Author: Dave Gomboc
Date: 00:29:23 01/20/03
Go up one level in this thread
On January 18, 2003 at 22:32:29, Robert Hyatt wrote: >On January 18, 2003 at 20:40:25, Dave Gomboc wrote: > >>On January 17, 2003 at 17:27:22, Dieter Buerssner wrote: >> >>>On January 17, 2003 at 17:08:11, Robert Hyatt wrote: >>> >>>>Remember that the mate scores in the search are "mate in N from the _root_ >>>>position." When you store a mate score in the transposition table, you have to >>>>correct it so that it is "mate in N from the current ply". Once you do that, >>>>you >>>>can use the scores easily as when you get an EXACT mate score from the table, >>>>you know it is mate in N from the current ply, so obviously it is mate in N+ >>>>something from the root position. Adjust it correctly and you are done. >>> >>>I agree - of course. >>> >>>>You can see how I do this in crafty if you look at hash.c... >>> >>>Last time I looked, you were just throwing valuable information in this regard >>>away. Crafty didn't store mate bounds due to the method you explained above, but >>>rather stored all mate bounds as some "Mate in very many moves". I am convinced, >>>that the adjustment to a mate score from root cannot be worse, and often will >>>ahve advantages (more cutoffs, faster search, ...). >>> >>>Regards, >>>Dieter >> >>I thought Bruce Moreland recommended this once, for non-PV cases where the root >>score is much less than mating (the idea being that since the line is just going >>to get cutoff anyway, why do the adjustment work for nothing?) >> >>Dave > > >I'm doing this once again. The point is that it can make the tree smaller >if you are searching thru a forest of varying depth mates... For clarity: you are updating the bound, or you are using a fixed bound? On ICC during CCT Dieter mentioned to me that he thought updating the bounds, if done properly, was never a loss and sometimes a win, which would contradict Bruce's idea -- that is, if I remembered Bruce's idea correctly to begin with. Dave
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.