Author: Laurence Chen
Date: 20:01:15 10/21/99
Go up one level in this thread
On October 21, 1999 at 20:22:44, Thorsten Czub wrote: >Example why it is dangerous to use a program >like Fritz6 in analysis for (e)mail-chess: > >In the moment i am playing a public game against >anonymous person in a newsgroup. > >during the analysis i came to the following situation: > >after the moves > >[Event "mail-chess-game"] >[Site "gambit-soft-board"] >[Date "1999.10.22"] >[Round "1"] >[White "Thorsten Czub"] >[Black "Anonymous called <Rudi>"] >[Result "*"] > >1. d4 d5 2. e4 dxe4 3. Nc3 Nf6 4. f3 exf3 5. Qxf3 Qxd4 6. Be3 Qh4+ >7. g3 Qg4 8. O-O-O Qxf3 9. Nxf3 c6 10. Bg2 Nbd7 11. Rhe1 e6 12. Nd4 Bc5 >13. Na4 Bxd4 14. Bxd4 O-O 15. b3 Re8 16. Nb2 Nb6 17. Nd3 Nfd5 18. c4 Nf6 >19. Bf3 Nfd7 20. Nf4 Kf8 21. Bb2 a5 22. Ba3+ Kg8 23. c5 a4 * > > > >the following position came on board of fritz6: > >Ich - Rudi >r1b1r1k1/1p1n1ppp/1np1p3/2P5/p4N2/BP3BP1/P6P/2KRR3 w - - 0 1 > >fritz6 evaluated the position and the chances as follows: > > >Analysis by Fritz 6: > >24.Txd7 > -+ (-2.97) Tiefe: 1/3 00:00:00 >24.Txd7 Sxd7 > -+ (-3.31) Tiefe: 1/4 00:00:00 >24.cxb6 > = (0.16) Tiefe: 1/4 00:00:00 >24.cxb6 Sxb6 > = (0.16) Tiefe: 2/4 00:00:00 >24.cxb6 > = (0.00) Tiefe: 3/5 00:00:00 >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > µ (-1.22) Tiefe: 3/9 00:00:00 >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > µ (-1.22) Tiefe: 4/6 00:00:00 >24.cxb6 axb3 25.Kb2 bxa2 26.h3 > µ (-1.12) Tiefe: 5/13 00:00:00 1kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.h3 > µ (-1.06) Tiefe: 6/16 00:00:00 7kN >24.cxb6-- > µ (-1.37) Tiefe: 7/11 00:00:00 8kN >24.cxb6 > µ (-1.37) Tiefe: 7/15 00:00:00 10kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.h3 e5 28.Td6 > -+ (-1.56) Tiefe: 8/16 00:00:00 20kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Lc5 Sd7 > -+ (-1.66) Tiefe: 9/17 00:00:00 43kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Te4 e5 28.Tde1 f6 > -+ (-1.69) Tiefe: 10/20 00:00:00 158kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Te4 e5 28.Tde1 f6 > -+ (-1.72) Tiefe: 11/26 00:00:01 506kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Lb2 f6 > -+ (-1.78) Tiefe: 12/27 00:00:06 1909kN >24.cxb6 axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Lb2 f6 > -+ (-1.81) Tiefe: 13/31 00:00:30 10018kN >24.b4 > -+ (-1.78) Tiefe: 13/31 00:01:02 20891kN >24.b4 Sc4 25.Lb2 Sxb2 26.Kxb2 e5 27.Sd3 f5 28.Lh5 Te7 > -+ (-1.69) Tiefe: 13/33 00:01:29 30248kN >24.b4 Sc4 25.Lb2 e5 26.La1 Sf6 27.Sd3 Le6 28.Sxe5 Sxe5 > -+ (-1.59) Tiefe: 14/31 00:04:11 84912kN > >As you can see from the analysis above, fritz6 thinks >it is 1.5 pawns up. >A few main-lines before it came to the conclusion that >cxb6 axb3 etc. is worth only 1.81 so it changes to >b4 because it thinks this would be a better idea for >white. > >Now we make 24.cxb6 and watch out what fritz6 says now : > >Ich - Rudi >r1b1r1k1/1p1n1ppp/1Pp1p3/8/p4N2/BP3BP1/P6P/2KRR3 b - - 0 1 > >Analysis by Fritz 6: > >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 > = (0.09) Tiefe: 4/10 00:00:00 1kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 > ² (0.28) Tiefe: 5/14 00:00:00 7kN >24...axb3 > = (-0.03) Tiefe: 6/14 00:00:00 20kN >24...axb3-- > ² (0.28) Tiefe: 7/16 00:00:00 49kN >24...axb3 > ² (0.28) Tiefe: 7/19 00:00:00 75kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 Sd7 28.Le4 e5 > = (0.09) Tiefe: 8/22 00:00:01 322kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Ld6 g5 29.Sd3 > = (0.00) Tiefe: 9/21 00:00:02 848kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Ld6 g5 29.Sd3 > = (-0.03) Tiefe: 10/24 00:00:08 2685kN >24...axb3 25.Kb2 bxa2 26.Kxa2 e5 27.Kb3 f6 28.h3 Kf7 29.Sd3 > = (-0.09) Tiefe: 11/28 00:00:33 9872kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 > = (-0.22) Tiefe: 12/29 00:02:01 37126kN >24...axb3 25.Kb2 bxa2 26.Kxa2 Sxb6 27.Kb3 e5 28.Ld6 f6 29.Lh5 > = (-0.19) Tiefe: 13/31 00:08:23 151133kN > > > >HOW can this be ? > >I guess it loads new piece-square-tables (preprocessing) >with new values and gets a totally different >impression from the position than ONE PLY before. > >it is obvious that this way the whole main-line >computation of ONE PLY BEFORE was senseless time-wasting, >since the new thought-computation gets a different >point of view about the position. > >Guess you are a newbie in computerchess, guess >you don't know about chess programs and guess you >have bought this very strong program, the package >says it got 2800 ELO performance against top-players. > >you trust this. >and you analyse your games with it. >you know that the longer you let fritz6 compute, >the better the results will be. > >you have to consider about the position. >and you get the one or the other evaluation from fritz6: > >one time it says black is 1.5 pawns ahead, >and ONE ply later it says : oh - not a problem, only 1/5 >of a pawn against white. > >So - many years we had genius having asymmetrically evaluations >and main-lines. >It had the same effect: you let it compute >a long time about a position, it says 1.5 for black (e.g.) >and you give it ONE PLY later and it says only 0.2 for >black. > >This time Fritz6 replaces the Genius misbehaviour: >i don't think it has an asymmetrical search, but i guess >the preprocessing makes it reevaluate only the root- >positions, without making sensible changes in the main-line. >cxb6 was in the main-line of the 24th move. >but this move was not moved on the board. >One ply later the root position is another, fritz has to load >new values in the tables, and suddenly the whole score changes >1.3 pawns ? > >How can a chess-program come to sensible information with >this type of behaviour ?? >How can you trust the analysis of a chess program >when it changes its mind this heavily ? >How can you create a sensefull main-line when >ONLY the root position gets somehow evaluated, and >not the moves in the main-line far from the 1st ply ? > > > >(BTW: fritz5.32 behaves the same way. it works the >same way, using heavy preprocessing methods.) > >Fritz5 and Fritz6 are strong programs. >When you let them play against humans or against >other chess programs. They are not the strongest >chess programs (ChessTiger and others are much stronger) >but they are much for the money they cost. > >But - as an analysis tool for chess players... >NO ! > > >i would like to see all my opponents use this type of programs >(nimzo, junior...) >so that i can easily beat them. This does not surprise me at all. I've seen this happens with other engines as well. I don't think it is fair to pick on Fritz and accuse it that it's stupid in analysis or it's very dumb, but the user behind it. It's like the old saying, the blind leads the blind and they both fall off into the ditch. Therefore, this is a warning sign to all users who relies solely and takes any computer chess analysis as the ULTIMATE truth. I personally never take any chess analysis to its face value, I always play another move, or another ply and see if the evaluation changes dramatically. If it remains the same, then I know that the assessment is correct, if it changes dramatically, then it's because of the horizon effect which all chess engines suffer, and I would then conclude that the chess engine is off in its evaluation. To someone new to chess engine this may be a new revelation. Therefore the moral is one should never trust any analysis of chess engines without further investigation. Laurence
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.