Author: Uri Blass
Date: 13:16:29 09/13/04
Go up one level in this thread
On September 13, 2004 at 15:49:16, Sandro Necchi wrote: >On September 13, 2004 at 06:21:04, Uri Blass wrote: > >>On September 13, 2004 at 01:25:25, Sandro Necchi wrote: >> >>>On September 12, 2004 at 16:52:48, Sune Fischer wrote: >>> >>>>On September 12, 2004 at 16:43:02, Sandro Necchi wrote: >>>> >>>>>OK, this is true, but think about this: >>>>> >>>>>1. Program A with his own book play at 2650 Elo >>>>>2. program A without any book (may) play at 2600 Elo >>>>>3. Program A with another book can play from 2500 to 2650 depending how suiting >>>>>his style the book is...This is due to the fact that an opening book can make a >>>>>program play worse than without if bringing the program in positions it does not >>>>>like. >>>>> >>>>>so, how can you know how strong is a program if you let it play with a different >>>>>book? >>>> >>>>I can't, that's why I never use books in testing but instead I used fixed >>>>positions. >>> >>>It is the same as we are talking about positions not suiting the program style. >>> >>>I.e.: If a program is not suiting to play the French defence, than it will play >>>that one bad. >>>If you force the position it is the same than to get it from a different book. >> >>I think that a program should know to play all positions well. > >This would be ideal, but so far no one did. Yes but the target is to improve the level of the program in all positions. >> >>I prefer not to test the program with book because I believe that if I improve >>the program in set of positions then it is probably possible to improve it also >>in games by the right book. >> >>I believe that stefan use similiar idea because I rememeber that you said about >>one new version of shredder that it is stronger but the book does fit the style >>of previous version so with the opening book that people buy it is nearly the >>same level. > >No, it is not as you say. > >When a programmer change the program a lot the book may need to be modified as >some positions may not be suiting the program anymore and others may be better. I agree. > >The problem is that to change the program may be relatively quick, but to check >the new potentiality of the program as well new weaknesses takes a lot of time >and even more to change the book partially or a lot. Yes but my point is that you needed to test that the new version that has new weaknesses to know that it has bigger potential with the right book and I think that the way to do it is to see if it can score better in games from positions that are not in the book(if it can score better in matches from predefined positions like the nunn match then it probably has better potential with the right book). > >The only way one can find out about the new possibility is to analyse the >program games and see it when it play what is pondering, expected replies, time >to find the correct move, time to find the correct reply, ability to find >difficult moves, ability to find the correct defence, ability to use the pieces >and so on...when the picture is complete, then it is possible to investigate >specifically. This may be useful in order to build a new book but the first problem that you have is to decide if it is a good idea to spend time on a new book. > >> >>I also think that testing program in many positions that are not in the original >>book can help to build a better book. > >No, see above. > >If you play soccer and you are not tall and I see that you are not good with the >head, it is time lost to test new schemes which involve you trying to make goal >with the head with high coming balls... > >I call this optimizing work/time usage... Yes but in order to know that you are not tall you need to see it in games and when you change the evaluation things may be change so the fact that you were not tall in the past does not mean that you are still not tall. I think that it is a waste of time for programmers to analyze games in order to work on optimizing the book for positions that their program understand better when some months later it may understand better different positions. It may not be the case for shredder because stefan does not work about the book but most programmers have not someone who is going to optimize the book for them and if they need to decide if to spend time about programming or about book then it is better if they spend time about programming. Based on my experience analyzing games by the programmers can be productive to show weaknesses of the program but fixing the weaknesses in evaluation or search is not dependent on book. Uri
This page took 0.01 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.