Author: Ross Boyd
Date: 16:40:22 03/17/05
Go up one level in this thread
On March 17, 2005 at 17:10:18, Jan K. wrote: >Hi,last week I played for couple of minutes with this position. I've tried 3 >replacement schemes:replace-always, that I use in Pseudo because I'm lazy to >write something better, replace-if-deeper and two smaller tables combined >approach(idea by K.Thompson??). Replace-always was _much_ better than the other >two, it went to depth 35 in 1 sec, other schemes got stuck somewhere at ply 25 >after 1 second, consuming lots of time compared to replace-always. It could be >improved by prefering beta(or more probably alfa, I don't remember:) ) entries >in the hash table, but still nothing extra. So I wonder if I had a bug or this >position is special somehow. ;) Btw, pawn hash table has nothing to do with >this, I think. Hi Jan, Interesting. I found the same thing. In this type of simple position with many transpositions, search depth gets very limited when you use depth-preferred replacement only. I use two slots (currently), a depth-preferred and an always replace. This helps a little to improve the depth. But I discovered a crude trick to 'unblock' the tt. I allow depth-preferred entries to be overwritten under certain conditions... for example, if the nodecount exceeds an arbitrary number and the root piece count is very low. OR even more crude, when the max_iteration_depth > 14. Try it, you'll zoom past ply 25 on the way to 35. But again, its really only useful for these very simple blocked pawn positions where transpositions go ballistic and so only a relatively few slots are getting probed and the tt data can get stagnated. Cheers, Ross
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.