Author: Uri Blass
Date: 06:57:44 01/05/06
Go up one level in this thread
On January 05, 2006 at 09:06:20, Tord Romstad wrote: >Hi Kirill, > >I agree that testing without books or with generic books can be just as >interesting as testing with the engine's own books (and indeed I run all >my own tests without books). However, I disagree with almost all the >reasons you give for prefering to test with the engine books: > >On January 05, 2006 at 08:02:37, Kirill Kryukov wrote: > >>1. Opening book quality is directly proportional to the amount of money invested >>into the chess program. Engine strength, on the other hand, is proportional >>mostly to the talent and efforts of the programmer. > >No. Engine strength is also to a large extent proportional to the amount >of money invested. If I had enough money, I could quit my job and work >on my engine many hours every day, instead of just an occasional hour on >late nights or in the weekends. I could buy a dozen computers to help me >run automated matches very quickly. Instead of clumsily attempt to program >in a low-level language which I don't really know at all, I could write >prototypes of my program in Lisp and hire a C or assembly language wizard >to do the dirty optimisation and low-level work. All of this would give >a *much* bigger strength boost to my engine+book package than if I hired >some opening book expert to work on my opening book. > >Writing a top chess engine isn't just about talent. It also takes lots >of time; probably too much time for the average hobbyist. Of course >talent helps, but I don't think this is any less true for book creation. >I know that I could never make a good opening book, no matter how much >time or money you gave me to do it. I think that you are wrong about this point. I also do not find it interesting to invest a lot of time on good book but I think that time of testing is clearly enough for writing a good book for comp-comp games. If you buy many computers and run tests against other engines with different lines you can easily find the lines that your engine score better in them and put them in your book. You only need a program to analyze the result of your games to decide which lines to put in your book. Example for white You may have the following lines and results against a lot of engines. 1.e4 e5 2.Nf3 score after 1000 games 60% 1.e4 e5 2.Nc3 score after 1000 games 55% 1.e4 e6 2.d4 score after 1000 games 62% 1.e4 c6 2.d4 score after 1000 games 60% 1.e4 c5 2.Nf3 score after 1000 games 40% 1.e4 c5 2.Nc3 score after 1000 games 45% 1.e4 c5 2.Na3 score after 1000 games 70% 1.d4 d5 2.c4 score after 1000 games 55% 1.d4 d5 2.Nf3 score after 1000 games 54% 1.d4 Nf6 2.c4 score after 1000 games 50% 1.d4 f5 2.Nf3 score after 1000 fames 50% If you need to decide based on the information which lines to put in your book in the first moves for white then you can decide to put in the book only 1.e4 and in the second move in the lines that I mention to put in your book only 1.e4 e5 2.Nf3 1.e4 e6 2.d4 1.e4 c6 2.d4 1.e4 c5 2.Na3 Of course with more computers you may collect more information and play millions of games and you get very good choice of the first moves in order to maximaize the results of your engine. Note that this idea does not include killer moves and only try to improve your book by better choice of the first moves. You can also improve your book by having many killer lines against other engines to get mate out of the book. I do not think that it is interesting to do it and I know of no commercial program that does it but I believe that with enough time of testing it is possible to get more than 100 elo improvement in the ssdf list by that method. 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.