Author: Uri Blass
Date: 10:05:05 02/07/02
Go up one level in this thread
On February 07, 2002 at 12:33:40, Sune Fischer wrote: >On February 07, 2002 at 12:12:19, Uri Blass wrote: > >>1)At some point after enough plies chess is solved so the returns is 0. >> >>2)It is clear that if the program use 0 seconds for book moves then book has >>increasing influence on the result when programs search deeper and there are >>cases that the result is decided by book. >> >>I think that this is the main reason for expected diminishing return. > >I disagree, I believe it has to do with the previous distribution of how often a >better move can be found at ply n+1 relative to ply n. >Once you have found the correct move, searching further doesn't help. >So the question is how often can the best move be found at ply n? I think that the difference at small depthes is small so the main reason for diminishing returns at small depthes is book. > > >>It may be more interesting if there is a diminishing return in the nunn match >>when I believe that the result is not decided by book in these positions(I >>believe that in these cases there is no obvious draw or obvious win for one >>side). >> >>My experiments of the nunn match between fritz and Tiger showed no evidence for >>diminishing returns but I did not got deeper than depth 13 and almost all my >>games were at depth<=10(I did only 2 matches of 13 plies against 13 plies and 13 >>plies against 12 plies and I posted the games of these matches here). >> >>I also posted my results at depth<=10. >> >>Note that I did not see clear diminishing returns relative to small depth and >>the difference in rating in the nunn position between 3 and 4 plies seems to be >>similiar to the difference between 12 and 13 plies and to the difference between >>9 and 10 plies). > >What are these "nunn positions"? These positions are positions that nunn composed to test chess programs without books. These positions are not result of long book lines of 40 plies that often happen in comp-comp games. I used a version of the nunn match that include 25 position and one match include 50 games. >So you saw the same elo jump in 3-4 as in 12-13? >I must say that surprizes me. Maybe I did not do enough games at 13 plies(I have only 100 games that are 40 games of 13 plies against 13 plies and 50 games of 13 plies against 12 plies but I have more games of 10 and 9 plies tiger(10 plies) played against Fritz(10 plies,Fritz 9 plies,Fritz 8 plies....) It was a sweningen tournament and every program played at least 300 games against the opponents I remember that there were total number of 3000 games not including the 100 games that were posted here(13 plies against 13 plies and 13 plies against 12 plies). Uri
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.