Author: Robert Hyatt
Date: 07:18:16 05/21/01
Go up one level in this thread
On May 21, 2001 at 06:52:52, Uri Blass wrote: >This position could happen in kasparov-Deeper blue but kasparov did not play >40.gxh5 > >Based on Deeper blue logfiles we know that Deeper blue evaluated the position >as 0.5 pawn advantage for white at depth 9(6)=15 > >Crafty's evaluation at depth 15 is 2.93 pawns for white so we can see that the >difference in evaluation is 2.43 pawns. > >Part of the difference can be explained by different evaluation but most of the >difference is clearly because of the fact that Crafty can see more than Deeper >blue. You should be able to see crafty's PV, which would clearly explain whether the 2.9 is position or material. Unfortunately, you can not see DB's PV so there +.5 is totally unknown. IE it could be just a bit better positionally, it could be material down and a really big positional plus. Who knows? and who can tell? > >Deeper blue found the line 40...Bxh5 41.Rh1 only at depth 10(6)=16 when Crafty >could see it at depth 14 and the line was not changed at the next 2 plies. This is the same problem. My eval is tuned to more risky play, because of the opposition I see on ICC all the time. In some types of positions I would expect it to get a much better picture of the real score than DB. But in others I would expect it to get a worse idea. And tactically, it would be no competition with DB having a huge edge. If DB searches 16, and crafty searches to 16, I would prefer the non-null-move 16 any day. It is _clearyly_ more accurate. > >Crafty can see the line 40...Bxh5 41.Rh1 Bf3 42.Rg1 Bh5 43.g7 Nf3 44.Rg3 c5 >45.bxc5 Rc8 46.f6 Rxc5 at depth 15 and 16 when Deeper blue can see it only at >depth 17,18(deeper blue can see a different line but the difference is only >41...Kg5 42.Rg1 Kh6 that leads to the same position as 41..Bf3 42.Rg1 Bh5 and >moves 43-46 are the same. > >It seems that Crafty is a very good program and it can outsearch deeper blue by >2 plies. That's a speculation that is based on too many what-ifs. IE what if you just let each program search for 3 minutes? Which is better? I don't like to compare search depth to search depth, as that is usually a totally wrong way to compare programs. IE you might try hiarcs at the same depth and find that it sees more or less. Plies between programs are not really interesting. > >I do not know the meaning of the plies of deeper blue but it seems to be not >brute force search. > >I also doubt if deeper blue could search 200M nodes per second otherwise I >cannot understand the reason that it is outsearched by other top programs again >and again based on the pv's What "top program" sees more in 3-4 minutes than they do? IE what program can solve all the nolot positions in reasonable time (maybe 1-2 exceptions as I think they missed one or two). But who can do all but 2 in 3-4 minutes? To say "top programs outsearch them" therefore seems strange. Unless you mean "top programs outsearch them given _enough_ time". And that is always true and meaningless at the same time. > >It is possible that the claims about 200M nodes