Author: Tord Romstad
Date: 04:02:17 05/05/04
Go up one level in this thread
On May 04, 2004 at 18:35:30, Dieter Buerssner wrote: >On May 04, 2004 at 10:39:12, Tord Romstad wrote: > >>Not really. The only thing you test with this kind of artificial position is >>whether the engine has checks in the qsearch. > >Perhaps, perhaps not only. Yace does not use checks in qsearch (at least not by >default) and will find it. But needs longer, 30 seconds. I have until now only >tried checks in first level qsearch (not in remaining levels). I tested soem >variations - like using check in qsearch only when directly coming form null >move, always, and few more. I never saw a significant difference (besides in >test positions, some are solved many plies earlier with checking moves in the >first call to qsearch). I always handle checking moves that capture in first >qsearch level somewhat more careful. For me, checks in qsearch are definitely effective, even in real games. I think there are two main reasons for this: 1. My engine is one of the slowest in the known universe, and needs to compensate somehow for its low N/s count in order to avoid being horribly outsearched. 2. The wild playing style of my engine often results in sharp kingside attacks and tactical complications, exactly the kind of positions where checks in qsearch are likely to help. Of course, I don't search all checks everywhere in the qsearch. I consider the path leading to the current position, the static eval (I don't search checks if the static eval is clearly winning, even if it is less than beta), the opponent's king safety, the material situation, and probably a few other factors I don't remember right now. In order to prevent search explosions, I also avoid searching checks if the number of qsearch nodes since the engine left the main search exceeds some threshold (currently 500 nodes). Not surprisingly, this scheme causes lots of search inconsistencies (I hash all qsearch nodes). I don't like this, but I've decided to try to live with it until I find a good solution. So far, it doesn't seem to be a big problem in practise. >What I wanted to say - just extending many checking moves in a row aggressively >should help to solve such a position. No special qsearch will be needed then. No >doubt, checks in qsearch will also help. Yes, you're right. Tord
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.