Author: Ernst A. Heinz
Date: 06:14:21 05/21/98
Go up one level in this thread
On May 21, 1998 at 08:29:51, Robert Hyatt wrote: >I see two problems to solve. First you can get a hash hit, but no >suggested >move... Just don't store entries without a suggested best move in the table. >because this is a PV node and you have no PV (hash) move to follow, you >have >to take some evasive action to avoid calling LookUp() in the iterative- >deepening call to Search()... or else you have to defer storing the >"open" >hash entry until you pass the IID part... >I tried it both ways... and didn't like the complexity... Right, we simply defer flagging the entries in the table until after the internal iterative deepening part. I don't think this is more complex than handling an extra move stack for repetition detection but there is surely room for disagreement here ... >I tried it both ways... and didn't like the complexity... Ken Thompson >used this however, but didn't do IID at all. My experiments offered no >speed improvement over the repetition_list approach (this can be fast if >you implement 50-move draws because you don't have to check the entire >repetition list and it stays very short as well). For "Crafty" this may be true -- for other programs like "DarkThought" the "open hashing" approach is definitely faster. It probably depends on the general memory requirements of the programm -- the more memory you use in other parts of the program the less it will suffer from an additional repetition list. =Ernst=
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.