Author: fca
Date: 05:35:31 08/07/98
Go up one level in this thread
On August 06, 1998 at 23:25:34, Harald Faber wrote: >Ed, could you explain why Rebel9 plays all the moves Rebel10 played in the Anand >game (Rebel black, tournament) except for only one move? This is amazing and >certainly no reason to get Rebel10 when the moves are 99% the same. I know this >question is difficult to answer, you need sales to survive. Don't see it as an >attack but as a reasonable question. Of course the question is valid. 1. Percentage would be (40-4-1)/(40-4) == 97%, *not* 99% (40 moves in game 7, first 4 from R10 book) :-) 2. You cannot know anyway as the times when R10 made the decision to switch to a given best move is not known. Time allocation is complex and never reproducible. permanent braiin and hash considerations make divergence more likely... There was no suggestion R9 would have found all those moves in the time actually had, or would have stuck to them. 3. Why select game 7 when in game 8 the divergence is more? 4. I no more draw conclusions from 1 game for this purpose than I try to evaluate ELO grade from just one game... Else R10 has ELO of 2750 from this game (draw) :-)) ... All 8 games should be considered.... (oof - 2900! :-)) ) 5. Elsewhere Ed has posted statistics showing an overview of analysis of many hundreds of positions, showing (among other things) the frequencies of evaluations being changed by anti-GM by various amounts, or the search times, or the actual move chosen. Summarising, anti-GM does seem to make quite a difference (good or bad is not indicated by these particular statistics :-) ) significantly often. Of course the type of position included in the set is critical, and these are I believe problem positions where you would expect anti-GM to perhaps be more relevant. 6. Anand's play may have been such that the problem-causing nature of anti-GM never got triggered so often as with most IMs or GMs. This is not anti-superGM. 7. R10 won the match anyway, and R9 may have won it anyway on the same hardware is quite a possible conclusion... If good moves would anyway be chosen by R9, why should R10 change them? :-) 8. Also, anti-GM is particularly triggered when the computer thinks it is *not* probably winning anyway (reasoning being: why put the win at risk?). And in this match (i.e. not just game 7), much of the time R10 was ahead in reality and even more often ahead in its opinion (much of game 8 for example) so anti-GM would be auto-suppressed. I will be buying my R10. :-) Kind regards fca
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.