Author: Dieter Buerssner
Date: 12:06:57 04/14/04
Go up one level in this thread
On April 14, 2004 at 14:00:47, Dann Corbit wrote: >Are there any PVS style programs that store both an upper and lower bounds (and >I imagine they will also need two depths as well.)? > >I am curious to find out if there is value so that I can write a very general >program where I can switch algorithms at will. I recently tried it. But I did not get it to work well. I had hoped, that especially in fail high/low situtions it might help. Also, I thought a depth preferred hash replacement scheme would work well with 2 bounds (and depth). Indeed, in some situations it seemed to give a bit smaller trees (perhaps even in general). But in other situations, trees got larger (I really cannot explain it). I suspected some bugs, reviewed the code many times, even investigated some really huge search tree dumps (with all the hash info inside). Didn't find anything. Just did not work as I expected. It was not easy to test. Some subtle things seem to be important. Like the problem we already have with repetitions (that are not handled correctly by HT in principle). With two scores, that problem seemed to be worse (I saw it naturally only in late endgames). I used one entry for upper bound values, and one entry sharing exact/lower bound. With only one entry, I also store moves in case of upper bound values (I guess most engines don't). It seems to help my program. With 2 scores, I still only left one slot for the move. I mainly tried only to store moves from lower bound/exact values. But also experimented a bit with storing moves from upper bounds, when the depth was bigger than the lower bound. Many variations - nothing really successful. Regards, Dieter
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.