Author: Robert Hyatt
Date: 07:00:46 09/26/01
Go up one level in this thread
On September 26, 2001 at 06:17:38, Sune Fischer wrote: >On September 25, 2001 at 20:06:26, Robert Hyatt wrote: > >>>>It is more likely that >>>>as you search, old things _will_ be overwritten by new things, due to the way >>>>I "age" the hash entries. >>> >>>How you do things in Crafty is a completely different matter. >> >>Why? The question is still the same. "Is bigger always better?" I don't >>really care about inefficient implementations. Or faulty implementations... > >No because if you can age the hash entries in some way, then that means you have >more information about the position than what is in the 64 bit (you know its >age!), which is equivalent in some way to using larger keys. Then I already use larger keys. Because I validity check the move that is in the table for safety. That is beyond 64 bits if you go that far. However, the 64 bits includes the location of all pieces, so this is really redundant. And I haven't seen a move from the hash table come up "illegal" in a couple of years. It only happens when I break something... which means that move is not really pseudo-increasing the size of the hash key... > >Using the standard principle behind 64 bit keys, it will not be possible _ever_ >to update _any_ of the old entries in a 2^64 table, which is why the example >works. > >>>With current kNs and 64 bit keys - then yes. >>>But generally the answer is no, smaller keys or more power would increase >>>collision rate to unacceptable leves with too large a table. >> >>Yes, but not for thousands of years into the future. When are we _really_ >>going to be able to search 16,000,000,000,000,000,000 nodes? (4 billion times >>4 billion)??? Today making the table larger does _not_ appreciably increase >>the chance for collisions. It _does_ increase the search efficiency however. >>The net is a _gain_... > >Yes for the present case, I thought we had this covered. > >>If you want to go theoretical and talk about solving the game, then yes, 2^64 >>is not big enough. The keys must be close to 2^160. Close enough that the >>collisions won't change the root score and screw things up. >> >>But for the question "is a bigger hash table always better?" today the answer >>is a simple "yes, so long as you don't make it so big that you start to page >>and slow down due to disk I/O." Collisions don't play any role in that decision >>whatsoever today... And won't for the next few thousand centuries... > >Ah but professor, you forget that progress in the scientific world is not >linear, it is exponential! >It will not take thousands of centuries to get there, the quantum computer is >perhaps only a decade or two away, don't know if it can do chess, but I intend >to find out;) >I seriously believe that within a 100 years, 64 bit will not be enough. > >-S.
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.