Author: John Merlino
Date: 13:19:02 10/22/02
Go up one level in this thread
On October 21, 2002 at 23:48:57, Robert Hyatt wrote:
>On October 21, 2002 at 21:34:12, John Merlino wrote:
>
>>On October 21, 2002 at 18:31:50, Dieter Buerssner wrote:
>>
>>>On October 21, 2002 at 00:00:42, Terry Ripple wrote:
>>>
>>>>I heard talk that if you don't have the complete set (all pawns) with pieces,
>>>>that it could very well do worst for the program than do better for it, is this
>>>>at all true? I'am especially interested in the 6 men as i have all the complete
>>>>5 men and now would like the 6 man, but i hear that it's not complete as yet!
>>>
>>>All available 6-men TBs are pawnless. In this case, you don't have to fear an
>>>problems. The typical problems arise, when there is a pawn, that must be
>>>promoted, and the TB with the promoted pawn is missing. This however can not
>>>happen with the available 6-men TBs. You need the 5-men TBs, so that after any
>>>possible capture, a TB position will be found.
>>>
>>>Regards,
>>>Dieter
>>
>>I'm pretty sure that Marc Bourzutschky is in the process of creating 6-man files
>>with pawns for Chessmaster 9000. I have no idea how long they will take, nor do
>>I know which files he's planning on building first....
>>
>>He posted a thread here which showed his build times for quite a few 6-man
>>files, but the only one I remember was 29 hours for KQQKQQ, and I don't remember
>>the CPU either :-(
>>
>>jm
>
>29 hours sounds incredibly fast, since it is going to require at _least_ 100
>iterations to find most of the draws...
Maybe so. I only have a report from Guy Haworth to go on it, in which he said:
[Your 29hrs for KQQKQQ compares with EN's 2 days for generation+validation on a
500MHz machine, maybe 1 day on a 1GHz machine.]
Additionally, thanks to Paul who sent me the contents of the above-mentioned
thread, here are the contents of that thread:
******************************************************************************
The Chessmaster FEG program (downloadable from www.chessmaster.com) is a very
clever tool that allows the generation of 6-man EGDB's with pawns on a 32 bit
machine with limited resource requirements. This would not be possible with
the Nalimov program without some code changes. I have generated KRPPKR which
occurs relatively frequently in practice (next I'll generate KRPKRP and KRKPPP.)
Below some positions that are misanalyzed in ECE, and also give Fritz trouble
even when using 5-man tablebases. Unfortunately the file format of the FEG
tablebases is not public, so no automated search for mutual zugzwang
positions nor a systematic comparison with Nalimov's results is yet possible.
The position below is from ECE II, 468 and has some subtle zuzwang play:
[D] 4r3/8/5K2/2Pk4/3P1R2/8/8/8 b
[White "?"]
[Black "?"]
[Fen "4r3/8/5K2/2Pk4/3P1R2/8/8/8 b - - 0 1"]
[Result "1-0"]
1... Rf8+ 2. Kg5 Rg8+ 3. Kh5 Rg7 4. Rg4 Rf7 5. Kg6 Rf8 6. Kh6 ({ECE has only}
6. Rh4 Rf1 7. Kg5 Rf8 8. Rf4 Rg8+ 9. Kh5 Rg7 {and claims a draw}) 6... Rf1
7. Kg5 Ke6 8. Re4+ Kd5 9. Rf4 Rd1 10. Kf6 Re1 11. Kf7 Re2 12. Kf8 Re6 13. Rh4
Re1 14. Rh7 Re6 15. Rd7+ Kc6 16. Rd8 Rh6 17. Kg7 Rh5 18. Rd6+ Kc7 19. d5 Rg5+
20. Kf6 Rh5 21. Kf7 Rg5 22. Ke6 Rg6+ 23. Ke7 Rg7+ 24. Kf6 Rh7 25. Rc6+ Kd7
26. Re6 Rh6+ 27. Ke5 Rh5+ 28. Kd4 1-0
The position below could have arisen in Blackburne-Lasker (ECE II, 533):
[D] 8/8/7k/6pp/2R5/1r6/6K1/8 w
[Event "St Petersburg"]
[Date "1914.??.??"]
[White "Blackburne"]
[Black "Lasker*"]
[Result "0-1"]
[FEN "8/8/7k/6pp/2R5/1r6/6K1/8 w - - 0 1"]
1. Rc6+ Kg7 2. Rc5 Kf6 3. Rc6+ Ke7 ({After} 3... Ke5 4. Rc5+ Kf4 5. Rc4+ Ke3
6. Rc5 {
Black has to retrace his steps. ECE therefore considers the position drawn.})
4. Rc5 Rb2+ 5. Kf3 g4+ 6. Kf4 Rf2+ 7. Kg5 (7. Kg3 Rf3+ 8. Kg2 h4 9. Rg5 h3+
10. Kh2 Rf2+ 11. Kg1 Rf4 12. Kh2 Kf6 13. Rg8 Kf5 {also loses}) 7... Rh2 8. Kf4
Rh3 9. Rc6 g3 10. Kf3 h4 11. Kg2 Rh2+ 12. Kg1 Rf2 ({Fritz7 needs 25 ply to
avoid} 12... Kf7 13. Rh6 {with a draw}) 13. Rh6 Rf4 14. Kg2 Kf7 15. Kh3 Rg4
(15... Kg7 {also wins}) 16. Kg2 Kg7 17. Rh5 Kf6 18. Rh8 Rf4 19. Rf8+ Ke5
20. Re8+ Kd5 21. Rd8+ Ke6 22. Re8+ Kf7 23. Re2 Kf6 24. Kh3 Rd4 25. Ra2 Re4
26. Ra8 Re1 0-1
The following position by Minev (ECE II, 564) features an interesting stalemate
defence:
[D] 8/7k/7P/5PK1/8/8/6R1/2r5 w
[White "?"]
[Black "?"]
[Result "1/2-1/2"]
[FEN "8/7k/7P/5PK1/8/8/6R1/2r5 w - - 0 1"]
1. f6 Rc5+ 2. Kf4 Ra5 {Fritz7 has White ahead by 5.25 after 26 plies.} 3. f7
Ra4+ 4. Kf5 Ra5+ 5. Kg4 Ra4+ 6. Kh5 Rf4 7. Rg7+ Kh8 8. Kg5 (8. Rg8+ Kh7 9. f8=Q
Rxf8 10. Rxf8 {stalemate}) 8... Rf5+ {with repeated check or stalemate.}
1/2-1/2
The position below is by Kopayev (ECE II, 586). White can win, but not in the
manner given in ECE:
[D] 8/r7/8/5k1P/6R1/5PK1/8/8 b
[White "?"]
[Black "?"]
[Result "1-0"]
[FEN "8/r7/8/5k1P/6R1/5PK1/8/8 b - - 0 1"]
1... Kf6 2. h6 Rh7 3. Ra4 (3. Rh4 {only draws after} Kf7 4. Kg4 Kg8 5. Kh5
Ra7 6. Rg4+ Kh8 7. f4 Ra5+ ({ECE only considers} 7... Ra1 {which loses after
} 8. f5 Kh7 9. Rg7+ Kh8 10. Rg4 Kh7 11. f6 Rh1+ 12. Kg5 Rxh6 13. f7 ) 8. Rg5
Ra1 9. f5 (9. Rb5 Rh1+ 10. Kg6 Rg1+ 11. Rg5 Rh1 12. Ra5 Rg1+ 13. Kf6 Kh7 {=})
9... Rh1+ 10. Kg6 Rxh6+ {=}) 3... Kg6 4. Ra6+ Kh5 5. Rb6 (5. Ra1 {!?} Kg6
6. Rh1 {?} Kf7 {!} {With the same draw as after 3.Rh4?}) 5... Kg5 (5... Ra7
6. f4 Ra3+ 7. Kf2 Kg4 8. Rf6 Rh3 9. Ke2 Rh1 10. Ke3 Re1+ 11. Kd4 Re7 12. Kd5
Kh5 13. Kd6 Rh7 14. f5 Kg5 15. Ke6 Ra7 16. Rg6+ Kh5 17. Kf6 Ra6+ 18. Kg7 Ra7+
19. Kh8 {+-}) 6. f4+ Kf5 7. Rb5+ Kg6 8. Kg4 Ra7 9. Rb6+ Kh7 10. Kg5 Ra1
(10... Rc7 11. Re6 Ra7 12. f5 {+-}) 11. Rb7+ Kg8 12. f5 Rg1+ 13. Kf6 Ra1
14. Re7 Ra2 15. Ke5 Re2+ 16. Kd6 Rd2+ 17. Ke6 Re2+ 18. Kd7 Rd2+ 19. Ke8 Rf2
20. Re5 Kh7 21. Kf7 Ra2 22. Re6 Ra8 23. Re8 1-0
The position below features interesting play with a bishop of the "wrong" color.
It is due to Minev and is misanalyzed in ECE III, 1196:
[D] 1r5k/R6P/6BK/8/8/8/8/8 w
[White "?"]
[Black "?"]
