Topic Rybka Support & Discussion / Rybka Discussion / handicaps for 32 piece tbs to im's or anyone else
If there is ever a 32 piece tbs,what handicap would it need for kasparov/kramnik to be confident of winning 90% of the time. knight ? rook ? queen ?
duncan
duncan
![[il]](/mwf/v12/flags/il.png "Israel")
32 piece tables is clearly not the way to beat humans in chess.
I do not understand why do you think that it will play better than 1.Na3 Na6 2.Nb1 Nb8 assuming that no tbs mistake is in this game.
Note that tb mistake is different than practical mistake.
5 piece tables can make stupid blunder by sacrificing a rook in KRB vs KR so beginners can get easy draw against it and I expect 32 piece tablebases to do similiar mistakes.
Uri
I do not understand why do you think that it will play better than 1.Na3 Na6 2.Nb1 Nb8 assuming that no tbs mistake is in this game.
Note that tb mistake is different than practical mistake.
5 piece tables can make stupid blunder by sacrificing a rook in KRB vs KR so beginners can get easy draw against it and I expect 32 piece tablebases to do similiar mistakes.
Uri
Your comment is true if the user of the 32-man TBs just looks at win-draw-lose information. A better way to use these TBs would be to generate a tree and pick the move in the best move category that gives the opponent the highest percentage chance of deviating to a worse result. As a first approximation, this could be done by counting nodes at some constant depth.
Alan
Alan
ok then what about a hypothetical computer that can look 200 moves ahead ?
duncan
duncan
![[mx]](/mwf/v12/flags/mx.png "Mexico")
> ok then what about a hypothetical computer that can look 200 moves ahead ?
If it sees that with perfect defense, d4, e4 and f3 lead to the same kind of drawish position, it could play f3 and give an easier game to the opponent. I think that there's a point in where too much depth can be harmful, as the engine starts seeing "ghost" moves in the position (For example, a very hard to find move in 60 plies that makes the engine not play the best move.) This was observed in the game Checkers, before it was solved (That the engine would perform better at 19 ply than at 21 ply, I think that reaching higher plies was causing the engine to be more pessimistic and have more draws, but I could be wrong.)
I do not see why.
If both 1.d4 and 1.e4 and 1.f3 lead to draw and if 1.d4 is the choice at small depth because it leads to small advantage then there is no reason that it changes it's mind to 1.f3
Even if it can see draw score for 1.d4 it is not going to play 1.f3 unless it can see positive score for 1.f3
Uri
If both 1.d4 and 1.e4 and 1.f3 lead to draw and if 1.d4 is the choice at small depth because it leads to small advantage then there is no reason that it changes it's mind to 1.f3
Even if it can see draw score for 1.d4 it is not going to play 1.f3 unless it can see positive score for 1.f3
Uri
My idea goes like this:
It sees that f3 is worth 0.2, d4 worth 0.4 and e4 worth 0.5 (for example.)
It finds a deep refutation for e4 and now the score for it is 0.0, so it prefers d4 now.
It finds a deep refutation for d4 and now the score for d4 is 0.0 as well, and we have:
f3 0.2
d4 0.0
e4 0.0
By the time it finds the refutation of f3, and makes its score 0.0, then the order doesn't have a reason to change (Because all moves are equal.) Then I think that the chosen move may be selected at random, or by the order of finding refutation, or some other factor, but if e4 is the optimal move, the engine would have problems playing it as at some point all moves look as "bad".
It sees that f3 is worth 0.2, d4 worth 0.4 and e4 worth 0.5 (for example.)
It finds a deep refutation for e4 and now the score for it is 0.0, so it prefers d4 now.
It finds a deep refutation for d4 and now the score for d4 is 0.0 as well, and we have:
f3 0.2
d4 0.0
e4 0.0
By the time it finds the refutation of f3, and makes its score 0.0, then the order doesn't have a reason to change (Because all moves are equal.) Then I think that the chosen move may be selected at random, or by the order of finding refutation, or some other factor, but if e4 is the optimal move, the engine would have problems playing it as at some point all moves look as "bad".
I undertand your idea but I think that finding a refutation to 1.f3 is easier then finding a refutation to 1.d4 so the case that you suggest is not going to happen.
Uri
Uri
> I undertand your idea but I think that finding a refutation to 1.f3 is easier then finding a refutation to 1.d4 so the case that you suggest is not going to happen.
