Author: Vincent Lejeune
Date: 04:24:36 06/14/99
Go up one level in this thread
On June 14, 1999 at 02:32:03, Thorsten Czub wrote:
>Where no one/man has gone before!
>
>
>Fritz5.32 (latest patch from the ChessBase site) is running on my k6/400 mhz
>machine, doing
>arround 370-420.000 NPS depending on position and hashtables used.
>On my second PC (same machine, k6/400) I let run ChessSystemTal 2.5 (downloaded
>from
>Oxford-site). ChessSystemTal2.5 is doing arround 5900-6500 NPS.
>
>The programs play a tournament gam, using 40 moves in 120 minutes.
>We are in the 19th move, ChessSystemTal has black. The material is even.
>
>Fritz5.32 evaluates the position with 0.00. CSTal 2.5 says +2.59.
>The programs have castled to different sides. In this moment CSTal2.5 plays h5
>to
>attack the white-king-position.
>Fritz looks into the tree, almost 13 plies it says, 40 selective plies, and does
>not see anything.
>Therefore it evaluates 0.00. It has no further criteria to evaluate a position
>where no concrete material
>is attacked, lost or whatever. It is quite simply. It has no criteria, therefore
>evaluates 0.00.
>
>Fritz trusts in its search. Doing 358.000 NPS it is calculating 55 times more
>positions (or moves)
>55 times faster than CSTal2.5.
>One would imagine it would wipe CSTal2.5 away instantly.But this does not
>happen.
>The evaluation of CSTal2.5 is almost speculative. In these 2.x pawns it gives,
>there
>are several points of the position looked up. CSTal recognizes that the
>different castle situations could be a problem. Also it has connected queen and
>Bishop onto an important diagonal towards the white king. White has moved his
>pawn/wall forward, this could maybe be a problem because black could
>attack these pawns.
>What CSTal2.5 evaluates is not real. it evaluates the CHANCES of the position to
>use in further moves. Move 22. Fritz scores 0.03 for white. CStal is in the 10th
>search and has a fail-high for the move 22.c4 that it expects white to play.
>Normal intelligent chess programs add positional values. For a bad thing they
>give a malus, for a good thing they give a bonus. Than they add the little
>values to a score. In old times the positional score was often limited in a
>range from 1-1.5 pawn units. todays chess-programs have extended this range up
>to a piece (3pawns) and some to a rook (5pawns).
>But this has not changed the method they generate their evaluation. it is still
>just a sum of different parts. As we all know, that in life the things are more
>than the sum of the parts, because something
>with life , with quality is born when a critical amount of parts combines in a
>certain way, this rule has not been realized in chess-programs. but it has been
>realized in chess-system-tal.
>In chess-system-tal the evaluation is not mechanical addition of minor-sums. it
>is something different.
>In the moment a tension in a position gets alive, suddenly the scores of cstal
>increase to an unbelievable amount. CSTal2.5 can play normal games, boring and
>with normal exchanges situations. but it can also happen that the position gets
>critical and something very different to normal chess-programming happens.
>Although it is 55 times slower than fritz5.32 , a leading program of the
>swedish-rating list, it does not die doing these evaluation-effects.
>25th move. Fritz evaluates 0.00. Whenever Fritz knows nothing, and does not see
>anything in the tree,
>it evaluates 0.00 and arround draw scores. This happens very very often. Fritz
>has increased NPS. It seems the position is more easy for it. In opposite the
>NPS rate of CSTal drops down, because the position gets more and more
>interesting for CSTal2.5 and it does extend more and more lines in the tree
>because they seem interesting.
>Fritz now calculates arround 450.000 NPS in the permanent-brain of move 26. it
>expects qxh3 and says again 0.00 !
>Guess this would be your mail-chess-game and you would have to find out
>something sensefull. all information you would get from Fritz would be, that the
>game is 0.00 !! Wherever you analyse.
>Maybe (this is Markus Gille and my theorie) fritz has very very often (due to
>special search algorithms) draw-by-repetition in the search-tree. so we could
>call these stupid effects a side-effect, a kind of system-immanent "bug"-like
>effect, caused by the search fritz trust so heavily in.
>
>it often seems to make the program blind because it only sees draws and
>draws-by-repetition everywhere.
>
>The 25th move, cstal2.5 has to calculate lots of things. the NPS decreases <5000
>NPS.
>Fritz captures cxb7+ and evaluates - guess what - 0.00.
>It seems chess is really easy. anthing is draw in chess :-))
>It expects cstal2.5 to move Kxb7 and capture the check-giving pawn. But Cstal2.5
>has a different idea.
>it likes Kb8.
>In reaction fritz evaluates the position now 0.03. Still a draw-score.
>Some exchanges follow. In the end, fritz in the 30.move evaluates 0.09. A little
>arround draw.
