Author: Don Dailey
Date: 10:24:19 06/23/98
Go up one level in this thread
On June 23, 1998 at 12:45:51, Bruce Moreland wrote: > >On June 23, 1998 at 09:21:58, Robert Hyatt wrote: > >>Dave Slate always said "don't store draws". Ken Thompson said "store >>everything". I go along with the latter... > >I store draws but not at the exact point that they occur. > >You can write something that will very carefully make sure that it doesn't ever >store a 0.00 in the transposition table. > >I don't think this is necessary, and it doesn't do anything about cases where a >draw score caused the value of a node to change score, without changing score to >the draw value. > >But on the other hand, what possible good does it do to store a 0.00 when the >reason you are storing that value is that you *just* detected a rep? When you >encounter this node later, I suggest that it is better to be able to check for a >rep again, and if there is one, return 0.00. If there isn't, then you avoided a >mistake, because the value of the node is almost certainly *not* 0.00. Same >with 50-move conditions. > >But yes, to go through and dissect out all of the 0.00's, that is a mistake I >think. > >bruce I try not to store draw scores, but I do store scores of zero. In my program a draw is a unique score that cannot occur normally. All normal positional scores are even numbers and a draw score is an odd number, usually -1 from the computers point of view unless I have reason to change the contempt factor. This doesn't solve all the problems associated with this phenomenon, perhaps it doesn't even help much at all, but I believe it is at least a small improvement because scores close to zero are the most common ones of all and I would hate to throw them all out. With a parallel program I cannot use the same tricks like using sticky bits for move history and calling hash table matches with these entries draws. So in the hash table I never store a direct draw score. - 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.