Author: Dave Gomboc
Date: 16:53:46 08/24/00
Go up one level in this thread
On August 24, 2000 at 18:49:16, José Carlos wrote: > In Averno, I never clear the hash tables. I store the total material (sum of >both sides) with the position, flags and move. So, when I'm going to store a new >posotion, I check if the material stored is bigger than the material at the root >of this search. In that case, I replace always. Otherwise, if the other tests >point me to store the position, I save the actual material (in that node of the >tree) with the position. > I know this could not be accurate when promoting occurs, but works fine most >of the time. > Just curious, why do you guys clear the hash tables and when? I read here some >days ago that Christophe clears before every search in Tiger, but is not going >to do it in further versions. Can you reveal your secret? :) > > Thanks in advance. > > José C. Increment a counter each time a new root position is searched (e.g. every time your program plays a move in the game). Include this counter in the information that is recorded when a position is stored in the hash table. This now allows you to see if a position you encounter in the hash table is from the current search, or how many searches ago it was from. Your counter can be just a few bits long. However, if you have a very large hash table relative to the number of entries that are being overwritten per search, then a too-short counter can lead to confusing some old positions for new ones. 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.