Author: Dann Corbit
Date: 21:53:31 05/28/01
Go up one level in this thread
On May 29, 2001 at 00:34:43, Peter Berger wrote: >On May 28, 2001 at 23:36:19, Dann Corbit wrote: > >>On May 28, 2001 at 20:13:55, william penn wrote: >> >>>On May 28, 2001 at 19:42:37, Dann Corbit wrote: >>> >>>>Play 'em that way if you want to. >>>> >>>>Silly to remove 80% of a program's strength. >>>> >>>>But that's just my opinion. >>> >>> >>> no way, most of your top programs today will make book moves even without an >>>opening book, >> >>There are many openings where the true outcome is hidden 18-20 plies deep and >>you will surely get sucked into a trap without an opening book. The Evans >>Gambit lines are a swell example of that. A computer playing such a sequence >>without a book will get slaughtered by one with a [good] book. >> >>>unless you mean that the program saves time, which makes it >>>stronger, since it can use more time later to calculate the middlegame >>>positions, >> >>That is a minor factor, and would not account for more than 30 ELO or so at a >>maximum. >> >>>still i don't see this accounting for 80 percent of its strength. >> >>At least 80%. Of course, it is just my estimate. But I am very sure of it. > >I think this number is completely bogus - did you do any tests ? I'd be _very_ >surprised if having a decent opening book is worth more than 70 ELO against >having no opening book at all but some basic opening rules and making sure that >the program varies moves slightly to escape book learning . This is little more than added weakness. Adding randomness to the eval will also slaughter your move ordering. All in all, a very bad idea. >Look at Chessmaster6000 performance for example - one could argue that this >program is very close to having no opening book at all . Still it did and does >very well . If you have no opening book, you will rapidly get killed by programs that learn. You could make the same argument for Chess Tiger. In fact, I think that at one time Chess Tiger had little or no opening book also. Against the latest generation of chess programs, that will mean that after a few thousand games you will lose every single game without fail. I have not performed the calculations. But I have observed chess programs without book (or book improperly configured) lose repeately to a program with a book properly configured. It's only a guess, but I think it might be more important than 80%. Look at how many games are lost in the current tournament because of opening book. I dare say nearly all of them when the opponents are approximate peers. If you're 300 ELO better you can have much higher odds of pulling out of your faux-pas. But even at that, vastly superior programs will get thrashed by a bad opening move. I suspect that a very good and deep opening book will add 500 ELO [minimum] and a very bad one will subtract at least as much or could even cause you to lose every single game in the absurd cases. If someone wants to test my conjecture then by all means do so. The enormous advantage of a good opening book is so obvious to me that I don't see it worth the bother to test. Of course, if I am wrong, it is a titanic mistake. But I doubt it.
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.