Author: Vasik Rajlich
Date: 03:14:17 06/18/05
Go up one level in this thread
On June 17, 2005 at 21:09:49, Antonio Dieguez wrote: > >>>Yes. However I remember what Ernst Heinz said to me once: "fail hard läuft >>>einfach besser". Seems not really logical - but I guess, he has a point. Perhaps >>>the point is, that it is much easier, to have some subtle bug in fail soft >>>search. alpha-beta is not really forgiving here - it will not crash with such >>>bugs, just not perform as good as it should (perhaps only in few positions). >>> >> >>Yeah, that's a pretty mysterious statement. I see no reason to ever fail hard. I >>don't even see how fail-soft is more bug-prone. > >one of the changes i did in last versions of my engine was changing from >fail-hard to fail-soft, that's partly why i bugfixed it twice (besides >incompetence..) because i wasn't carefull when returning a score below alpha. > >The closer you are from something like, for example: > >SortMoves(); >then loop in a for trough the moves > >the harder to have a bug, but if you have something more bloated or bad coded >you could not even see that in some place you aren't setting the score below >alpha or doing it incorrectly. When discarding a move you want to set its score >the lower possible too, so you have to do more work. May be is a basic mistake, >but that would do for being "more bug-prone". >Just my 1 cent. I could be absolutely wrong as always. BTW - normal fail-soft doesn't mean doing more work to get a softer value. It just means returning it if it happens to exist. I have some ideas for an "ultra-fail-soft", where you would voluntarily do more work in some cases to get a softer value. Combined with MTD (f), this could potentially combine the strong points of PVS and MTD (f). I can try to explain this in more detail but at the moment I don't think there is a huge potential benefit there. Really the main thing in a chess program is the eval, and maybe some selectivity in the search .. Vas
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.