Author: José Carlos
Date: 07:51:28 12/30/01
Go up one level in this thread
On December 30, 2001 at 10:19:59, Gian-Carlo Pascutto wrote: >On December 30, 2001 at 10:12:49, José Carlos wrote: > >>a) there are tricks to save space > >True > >>b) it only applies to depth-preferred table, so crap nodes (which are quickly >>discarded) don't stay long because of depth. And there's also always-replace >>table. > >But if you never adjust the depth, this does not address the issue >of positions that are totally different hanging around because of >equal material. I think didn't say it all :) I use the abitual depth criteria. If the position stored in the hash table has depth<= than the actual position I replace it. But for old positions, from previous searches, I _know_ a position is not useful anymore when it has more material than the root position for this search. So useless positions, which are not search very deeply thanks to null move, never get stored with large depth, and so, even if material is valid, they are replaced due to depth criterium. >>c) block positions usually get open by pawn exchanges, so material decreases and >>crap nodes get replaced. > >I would say that blocked positions are just those where this isn't possible :) You know what I mean ;) If I have a possible pawn break that appears to be bad, and you can do nothing to open the position, we'll be moving pieces around until 50 moves rule is near. During that time, the positions in the tree searches will be the same search after search, so they are useful to keep. Once we get to 50 moves, if I think after the pawn break I'm up by 0.01, I'll push the pawn, you'll take it and all the very deep hash entries I have in my hash table will become ready to be replaced because of material criterium. >> Also, I only wanted to be useful by commenting my idea. All comments are >>welcome. > >Sure, and it's appreciated. Don't think me attacking your idea means >that I don't like your comments! Thanks. >-- >GCP José C.
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.