root), sometimes it is higher. I never saw it lower. I don't know, whether other commercial engines display lower bound scores at the point the fail high at the root was found. But the situation, that a fail high was followed by a fail low at the same depth never caught my eye in analysis posted here from commercial engines, while one can see it now and then in the analysis of amateur engines. Regards, Dieter
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.