Yes, but it depends on the implementation. If the engine is only interested in the main move first, it may search more thoroughly e4 until it finds a refutation, and then not find a refutation for d4 until the not so thoroughly searched f3 appears over e4. Then the same can happen for d4 and the engine may not start searching thoroughly f3 until it's clear first. In the end, the engine could be finding refutations backwards (Stronger moves look more interesting so the engine finds their refutation first) and get a bad move ordering.
I don't know how a 200 ply engine would work but this is a possibility. And we're talking about the opening position only, in where for the human eyes, e4 and d4 are clearly better than f3, but on later positions the engine could end playing really dumb moves just for finding refutations in the wrong order, so, for it to be stronger, the programmer has to have this and other factors in mind, not just reach 200 ply "brutally." (And the optimizations such engine would need aren't clear.)
![[cr]](/mwf/v12/flags/cr.png "Costa Rica")
I think that if Rybka 3 could search 200 moves ahead, difficultly it would lose.
If you suppose that even some opponent can be stronger, I think that it would not be much problem for Rybka. I believe that even this way Rybka would find - in a very high percentage- almost always some path to save the game and to make a draw..
In fact, 200 moves is a very high number, I think it should be much less.
Regards,
Gaмßito.
If you suppose that even some opponent can be stronger, I think that it would not be much problem for Rybka. I believe that even this way Rybka would find - in a very high percentage- almost always some path to save the game and to make a draw..
In fact, 200 moves is a very high number, I think it should be much less.
Regards,
Gaмßito.
I think that Rybka can already see up to 200 moves, with extensions. But are we talking about a 200 ply brute force Rybka*? To beat it you need a 201 move tactic that sees beyond 200 move Rybka, and with pruning you can find this tactic much sooner than Rybka looks at all the possible 200 moves with brute force (After all, the major part of the moves that need to be searched for a depth of 200 moves are garbage.)
(*) Or maybe 400 ply, if we're talking about 400 halfmoves = 200 moves.
(*) Or maybe 400 ply, if we're talking about 400 halfmoves = 200 moves.
I think that you are wrong and rybka cannot see 200 moves forward even with extensions but only Vasik knows.
Most program will never see 200 plies forward regardless of depth simply because they are not designed to do it.
Rybka's depth is limited to something like 60(do not remember exactly the maximal depth) and I guess that rybka's extensions are also limited.
Rybka mainly does pruning and not extensions.
Uri
Most program will never see 200 plies forward regardless of depth simply because they are not designed to do it.
Rybka's depth is limited to something like 60(do not remember exactly the maximal depth) and I guess that rybka's extensions are also limited.
Rybka mainly does pruning and not extensions.
Uri
Maybe 200 ply is just too high of a number and outside of reality, and we should be discussing lower numbers that make more sense.
what I was trying to ask is if a hypothetical computer could play perfect chess. that is if the best move will give it a .2 of an advantage in 200 or 2000 moves, it will see it and go for it, how much material loss would you have to handicap it for kasparov to defeat it regularly.
duncan
duncan
![[de]](/mwf/v12/flags/de.png "Germany")
A knight or a bishop would be not enough I think, a rook would do it.
Assuming, that the 31 piece tables play always the move which lead to the longest distance to mate (assuming that home position without knight is mate :-)).
Assuming, that the 31 piece tables play always the move which lead to the longest distance to mate (assuming that home position without knight is mate :-)).
I believe that knight or bishop is going to be enough and super GM will be able practically to win when they have one knight more than their opponent.
Any idea about DTM (distance to mate) for home position without Nb1/Nb8 with perfect play bothsides? Less than 30, 50, 75 or 100 moves?
Please assume perfect player has White without Nb1. How many moves need 2800 Elo player with Black on average for practical win?
Please assume perfect player has White without Nb1. How many moves need 2800 Elo player with Black on average for practical win?
I feel sure that it is more than 30 moves and I also believe that it is more than 50 moves
Trying one blitz rybka-rybka game black mated in 72 moves.
