Author: Daniel Clausen
Date: 13:41:27 12/09/99
Go up one level in this thread
Hi On December 09, 1999 at 16:30:29, Dann Corbit wrote: >On December 09, 1999 at 16:21:22, Daniel Clausen wrote: >[snip] >>>How are collisions handled? Do you compute the value for the current key and >>>replace or??? >> >>Depends: >> >>Storing a value into the HT: >> If the depth of the position in the HT is less then the current depth, I >>overwrite. >How do you know that the deeper one is more important than the shallow one? It >seems to me that a shallow one might be examined again and again, and a deep one >might be some chance leaf search of little value. I should have been more specific. You usually overwrite an entry, if the stored position is deeper in the tree than the new position. So positions closer to the root-node have a higher priority. The reason for this is, that a successful HT-probe closer to the root saves you more work than a hit deeper in the tree. >Has anyone tried using an equivalent algorithm to a LRU cache for hash values, >but instead of flushing to disk, just discarding the unused entries? After a search, most programs set a 'dirty' bit for all occupied slots in the hash-table. The next time you wanna store something in the hashtable and you see that there's already a position stored with the dirty-bit set, you overwrite it, regardless of the depth. I think this is a lil bit like LRU. Kind regards, -sargon
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.