Author: Don Dailey
Date: 14:51:23 05/11/98
Go up one level in this thread
On May 11, 1998 at 03:48:03, Ulrich Tuerke wrote: >On May 08, 1998 at 14:20:58, Don Dailey wrote: > >>... >>I've been considering this type of approach for a long time. I think >>it definitely has merit. It would suffer from hash table anomolies >>but there are ways to deal with this, the primary one to simply ignore >>them. > >I see what you mean. The score of a position would also depend on the >iteration (and not only on the distance to the "horizon"). Yes, a way >out could be (a) ignore (?) or (b) init your hash table at start of a >iteration (which seems cleaner but looks a little bit unfortunate cause >of performance loss). > >> >>There are a few issues to be dealt with but my feeling is that you >>might do well with this scheme. Scoring might be a little fuzzy at >>times but that might be ok. Probably the scheme would be to >>pre-process at every level up to MAX_DEPTH-n, where n might be at >>least 3 or 4 ply. Keep in mind that pre-processing is incredibly >>expensive even compared to fully dynamic processing, so I suspect you >>cannot get very close to the leaf nodes. Of course 3 or 4 ply away >>might be considered pretty close if you were doing 14 ply searches. > >Yes, that's probably more reasonable than at half of the search depth. > >>... > >Thank you very much indeed for your detailled reply, Don. Certainly, the >idea is quite natural. May be, other programmers have played with it or >are using it. Your remarks sound quite encouraging. May be I'll soon >give this a try. >Regards, Uli I think others programmers are definitely using it. Amir pretty much said that he was. I just wrote a little chess program as an example for our next Cilk language release. It is FULLY pre-processed. I did this because it was easy, and also because I wanted to do some experimentation with pce/square tables again. I haven't written a preprocessed evaluation for a few years now. I will definite try this idea and get back to the group. Hopefully Amir and others who probably already know what to expect will pitch in and give us their insights. - Don
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.