Author: Moritz Berger
Date: 15:01:04 05/29/98
Go up one level in this thread
I took the liberty to analyze the game you posted here. 1.b3 e5 2.h3 d5 seems slightly unusual and smells like a typical computer killer opening, the more interesting is the fact that a theme from e.g. the QG Tarrasch defense (poised rook on a1 offered to a black knight) became essential in this game. To offer the Ra1 was a very human move (and every human player would have smelled the danger of playing Nc2 NxRa1). Maybe a bug in CSTal ? ;-) Still I find it strange that you chose a game where Fritz plays 'as bad' as every other program on this planet. The rook offer was a nice bug (just kidding ;-)), but you promised to show us some really bad chess from Fritz. I have a couple of Fritz 5 games here that are really bad and might have been better to illustrate your point. The only issue I take is that I can easily produce such games for *ALL* programs out there. And then it's no bean counting for me if Fritz scores approx. 60% in >1000 games vs. other top programs. You get good and bad moves from every program. I'm sure you agree. The following mathematical theorem comes to my mind: Every sequence that isn't absolutely convergent has a bijection to make it convergent to any arbitrary number. Application: By picking good and bad moves (since there are both kinds of moves) or good and bad games (sum of moves) I can claim everything I want about any program. That's why we can continue forever with our little discussion as long as you refuse to accept strong quantitative arguments instead (score in >1000 games at 40/120) and keep referring to 'secret autoplayer mumbo jumbo' etc. This doesn't make a meaningful discussion easier to conduct for all participants and to comprehend for our more and more bewildered 'silent' audience. We could also talk about typical positions that Fritz doesn't understand (e.g. certain endgames). But again: Lots of EPD positions instead of very few PGN gamescores without search depth/evaluation or any comments about keymoves serve this purpose much better. I hope you are able to discuss on this level, instead of calling other people names and treating them like ignorant idiots and implying sinistre motivations. Lots of name-calling and unfair accusations has happend to you, too, on r.g.c.c. and you didn't exactly like it, as far as I remember ... Let's see what other programs might have played instead of the moves Fritz played in the actual game: 14...Nc2 is irresistible for Fritz and Hiarcs alike (as well as the other moves that followed), here's the logfile from Shredder 2 (all programs on my P233MMX with 128MB RAM): 12 -> 1:56.96 +2.05 Nc2 Qb2 Nxa1 Qxa1 O-O Nbd2 c5 Qc1 Qa5 Qc4 Rad8 (6.679.333) 13.01 7:37.48 +1.81 Nc2 Qd6 Nxa1 Nbd2 Qd8 Qe5+ Qe7 Qxe7+ Kxe7 Rxa1 Rhd8 Rc1 Nh5 g3 Rac8 Rc4 (24.348.449) best move: Nb4-c2 time: 7:40.713 min n/s: 53.287 nodes: 24.550.241 eval hash hits 14%; search/eval hash 97% / 96% filled 15...Nxa1 is the logical consequence of Nc2 (again Shredder 2): 12.01 3:46.23 +1.80 Nxa1 Nbd2 Nc2 Rc1 Ne3 fxe3 dxe3 Nc4 Qa6 Nxe3 (12.593.420) 12 -> 6:48.97 +1.80 Nxa1 Nbd2 Nc2 Rc1 Ne3 fxe3 dxe3 Nc4 Qa6 Nxe3 (22.497.146) best move: Nc2xa1 time: 6:53.465 min n/s: 54.954 nodes: 