[Event "Shootout (Rybka232a32-bit, Blitz:4'+2") "]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "Rybka 2.3.2a 32-bit"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "144"]
1. e4 {-1.79/12 9} Nf6 {-1.75/13 12} 2. e5 {-1.76/13 4} Nd5 {-1.75/13 2} 3. Bc4
{-1.70/13 4} Nb6 {-1.79/14 7} 4. Bb3 {-1.80/14 8} d5 {-1.80/15 56} 5. exd6 {
-1.80/15 56} Qxd6 {-1.84/14 2} 6. Nf3 {(Ng1-e2) -2.00/15 32} a5 {-1.95/15 11}
7. c3 {-1.95/15 12} Qf6 {(a5-a4) -1.93/15 33} 8. h3 {-2.00/15 22} Bf5 {
-2.01/15 9} 9. O-O {-1.96/15 8} Bd3 {-2.02/15 10} 10. Re1 {-1.87/15 7} Nc6 {
-1.88/15 12} 11. Re3 {-1.95/15 7} O-O-O {-1.96/15 16} 12. Nd4 {-1.96/15 12} Ne5
{-2.06/15 19} 13. Qe1 {-2.04/15 6} Nbd7 {-2.11/15 19} 14. Ba4 {-2.07/15 18} c5
{-2.07/15 6} 15. Bxd7+ {-2.03/14 7} Rxd7 {-2.07/16 5} 16. Rxe5 {-2.11/15 15}
cxd4 {-2.15/14 1} 17. Rxa5 {(b2-b3) -2.29/11 1} b6 {-2.50/12 5} 18. Ra3 {
(Ra5-a4) -2.53/12 2} Rd5 {-2.56/12 2} 19. cxd4 {-2.55/13 3} Qxd4 {-2.60/13 2}
20. Ra8+ {-2.62/13 5} Kd7 {-2.37/13 0} 21. Ra7+ {-2.68/13 3} Kc6 {-2.66/14 3}
22. Qe3 {-2.66/14 4} Qe5 {-2.67/14 5} 23. Qxe5 {-2.76/14 10} Rxe5 {-2.54/13 0}
24. b3 {-2.85/14 3} Re1+ {-2.85/14 5} 25. Kh2 {-2.58/13 0} f6 {-2.80/13 0} 26.
Bb2 {-2.85/14 4} Rxa1 {-2.88/14 3} 27. Bxa1 {-2.93/15 4} e5 {-2.92/14 12} 28.
Bb2 {-2.86/14 4} h5 {-2.84/14 4} 29. Kg1 {-2.98/14 7} Rg8 {-2.96/14 9} 30. Ra8
{-2.99/14 1} Bh7 {-3.04/14 2} 31. d4 {-3.06/14 2} Bd6 {-3.06/14 1} 32. Rxg8 {
-3.10/14 2} Bxg8 {-2.85/13 0} 33. dxe5 {-3.20/14 1} fxe5 {-2.90/13 0} 34. f3 {
-3.25/14 4} b5 {(Bg8-h7) -3.44/14 3} 35. Kf2 {-3.56/15 2} b4 {-3.54/15 2} 36.
Ke2 {-3.54/15 0} Bh7 {-3.67/15 2} 37. h4 {-3.67/15 0} Bb1 {(Bh7-f5) -3.90/15 2}
38. Kd1 {-3.76/16 4} Kd5 {-4.22/15 3} 39. Kc1 {-4.25/15 1} Bf5 {-3.85/14 0} 40.
Kd1 {-4.42/16 1} Be7 {-4.00/15 0} 41. g3 {-4.51/15 0} Bc5 {-4.58/15 0} 42. Ke2
{-4.67/16 1} Bb1 {-4.71/16 0} 43. Kd1 {-4.95/17 1} Bf2 {-4.98/17 1} 44. g4 {
-4.98/13 1} hxg4 {-4.98/16 1} 45. fxg4 {-4.98/17 4} Bxh4 {(Bb1-e4) -5.25/15 2}
46. a3 {(Bb2-c1) -5.48/13 2} bxa3 {-5.68/14 1} 47. Bxa3 {-6.09/15 2} e4 {
-6.09/14 1} 48. Bb4 {-6.09/13 1} Kd4 {-6.29/14 2} 49. Ba5 {-6.93/14 4} Bg5 {
(Kd4-d3) -6.96/15 11} 50. Bd2 {(b3-b4) -6.94/15 2} e3 {-6.94/13 0} 51. Be1 {
-7.97/16 3} Kd3 {-7.97/14 2} 52. Kc1 {-7.97/14 1} Bc2 {-7.80/12 0} 53. Kb2 {
-8.30/13 2} Bf6+ {(e3-e2) -7.70/11 2} 54. Ka3 {(Kb2-c1) -8.07/12 2} Bc3 {