[Result "1-0"]
[FEN "1r5k/R6P/6BK/8/8/8/8/8 w - - 0 1"]
1. Ra1 Rd8 2. Bf5 {White's winning plan is to bring the bishop to the a2-g8
diagonal and threaten a deadly rook check on g8. At the same time White has to
ensure that he is left with a winning KRBKR ending if Black captures the pawn
on h7. Timman in ECE only gives 2.Bf7?, when Black can draw with 2... Rd6+ 3.
Kg5 Kxh7.} Rf8 3. Kg5 Re8 (3... Kg7 4. h8=Q+ {leads to a won KRBKR ending.})
4. Bh3 {In the sequel Black has several chances of capturing the h7 pawn,
which all lead to lost KRBKR endings.} Rd8 5. Be6 Rb8 6. Kg6 Rf8 7. Ra2 Rc8 8.
Kf7 Rb8 9. Ke7 Rb7+ 10. Kf6 Ra7 11. Rd2 Ra8 12. Rg2 1-0
******************************************************************************
Below is a list of the endings I have generated so far, along with
some statistics. The number of seconds is somewhat misleading because
I used 3 different computers: a 1.9GHZ and 2.4GHZ P4, and an an old
800MHZ P3. For example, the KQRKQR ending was generated on the old
machine which is why it took so long. I started this whole project
September 12.
As far as dependencies are concerned, FEG automatically stops when
dependent files are missing, so you can only generate endgames with
pawns if all possible promotions have been created.
The main difference to the Nalimov scheme is that each EGDB only
contains WTM to win or not to win, not distinguishing between a draw
or a loss. So to get comparable info one needs to generate both
KXKY and KYKZ. For KRPPKR I have not yet generated KRKRPP because
I was mainly interested whether White can win or not, ignoring
pathological positions where White actually loses. I'll get KRKRPP
later anyway as a byproduct of generating KRKPPP, which is in the
works. Because KRKPPP can go either way, I'll also generate KPPPKR.
Ending # Secs WTM Size WTM Ply BTM Size BTM Ply
------------------------------------------------------------
KQQRKR 248337 181541366 73 454405666 74
KQRBKR 81755 705485974 135 1060868934 136
KQRKQR 510320 1161463938 233 426465391 234
KQRKRB 266976 1044489055 141 1252885945 142
KQRKRN 265612 1025129450 133 1353459632 138
KQRKRR 424856 696088906 113 563970032 112
KQRNKR 110866 678962160 87 1023118760 88
KQRPKR 329993 2069855830 149 3746111779 150
KQRRKR 81998 246328793 81 683209859 82
KRBBKR 121787 815254179 139 795943021 140
KRBKQR 54376 60403494 129 7176017 128
KRBKRB 62467 558544458 129 68163464 128
KRBKRR 56590 46934987 133 6850121 132
KRBNKR 131406 1621032003 135 1249681979 136
KRBPKR 311575 4691404903 151 4585986662 152
KRKQBB 19942 6021257 61 5967701 6
KRKQBN 23232 7374862 79 6402464 60
KRKQBP 67545 49797631 77 20721167 68
KRKQNN 24266 8042923 81 5996962 68
KRKQNP 104173 123187152 133 21034570 126
KRKQPP 80602 387265124 83 20308922 78
KRKQQB 17771 5799312 51 5845019 6
KRKQQN 21387 5884460 77 5856010 6
KRKQQP 54465 20018808 55 18789740 6
KRKQQQ 9890 5648101 7 5628973 6
KRKQQR 9686 5760118 7 5794036 6
KRKQRB 18535 6157489 53 6334517 6
KRKQRN 22505 6409825 79 6305985 6
KRKQRP 52386 26195616 57 20419355 48
KRKQRR 9968 5839264 7 5884821 6
KRNKQR 31582 40756874 73 6891222 70
KRNKRB 181544 532653480 73 47539498 72
KRNKRN 182437 633633087 103 59911833 102
KRNKRR 53265 31247335 73 6607222 70
KRNPKR 322138 4959158626 169 4558331725 170
KRPKQR 257582 660773785 235 33211534 224
KRPKRR 220795 734378250 149 72391611 148
KRPPKR 258491 1806511454 189 2200757722 190
KRRBKR 82773 629065508 137 559091788 138
KRRKQR 49265 277825100 65 35754296 64
KRRKRB 83225 614558904 133 150419097 132
KRRKRN 138858 753078894 177 315051543 178
KRRKRR 61544 258742167 79 45661712 78
KRRNKR 74479 613173287 89 517953298 90
KRRPKR 250624 1739999970 149 2354960874 150
KRRRKR 83699 139427358 65 264648675 66
******************************************************************************
jm
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.