22.721.800 eval hash hits 16%; search/eval hash 96% / 94% filled 16...Nxb3 is also a popular move: 10.01 0:18.32 +2.10-- Nxb3 (1.082.073) 10.01 0:27.27 +1.74 Nxb3 Nc4 Qb5 axb3 Qxb3 Qc5 O-O-O Qxa7 Qxd3 Qa8+ Kd7 Nce5+ Ke6 Qxb7 Kd6 (1.615.210) 10.02 1:11.97 +1.75++ Qa6 (4.010.390) 10.02 1:33.84 +1.75 Qa6 Rxa1 Qxd3 Re1 h6 Nc4 Rd8 Qe5+ Kf8 Nd6 Nd7 Qxd4 Qxd4 Nxd4 (5.166.055) 10.08 2:02.75 +1.76++ Qd8 (6.700.728) 10.08 2:27.44 +1.93 Qd8 Nc4 Qxd6 Nxd6+ Kd7 Nxf7 Nc2 Rc1 Nb4 Nxh8 Rxh8 (7.943.945) 10 -> 2:41.87 +1.93 Qd8 Nc4 Qxd6 Nxd6+ Kd7 Nxf7 Nc2 Rc1 Nb4 Nxh8 Rxh8 (8.735.239) 11.01 3:18.91 +1.70 Qd8 Nc4 Qxd6 Nxd6+ Kd7 Nxf7 Nc2 Rc1 Nb4 Rc4 c5 (10.645.311) 11.07 5:51.95 +1.71++ Qa5 (18.488.903) 11.07 6:08.06 +1.76 Qa5 Rxa1 Rd8 Nc4 Rxd6 Nxa5 Rd7 Rc1 Nh5 g3 O-O Rc4 Rfd8 Nxd4 (19.442.088) 11 -> 6:57.48 +1.76 Qa5 Rxa1 Rd8 Nc4 Rxd6 Nxa5 Rd7 Rc1 Nh5 g3 O-O Rc4 Rfd8 Nxd4 (22.035.167) best move: Qb6-a5 time: 7:24.299 min n/s: 52.831 nodes: 23.473.000 eval hash hits 14%; search/eval hash 97% / 96% filled Shredder would have played Qa5 instead at 11ply, but Nxb3 was very close all the time, I checked this in single move analysis mode where it only failed 'really' low at 14ply: analysing Na1xb3 ... 9 -> 0:08.76 +2.35 Nxb3 Nc4 Qb5 axb3 Qxb3 Qc5 Qc3 Nxd4 Qxd3 Nd6+ Kf8 Nxb7+ (516.442) 10.01 0:10.94 +2.10-- Nxb3 (630.264) 10 -> 0:20.41 +1.74 Nxb3 Nc4 Qb5 axb3 Qxb3 Qc5 O-O-O Qxa7 Qxd3 Qa8+ Kd7 Nfe5+ Ke8 Qxb7 Ng8 (1.196.985) 11 -> 0:41.31 +1.63 Nxb3 Nc4 Qb5 Rb1 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 (2.471.802) 12 -> 1:26.82 +1.63 Nxb3 Nc4 Qb5 Rb1 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 (5.283.449) 13 -> 6:30.36 +1.73 Nxb3 Nc4 Qb5 Rb1 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 (21.941.575) 14.01 8:15.87 +1.48-- Nxb3 (27.544.386) 14 -> 18:34.43 +1.10 Nxb3 Nc4 Qd8 axb3 Qe7 Nxd4 Kf8 Qxe7+ Kxe7 (61.489.539) 17.Qb5 is ok, but now Shredder starts to smell the danger: 7 -> 0:04.30 +2.35 Qb5 axb3 Qxb3 Qc5 Qc3 Nxd4 Qxd3 Nd6+ Kf8 Nxb7+ (236.604) 8.01 0:05.69 +2.10-- Qb5 (308.539) 8 -> 0:12.45 +1.74 Qb5 axb3 Qxb3 Qc5 O-O-O Qxa7 Qxd3 Qa8+ Kd7 Nce5+ Ke7 Qxb7+ Kd6 (688.718) 9 -> 0:26.49 +1.63 Qb5 Rb1 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 (1.488.407) 10 -> 0:51.84 +1.63 Qb5 Rb1 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 (2.997.880) 11 -> 2:36.38 +1.69 Qb5 axb3 Qxb3 Qc5 Nd7 Qxd4 Nf6 Nfd2 a5 Nxb3 Nxe4 (8.816.886) 12.01 7:11.57 +1.44-- Qb5 (23.490.700) 12.01 8:43.82 +0.94-- Qb5 (28.351.479) 18.Ne4 is the final turning point where all programs need about 5 minutes to prove that after Ne4 white is better (Fritz 5 after 257" fail low (+= 0.44)). Here again Shredder's logfile (it doesn't even concede white's superiority after Nxe4): 9 -> 0:25.74 +1.73 Nxe4 Re1 Nbc5 Nxd4 Ne6 Nxb5 Nxd6 Nbxd6+ Kf8 Nxb7 Rb8 (1.477.180) 10.01 0:26.49 +1.48-- Nxe4 (1.520.708) 10.01 0:26.88 +0.98-- Nxe4 (1.545.865) 10.01 0:34.64 +0.79 Nxe4 Re1 Nbc5 Nxd4 Ne6 Rxe4 Qb1+ Kh2 Qa1 Nxe6 (2.095.248) 10 -> 1:47.69 +0.79 Nxe4 Re1 Nbc5 Nxd4 Ne6 Rxe4 Qb1+ Kh2 Qa1 Nxe6 (5.926.991) 11.01 1:51.86 +0.54-- Nxe4 (6.151.757) 11.01 2:16.06 +0.43 Nxe4 Re1 Nbc5 Nxd4 Ne6 Rxe4 Qb1+ Kh2 g6 Nxe6 fxe6 Rxe6+ Kf7 (7.631.715) 11.06 5:33.98 +0.44++ Rd8 (18.208.020)
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.