(Bf6-e5) -8.64/13 2} 55. Bh4 {(Be1-g3) -11.78/14 4} Ke4 {(e3-e2) -14.54/13 5}
56. b4 {-14.40/12 4} Kf4 {-14.54/11 3} 57. b5 {(g4-g5) -14.40/9 1} g5 {
-15.85/10 2} 58. Be1 {-16.08/10 3} Bxe1 {-15.98/8 0} 59. b6 {-16.07/8 1} e2 {
-16.07/7 1} 60. b7 {-16.07/8 2} Bb4+ {-4.62/8 0} 61. Kxb4 {(Ka3-b2) -16.07/8 1}
e1=Q+ {-16.07/7 1} 62. Kc4 {-16.36/8 4} Qb1 {(Qe1-e5) -16.32/7 3} 63. Kc3 {
-16.32/6 2} Be4 {(Qb1-b3+) -16.96/7 1} 64. b8=N {(b7-b8Q+) -17.29/7 2} Qxb8 {
-17.30/7 1} 65. Kc4 {-26.76/8 1} Kxg4 {-26.76/7 1} 66. Kd4 {-#9/7 0} Kf4 {
-#8/6 0} 67. Kc4 {-#8/5 0} g4 {-#7/4 0} 68. Kc3 {-#6/3 0} g3 {-#5/3 0} 69. Kd2
{(Kc3-c4) -#4/3 0} g2 {(Qb8-b4+) -#4/3 0} 70. Kc3 {(Kd2-e2) -#3/3 0} g1=Q {
-#3/3 0} 71. Kc4 {-#2/3 0} Qc1+ {-#2/3 0} 72. Kd4 {-#1/3 0} Qb4# {-#1/3 0} 0-1
Uri
Trying one blitz rybka-rybka game black mated in 72 moves.
[Event "Shootout (Rybka232a32-bit, Blitz:4'+2") "]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "Rybka 2.3.2a 32-bit"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "144"]
1. e4 {-1.79/12 9} Nf6 {-1.75/13 12} 2. e5 {-1.76/13 4} Nd5 {-1.75/13 2} 3. Bc4
{-1.70/13 4} Nb6 {-1.79/14 7} 4. Bb3 {-1.80/14 8} d5 {-1.80/15 56} 5. exd6 {
-1.80/15 56} Qxd6 {-1.84/14 2} 6. Nf3 {(Ng1-e2) -2.00/15 32} a5 {-1.95/15 11}
7. c3 {-1.95/15 12} Qf6 {(a5-a4) -1.93/15 33} 8. h3 {-2.00/15 22} Bf5 {
-2.01/15 9} 9. O-O {-1.96/15 8} Bd3 {-2.02/15 10} 10. Re1 {-1.87/15 7} Nc6 {
-1.88/15 12} 11. Re3 {-1.95/15 7} O-O-O {-1.96/15 16} 12. Nd4 {-1.96/15 12} Ne5
{-2.06/15 19} 13. Qe1 {-2.04/15 6} Nbd7 {-2.11/15 19} 14. Ba4 {-2.07/15 18} c5
{-2.07/15 6} 15. Bxd7+ {-2.03/14 7} Rxd7 {-2.07/16 5} 16. Rxe5 {-2.11/15 15}
cxd4 {-2.15/14 1} 17. Rxa5 {(b2-b3) -2.29/11 1} b6 {-2.50/12 5} 18. Ra3 {
(Ra5-a4) -2.53/12 2} Rd5 {-2.56/12 2} 19. cxd4 {-2.55/13 3} Qxd4 {-2.60/13 2}
20. Ra8+ {-2.62/13 5} Kd7 {-2.37/13 0} 21. Ra7+ {-2.68/13 3} Kc6 {-2.66/14 3}
22. Qe3 {-2.66/14 4} Qe5 {-2.67/14 5} 23. Qxe5 {-2.76/14 10} Rxe5 {-2.54/13 0}
24. b3 {-2.85/14 3} Re1+ {-2.85/14 5} 25. Kh2 {-2.58/13 0} f6 {-2.80/13 0} 26.
Bb2 {-2.85/14 4} Rxa1 {-2.88/14 3} 27. Bxa1 {-2.93/15 4} e5 {-2.92/14 12} 28.
Bb2 {-2.86/14 4} h5 {-2.84/14 4} 29. Kg1 {-2.98/14 7} Rg8 {-2.96/14 9} 30. Ra8
{-2.99/14 1} Bh7 {-3.04/14 2} 31. d4 {-3.06/14 2} Bd6 {-3.06/14 1} 32. Rxg8 {
-3.10/14 2} Bxg8 {-2.85/13 0} 33. dxe5 {-3.20/14 1} fxe5 {-2.90/13 0} 34. f3 {
-3.25/14 4} b5 {(Bg8-h7) -3.44/14 3} 35. Kf2 {-3.56/15 2} b4 {-3.54/15 2} 36.
