Author: Andrew Williams
Date: 13:42:22 03/29/99
Go up one level in this thread
On March 28, 1999 at 18:21:54, Paul Richards wrote: >On March 28, 1999 at 11:05:09, Dave Gomboc wrote: > >>Crafty is capable of using 16 processors, maybe more, I don't know exactly. >>Fritz would probably love to have a gigabyte of ram. Whatever. If you are >>trying to compare program quality, you can't cripple one of them, then say >>"look, the other one is way better". Well, actually you can, but it sounds >>silly. > >You don't have to cripple anything to make a comparison. Here is an >easy example: 1.e4 e5 2. Nf3 Nc6 3. Bb5, the start of the Ruy Lopez. >Using Crafty 16.5 engine in Fritz interface, I set this position and >started Infinite Analysis. In the 10-12 ply range Crafty recommends >the positionally atrocious 3...Bd6? as black's best. Sure it develops >the bishop and defends the pawn, but the pawn does not need defense. >White exchanging on c6 fails to Qd4, so white will never take the >pawn. So all Bd6 accomplishes is to block the d pawn and light squared >bishop. This is obvious even to non-masters, and a Kasparov seeing >something like this would be enough to conclude how Crafty plays. >I let it continue to think to 15 ply, and when I came back after a >number of hours it had chosen the sensible 3...a6. But 12 plies is >already a significant depth, and Bd6 should have been rejected before >this. For Fritz Bd6 is never a serious contender even at shallow >search depths. So this is a difference in how positions are evaluated >and has nothing to do with hardware, optimal compilation or what >have you. > With respect, I don't think this argument has a lot of merit. For one thing, I'd guess that any reasonable opening book for crafty would stretch well beyond 3...Bd6, so the point is moot. However, even if you ignore this point, my program found 3...a6 at ply 10 and sticks with it to depth 12. By your argument, this makes PostModernist stronger than crafty. While I am proud of my program, I can assure you that it is still some way off that standard. For evidence, search on ICC for games between PostModernist and the various crafty clones - I have never beaten or drawn with a crafty clone on equal or better hardware. >I reran this test several times, and on one occasion Crafty did like >3...a6 up until 12 ply, where it switched back to Bd6. I don't know >why it isn't consistent, but I don't have a standalone version of >Crafty to test. I just tested 16.6 engine on the same position, and see Unless you're running on an SMP computer, I don't think this should happen. Perhaps I'm wrong, but I would be very concerned if my program produced two different best moves for the same position. Bob would be better placed to discuss this, of course. >a clear improvement in the recommendations. Its choices at 12 >ply are a6, f5, Bc5, and only then Bd6. The point is that this is >the sort of thing that needs to be tuned, not whether Crafty can >play on 100 processors. That part is already superior to commercial >programs, it is the weaknesses that need attention. I think you're simply identifying the fact that some programs behave oddly in the opening without an opening book. I'm absolutely sure that my program's eval.c would look significantly different if I had to play without an opening book (assuming I wanted to play under those conditions). For an in-depth discussion on the role of opening books, look at RGCC over the last few weeks. There was some reasonable stuff in the debate there until the lunatics started to "contribute". Best regards Andrew Williams
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.