per second were only a >psychological war against kasparov Not a chance... I have sat behind them and watched too many times... > >Here is the analysis of Crafty > >kasparov - Deeper blue >[D]3rr3/8/2p3Pk/1p2nP1P/pP2p3/P1B1Nb1B/2P2K2/5R2 b - - 0 1 > >Analysis by Crafty 18.08: > >40...Kxh5 > +- (1.42) Depth: 1/8 00:00:00 >40...Kxh5 41.g7 Ng4+ 42.Bxg4+ Bxg4 > +- (1.68) Depth: 2/8 00:00:00 >40...Kg7 41.f6+ Kxf6 > +- (1.53) Depth: 2/8 00:00:00 >40...Kg7 41.f6+ Kxf6 42.Kg3 > +- (1.53) Depth: 3/8 00:00:00 >40...Kg7 41.Ng4 Bxg4 42.Bxg4 c5 > +- (2.58) Depth: 4/10 00:00:00 >40...Rd6 > +- (2.18) Depth: 4/10 00:00:00 >40...Rd6 41.Rg1 Bxh5 42.g7 Bf7 > +- (2.03) Depth: 4/10 00:00:00 >40...Kxh5 > +- (1.63) Depth: 4/10 00:00:00 >40...Bxh5 41.Bg2 Bf3 42.Bh3 > ± (1.34) Depth: 4/10 00:00:00 >40...Bxh5 41.Bg2 Nxg6 42.fxg6 Rf8+ 43.Ke1 Rxf1+ 44.Kxf1 Bxg6 > ± (1.33) Depth: 5/16 00:00:00 >40...Bxh5 41.Kg3 Kg5 42.Rf4 Bf3 43.Rh4 > +- (1.62) Depth: 6/16 00:00:00 88kN >40...Bxh5 41.Rg1 Kg7 42.Bg4 Kh6 43.Bxe5 Rxe5 44.Rh1 > +- (1.76) Depth: 7/19 00:00:01 222kN >40...Bxh5 41.Rg1 Nf3 42.Rh1 Kg5 43.g7 Rg8 44.f6 Kf4 > +- (1.81) Depth: 8/20 00:00:01 480kN >40...Bxh5 41.Rg1 Nf3 42.Rh1 Ne5 43.Bg2 Nxg6 44.fxg6 Kxg6 > +- (1.78) Depth: 9/23 00:00:03 1223kN >40...Kxh5 41.Rg1 Kh4 42.Bf1 Kh5 43.g7 Ng4+ 44.Nxg4 Bxg4 45.Rh1+ Kg5 46.f6 e3+ >47.Kg3 > +- (1.77) Depth: 9/23 00:00:04 1778kN >40...Kxh5 41.Rg1 Kh4 42.Bg2 Bxg2 43.Rxg2 Kh5 44.g7 Nf7 45.Rg6 Nd6 46.Bd4 > +- (2.01) Depth: 10/26 00:00:07 3318kN >40...Bxh5 41.Kg3 Kg5 42.g7 Nf7 43.f6 Rg8 44.Rf4 Bf3 45.Be6 Kg6 > +- (1.94) Depth: 10/26 00:00:14 6548kN >40...Bxh5 41.Kg3 Kg5 42.g7 Nf7 43.f6 Rg8 44.Be6 Bg6 45.Rh1 Rd6 46.Bg4 > +- (2.16) Depth: 11/27 00:00:24 11929kN >40...Kxh5 41.Rg1 Kh4 42.Bg2 Bxg2 43.Rxg2 Kh5 44.g7 Nf7 45.Bf6 Rd2+ 46.Kf1 Rxg2 >47.Kxg2 > +- (2.14) Depth: 11/27 00:00:33 16243kN >40...Kxh5 41.Rg1 Kh6 42.g7 Nf7 43.Rg6+ Kh7 44.Rxc6 Rc8 45.Rxc8 Rxc8 46.Bd4 Rd8 >47.Bc3 Kg8 > +- (2.53) Depth: 12/29 00:01:24 41952kN >40...Bxh5 41.Kg3 Kg5 42.g7 Nf3 > +- (2.45) Depth: 12/29 00:01:48 54476kN >40...Bxh5 41.Kg3 Kg5 42.g7 Nf7 43.Rf4 Nd6 44.f6 Bg6 45.Rg4+ Kh5 46.Kf4 Rg8 >47.Rg5+ Kh4 48.Be6 > +- (2.71) Depth: 13/32 00:02:59 89583kN >40...Bxh5 41.Rh1 Kg5 42.Rg1+ Kf6 43.Bg2 Bf3 44.Bxf3 exf3 45.g7 Kf7 46.g8Q+ Rxg8 >47.Rxg8 Rxg8 48.Bxe5 Re8 49.Ng4 > +- (2.61) Depth: 14/34 00:08:56 267478kN >40...Bxh5 41.Rh1 Bf3 42.Rg1 Bh5 43.g7 Nf3 44.Rg3 c5 45.bxc5 Rc8 46.f6 Rxc5 >47.Bb4 Rg5 48.Rxg5 Kxg5 > +- (2.93) Depth: 15/39 00:23:36 722662kN >40...Bxh5 41.Rh1 Bf3 42.Rg1 Bh5 43.g7 Nf3 44.Rg3 c5 45.bxc5 Rc8 46.f6 Rxc5 >47.Bb4 Rg5 48.Rxg5 Kxg5 49.Be7 > +- (3.06) Depth: 16/40 00:43:40 1309529kN > >(Blass, Tel-Aviv 21.05.2001) That shows that Crafty can get to depth=16 after searching a billion nodes. With Null-Move (R=3) that equates to a brute-force of 13 plies. I'm not sure how to reach any comclusion from that. There are _always_ positions that Crafty will evaluate better than _anybody_. There are always positions where it will evaluate worse. Ditto for tactics. Unfortunately, one position doesn't win a game, although one can lose 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.