Ke2 {-3.54/15 0} Bh7 {-3.67/15 2} 37. h4 {-3.67/15 0} Bb1 {(Bh7-f5) -3.90/15 2}
38. Kd1 {-3.76/16 4} Kd5 {-4.22/15 3} 39. Kc1 {-4.25/15 1} Bf5 {-3.85/14 0} 40.
Kd1 {-4.42/16 1} Be7 {-4.00/15 0} 41. g3 {-4.51/15 0} Bc5 {-4.58/15 0} 42. Ke2
{-4.67/16 1} Bb1 {-4.71/16 0} 43. Kd1 {-4.95/17 1} Bf2 {-4.98/17 1} 44. g4 {
-4.98/13 1} hxg4 {-4.98/16 1} 45. fxg4 {-4.98/17 4} Bxh4 {(Bb1-e4) -5.25/15 2}
46. a3 {(Bb2-c1) -5.48/13 2} bxa3 {-5.68/14 1} 47. Bxa3 {-6.09/15 2} e4 {
-6.09/14 1} 48. Bb4 {-6.09/13 1} Kd4 {-6.29/14 2} 49. Ba5 {-6.93/14 4} Bg5 {
(Kd4-d3) -6.96/15 11} 50. Bd2 {(b3-b4) -6.94/15 2} e3 {-6.94/13 0} 51. Be1 {
-7.97/16 3} Kd3 {-7.97/14 2} 52. Kc1 {-7.97/14 1} Bc2 {-7.80/12 0} 53. Kb2 {
-8.30/13 2} Bf6+ {(e3-e2) -7.70/11 2} 54. Ka3 {(Kb2-c1) -8.07/12 2} Bc3 {
(Bf6-e5) -8.64/13 2} 55. Bh4 {(Be1-g3) -11.78/14 4} Ke4 {(e3-e2) -14.54/13 5}
56. b4 {-14.40/12 4} Kf4 {-14.54/11 3} 57. b5 {(g4-g5) -14.40/9 1} g5 {
-15.85/10 2} 58. Be1 {-16.08/10 3} Bxe1 {-15.98/8 0} 59. b6 {-16.07/8 1} e2 {
-16.07/7 1} 60. b7 {-16.07/8 2} Bb4+ {-4.62/8 0} 61. Kxb4 {(Ka3-b2) -16.07/8 1}
e1=Q+ {-16.07/7 1} 62. Kc4 {-16.36/8 4} Qb1 {(Qe1-e5) -16.32/7 3} 63. Kc3 {
-16.32/6 2} Be4 {(Qb1-b3+) -16.96/7 1} 64. b8=N {(b7-b8Q+) -17.29/7 2} Qxb8 {
-17.30/7 1} 65. Kc4 {-26.76/8 1} Kxg4 {-26.76/7 1} 66. Kd4 {-#9/7 0} Kf4 {
-#8/6 0} 67. Kc4 {-#8/5 0} g4 {-#7/4 0} 68. Kc3 {-#6/3 0} g3 {-#5/3 0} 69. Kd2
{(Kc3-c4) -#4/3 0} g2 {(Qb8-b4+) -#4/3 0} 70. Kc3 {(Kd2-e2) -#3/3 0} g1=Q {
-#3/3 0} 71. Kc4 {-#2/3 0} Qc1+ {-#2/3 0} 72. Kd4 {-#1/3 0} Qb4# {-#1/3 0} 0-1
Uri
![[us]](/mwf/v12/flags/us.png "United States")
In spite of the fact that a 2000-rated player cannot generally beat a 3000-rated computer at knight/bishop odds, I think that a 2800-rated player can regularly beat a 3800-rated (perfect) player at those odds because of the fact that the position is theoretically won for the side with the extra piece, and the theoretical evaluation of a position corresponds more and more with the tendency of game outcomes as the ratings get higher.
This is clearly true. The higher up the rating scale you go, the smaller are the errors that typically decide games, and therefore the proper handicap for a given rating difference gets smaller as you go higher.
32 piece tables is clearly not the way to beat humans in chess.
But this is not the question! 31! :-) piece tables should not win but not loose more than 89%.
But this is not the question! 31! :-) piece tables should not win but not loose more than 89%.
31-piece (or 30-piece) tablebases will never lose against any opponent if they are equipped with a half-decent computer (which will obviously exist if we're talking about such things)--trades can be forced in the opening.
Topic Rybka Support & Discussion / Rybka Discussion / handicaps for 32 piece tbs to im's or anyone else
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill