Author: Uri Blass
Date: 02:44:37 11/16/04
Go up one level in this thread
On November 16, 2004 at 04:47:52, Joachim Rang wrote: >On November 16, 2004 at 04:38:57, Günther Simon wrote: > >>On November 16, 2004 at 03:16:35, Joachim Rang wrote: >> >>>On November 16, 2004 at 01:46:47, Kurt Utzinger wrote: >>> >>>>On November 15, 2004 at 15:42:01, Albert Silver wrote: >>>> >>>>>On November 15, 2004 at 13:37:01, Kurt Utzinger wrote: >>>>> >>>>>>Nunn2 match (60m+10s) 40 games >>>>>>Pro Deo 1.0 vs Fritz 8.0.0.26 >>>>>>http://www.utzingerk.com/pro_deo.htm >>>>>> >>>>>>A little test should show the influence of the option [clear hash table = >>>>>>always] and [clear hash table = never]. Below the result against Fritz 8 as Pro >>>>>>Deo had the worst result against this engine: >>>>>> >>>>>>Option [Clear Hash Table = Always] >>>>>>PD = 32.5 % >>>>>>13.0 - 27.0 >>>>>> >>>>>>[Clear Hash Table = Never] >>>>>>PD = 36,2 % >>>>>>14.5 - 25.5 >>>>> >>>>>I ran some Nunn2 blitz matches as you know, with and without, and the results >>>>>were not clear. Against Aristarch and Ruffian the results were better, and >>>>>against List, PD scored 1 point less. >>>>> >>>>> Albert >>>> >>>> >>>> Hi Albert >>>> Is it worth to replay all remaining 280 games >>>> at time control 60m+10s with clear hash table >>>> = never http://www.utzingerk.com/pro_deo.htm >>>> Kurt >>> >>> >>>why should clear Hash Table = Never be an improvement? ProDeo looses important >>>information when it clears the hashtable? Why is this option tested? Did Ed >>>mention that he is not clear whether it helps or not? >>> >>>regards Joachim >> >>Hi Joachim, >> >>You sure meant 'clear hash table = always' in your statement above? >>The reason for not using that option is that ProDeo seems to have some >>kind of hashtable bug and sometimes plays weird blunder moves from >>the hashtable. (users reported such blunders here) >>I have experienced such blunders too in RWBC so far, nevertheless I still >>use 'never', as Ed suggested this in relief that the overall success >>should be better, even with some occasional big blunders. >> >>Best regards, >>Günther > > >Yes Günther you are right. I confused Never/Always. The only think I can imagine >clearing the hashtables should help is if there is a bug, like you Uri and other >mentioned. I didn't know that there are suspicions of such a bug in Rebel. > >regards Joachim You should always suspect about bugs. I suspect that the top programs can be more than 100 elo better if the programmers only fix the bugs and change the personality. Part of the problem is that the programmers do not know where are the bugs and sometimes bugs can be even be productive for playing strength unless you change the personality(for example if a program does not prune enough some bug of pruning more lines based on conditions that are irrelevant for pruning may be productive because the bug may mean that the program at least prune enough in part of the cases when the bug increase pruning). I know it because it is exactly what happened with movei when the program played worse because of fixing some bug and it took me time to understand the reason. After changing parameters the program seemed to play better again but I did not dare to change the pruning paramaters enough so I gave beta testers to test possibility of more pruning and compare with the latest version that I tested and result so far are very good see http://www.husvankempen.de/nunn/ratingaegt1511.txt 14 Movei 00_8_291a : 2640 52 42 161 54.0 % 2612 31.1 30 Movei 00_8_291 : 2578 68 55 100 52.5 % 2561 Note that the only difference between movei00_8_291a and movei00_8_291 is that 00_8_291a has more pruning based on evaluation and other factors when the remaining depth is small. 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.