Author: Sandro Necchi
Date: 22:40:16 02/18/05
Go up one level in this thread
On February 18, 2005 at 19:19:31, Uri Blass wrote: >On February 18, 2005 at 18:52:58, Arturo Ochoa wrote: > >>On February 18, 2005 at 18:12:18, Sune Fischer wrote: >> >>>On February 18, 2005 at 13:29:56, Arturo Ochoa wrote: >>> >>>>>Yes but what is "help a lot"? >>>> >>>>Look the answer: 30% of the total score reached by Diep in testings and 25% of >>>>the total score reached by Zappa in private tests. The books was responsible of >>>>30% and 25% of the score reached for every mentioned engine. >>> >>>I'm not quite sure what that means actually. >>> >>>If your score is 20% and you improve that by 30% you score will be 26% which is >>>a rating increase of 59 Elo. >>> >>>If the score improves 30% from 50% to 65% it's a 107 Elo. >>> >>>If the score improves by 30% from 35% to 65% it's 240 Elo. >>> >>>If the score improves by 30% from 60% to 90% it's 320 Elo. >>> >>>-S. >> >>These assumptions are absolutely wrong. It is a common problem in this Forum of >>asserting things that I have not said. >> >>"Look the answer: 30% of the total score reached by Diep in testings and 25% of >>the total score reached by Zappa in private tests. The books was responsible of >>30% and 25% of the score reached for every mentioned engine. >>I'm not quite sure what that means actually." >> >>Example: If Diep played 10 games, and it won 10 games, 3 games were because of >>the book. Do you understand? A direct win because of the book. >> >>AO-- > >It means that diep scored 10/x in your testing with book when 7/x was without >book when x is unknown. No, not necessarely. We weed to define what one means to the book first. It could be thanks to the book even if the program does not come out of the opening with a winning advantage: i.e. a small plus but the program play very well out of that position. If the program got a position that handles very well with book x, than book x did very well and has part of the win merit. > >Of course even if we know x it cannot mean nothing about rating points because >the condition of your testing have to be different than the condition in the >tournament. To make a generic statement that a good book can give up to 700 Elo is extremely vague. A good opening book will give a different plus depending if we are using the program against humans, on the program strenght, on the hardware used, against a specific opponent, against computers, on a specific no. of games and if selection is not automatic but guided by the operator (I mean which variation to play against a specific opponent rather than let the program choose; of course this selection should be made before the games start). > >In tournaments part of the opponents are not passive target and it is possible >that your plan against version X does not work for the simple reason that the >author upgraded to version X+1. This is true as computers, more than humans will not play the same moves of a previous version with an upgraded program and/or different hardware. This is why it is very difficult and dangerous to prepare specifically against a specific opponent. Even if someone told I did, I never did. On the contrary I did improve some lines the opponents where playing agaist us, but only to remove some of our weaknesses. It is different. I mean if my car is slow compared to the competition, by making it faster I do not prepare against a specific opponent, but I remove/reduce a weakness. The lines in the book were made to play against anybody, and not a specific opponent or more than one. I remember when we won the WCCC in 1995, with MChess Pro. 5.0, that the last game of the 2 games final match Genius used a specific prepared book against us, but our program played the same moves until move 20th, if I rememeber correctly, and did improve the line after that winning the game. > >Not easy to predict how many prepared direct wins you will lose because of that. Yes, the opening book is a very complicated matter. It takes a huge amount of time to work on it and is never finished. Also, no matter how good you think it is, you will always find someone complaing or criticizing it. To criticize is very easy...to make things not so. So going back to a good opening book it is necessary to define first what one means by a good opening book and for which it has been designed. > >Uri Sandro
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.