>In the 31st move it says 0.19. maybe it believes all endgames are draws. this is
>the program
>kasparovs calls the reference in tactics. i begin to understand why.
>in my opinion, when the day comes the last human beeing can win against such a
>"reference in tactics", then the day comes such programs like cstal kill those
>tactical-programs.
>maybe humans have to learn more from cstal than cstal has to learn from humans.
>we always thought it is vice versa and we had to teach the program how to play
>human-like.
>but all those humans lose against programs like fritz. cstal does not. so maybe
>we have to overthink chess in general. when kasparov says it was higly
>underestimated to what a degree tactice is important for chess, this maybe only
>shows the key to the reason why players like him lose against those stupid chess
>programs. because these human beeings believe something that is not true.
>of course tactics is important. of course. but how can cstal handle it with 5500
>NPS and 55 times slower, when tactics , played by the "reference in tactics" is
>SO IMPORTANT ???
>Maybe we suddenly see the reason why those players get higher and higher
>elo-ratings, but do not play better chess at all. because this is the same thing
>that happens to brute-force fast-search-chess-programs in the swedish elo-lists.
>The future looks promising. human beeings will have to learn chess again from
>machines, who took over the leadership. human-beeings have to rethink the way
>they played chess. they have to understand that tactics is only important, when
>you call it important. and when you stress it over and over again. the classics
>have never done this way. lasker or capablanca, they have played a different
>chess. not with inflationary elo-ratings.
>A nice future. chess players of the future will have to learn from intelligent
>programs how to do it. how to beat the fast-searchers. the rules
>chess-programmers implement to make their programs stronger but faster, will
>lose its power in the moment human beeings find out how intelligent programs
>deal with it.
>it is the 32nd move. fritz evaluates 0.06. arround draw. CSTal 2.5 says +2.15.
>Maybe the dawn of the intelligent programs begins with the new century, in the
>moment of the highest successes of fast-searchers. because with any victory the
>fast-searchers write-down in computerchess history, they commit suicide in the
>race of the better chess.
>
>33rd move, fritz says -0.06 against it. cstal2.5 says +1.86. Fritz changes to
>-0.03 score. impressive.
>34th move, fritz evaluates 0.00, than -0.06. Move 35. Fritz suddenly drops to
>-0.25. CSTal2.5 evaluates +2.09 pawns.
>
>Move 35. Fritz says 0.00 again. Sure. Endgames are often draws. CSTal says
>+2.12.
>
>In the following moves Fritz increases its evaluation, until it finally drops in
>move 50.
>In the end fritz lost the "drawish"-game.
>
>As kasparov says: a tactical reference ! :-))
>
>Running on 4 xeon quad-cpu's fritz would find these draw scores about 4 times
>faster.
>
>where does this way bring us? what is the message ?
>I have my answer. And I can wait. this century is nearly over. only a few days
>to go.
>
>
>
>[Event "Stufe=120'/40+60'/20+0'0. "]
>[Site "k6-2/400 both"]
>[Date "1999.06.14"]
>[Round "1"]
>[White "Fritz 5.32"]
>[Black "ChessSystemTal2.5"]
>[Result "0-1"]
>[ECO "B01"]
>[Annotator "0.09"]
>[PlyCount "158"]
>[EventDate "1999.05.28"]
>
>{41984kB, General.ctg. AMDk6
>} 1. e4 {0} 1... d5 {1} 2. exd5 {0} 2... Qxd5 {1}
>3. Nc3 {0} 3... Qa5 {1} 4. d4 {0} 4... c6 {1} 5. Nf3 {0} 5... Nf6 {1} 6. Bc4 {0
>} 6... Bf5 {1} 7. Bd2 {0} 7... e6 {295} 8. Nh4 {0} 8... Bg4 {
>letzter Buchzug 168} 9. Nf3 {0.09/1 489} 9... Qb6 {369} 10. Bb3 {-0.44/13 267}
>10... Nbd7 {1} 11. Be3 {-0.22/11 153} 11... Qa5 {144} 12. h3 {-0.25/12 145}
>12... Bh5 {281} 13. O-O {0.00/12 67} 13... Bd6 {303} 14. Bd2 {-0.31/12 601}
>14... Qb6 {1} 15. Be3 {-0.19/12 196} 15... Qc7 {1} 16. Qe2 {-0.28/12 309} 16...
>O-O-O {378} 17. Rfe1 {-0.28/12 176} 17... h6 {182} 18. g4 {-0.25/11 114} 18...
>Bg6 {375} 19. Rad1 {0.00/13 20} 19... h5 {161} 20. g5 {0.00/13 516} 20... Nd5 {
>1} 21. Nxd5 {0.09/13 235} 21... exd5 {1} 22. c4 {0.03/13 454} 22... dxc4 {1}
>23. Bxc4 {0.00/13 240} 23... Nb6 {1} 24. d5 {0.00/12 219} 24... Qd7 {7} 25.
