Author: Larry Griffiths
Date: 10:56:39 02/19/01
Go up one level in this thread
On February 19, 2001 at 11:35:10, Carmelo Calzerano wrote: >On February 19, 2001 at 09:55:04, Larry Griffiths wrote: > >>On February 19, 2001 at 06:01:36, Carmelo Calzerano wrote: >> >>>On February 18, 2001 at 11:29:39, Larry Griffiths wrote: >>> >>>>Hi, >>>> >>>>I use a reference indicator in my hash entries when the hash key matches. >>>> >>>>Only 20 to 40% of the hash entries are being referenced during a tree search. >>>> >>>>Is this a normal range? >>> >>> >>>Surely not, unless the number of entries in the table is much bigger than the >>>number of nodes you visit in the search (which is usually not the case). > >I have to correct myself: 40% is _not_ a bad value, unless the number of nodes >visited is much bigger than the number of hash entries :-) > >>>Hash entry references must be equally distributed; i.e., if you have a 1 M >>>entries in your hash and search 100 M nodes, you should find each hash entry >>>referred about 100 times (with small statistical fluctuations of course). >>> >>>HTH > >>I used Bob's Random64() function to produce the Hash keys. I also created two >>hash tables so that black and white positions would not overlay each other. >>I even use slots and will do a hash add within the next 7 slots if the primary >>hash slot add fails. > >But what's the ratio between nodes visited/hash size? >I.e., if you have 1 M hash entries and visit 5 M nodes, 20-40% sounds ok. >OTH, if you have 1 M entries but visit 500 M nodes, then 20-40% is >_definitely_ a very bad value... >:-) > >>It may be that positions that caused cutoffs are never seen again because >>succeeding cutoffs occur. > >I'm not sure I understand what you mean Lets say a position causes a cutoff and is placed in the hash table. This entry is added with the reference bit off. This position is never referenced because another cutoff occurs that is better, or a hash entry nearer the root is used and hash entries near the leaves never get seen again. I think I need to get snapshots of the hashtable during the search to see if this occurs mainly after the search has ended. > >Bye, >Carmelo
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.