>dxc6 {0.00/11 132} 25... Qxh3 {553} 26. cxb7+ {0.00/12 57} 26... Kb8 {826} 27.
>Rd4 {0.03/12 251} 27... Nxc4 {1} 28. Rh4 {0.25/12 80} 28... Bd3 {70} 29. Rxh3 {
>0.47/13 2} 29... Bxe2 {153} 30. Rxe2 {0.16/12 150} 30... Rde8 {1} 31. Rh4 {
>0.22/12 480} 31... Nxe3 {330} 32. fxe3 {0.16/13 81} 32... Bc5 {187} 33. Nd4 {
>-0.03/12 416} 33... g6 {1} 34. b4 {-0.09/11 154} 34... Bd6 {320} 35. Rd2 {
>-0.12/11 190} 35... Kxb7 {493} 36. Kf2 {-0.03/12 9} 36... Re5 {1010} 37. Nf3 {
>-0.03/12 267} 37... Re6 {1} 38. Rhd4 {0.09/11 248} 38... Rd8 {1} 39. R4d3 {
>0.13/11 234} 39... Kc7 {358} 40. Nd4 {0.31/11 164} 40... Re5 {205} 41. Rc2+ {
>0.53/12 23} 41... Kd7 {353} 42. a4 {0.47/13 169} 42... Rc8 {141} 43. Rcd2 {
>0.81/13 356} 43... Ke7 {1} 44. b5 {0.66/13 410} 44... Rec5 {1} 45. Kf3 {
>0.59/12 322} 45... Kf8 {165} 46. Nc6 {0.22/12 126} 46... Rf5+ {1} 47. Ke2 {
>0.06/13 117} 47... Bc5 {79} 48. a5 {0.13/13 256} 48... Re8 {555} 49. Rd8 {
>0.84/13 157} 49... Rxd8 {175} 50. Rxd8+ {0.84/16 1} 50... Kg7 {1} 51. b6 {
>0.66/14 123} 51... axb6 {337} 52. a6 {0.13/16 201} 52... b5 {1} 53. a7 {
>-0.09/14 151} 53... Bxa7 {15} 54. Nxa7 {-0.25/14 116} 54... Rxg5 {117} 55. Rd4
>{-0.22/13 18} 55... Rg2+ {114} 56. Kd3 {0.00/15 243} 56... Rb2 {141} 57. Kc3 {
>-0.37/14 396} 57... Rf2 {1} 58. Nxb5 {-0.12/13 179} 58... g5 {106} 59. Rd5 {
>-0.03/12 4} 59... Kg6 {242} 60. Rd8 {-0.09/12 2} 60... Kf5 {185} 61. Nd6+ {
>0.06/13 444} 61... Ke5 {1} 62. Nc4+ {0.22/12 227} 62... Ke4 {142} 63. Re8+ {
>0.22/12 63} 63... Kf3 {292} 64. Rg8 {0.22/13 44} 64... f6 {219} 65. Kd3 {
>0.00/14 261} 65... h4 {409} 66. Rg6 {-0.41/15 359} 66... Kg2 {282} 67. e4 {
>-0.44/15 258} 67... h3 {1} 68. Ne3+ {-0.25/14 125} 68... Kg3 {82} 69. Nf5+ {
>-0.31/14 49} 69... Rxf5 {175} 70. exf5 {-1.31/20 47} 70... h2 {123} 71. Rh6 {
>-5.81/17 429} 71... Kg2 {1} 72. Ke3 {-1.44/17 716} 72... h1=Q {1} 73. Rxh1 {
>-7.53/15 95} 73... Kxh1 {69} 74. Kf3 {-7.69/28 38} 74... Kh2 {196} 75. Kg4 {
>-8.41/28 97} 75... Kg2 {156} 76. Kh5 {-5.84/4 0} 76... Kf3 {630} 77. Kg6 {
>-7.75/26 0} 77... g4 {383} 78. Kxf6 {-7.87/24 226} 78... g3 {1} 79. Ke7 {
>-8.34/20 215} 79... g2 {1} 0-1
Let's see ...
80. f6 g1=Q 81. f7 Qg5+ 82. Ke8 Qe5+ 83. Kf8 Kg4 84.
Kg8 Qg5+ 85. Kh8 Qf6+ 86. Kg8 Qg6+ 87. Kh8 Qf6+ 88. Kg8
I see that this position is a theorical draw ...
I know that Chess System Tal have lot of knowledge , now you can had this draw
in ;)
My opinion is : The knowledge based programs are the futur (Chess System Tal,
Diep, ...) but now they needs still some months to mature and in the meantime
the classic program will improve too, but I think that the knowledge based
programs will win the race ...
An interresting paradox is : when game is draw Fritz display -8.34 :)
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.