Little Goliath Evolution bug

[Guenther Simon]

New LG has still hash key bugs like the previous version.
In the position below it will always play ...Qa7?? without thinking.

Guenther

[diag]1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -[/diag]
1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -

[Anonymous]

New LG has still hash key bugs like the previous version.
In the position below it will always play ...Qa7?? without thinking.

Guenther

[diag]1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -[/diag]
1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -

I've noticed this behavior also. See this post on CCC ...

http://www.talkchess.com/forums/1/message.html?426326

Regards
Dave

[Uri Blass]

New LG has still hash key bugs like the previous version.
In the position below it will always play ...Qa7?? without thinking.

Guenther

[diag]1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -[/diag]
1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -

2 questions:

1)How do you know that it is a hash bug?
I can only know that it is a bug.

I expect hash bug to cause the program to play stupid mistakes after long search and not to cause the program to play stupid mistakes with no thinking time.

2)What previous version you talk about?

Uri

[Guenther Simon]

I should have said more precisely that it is an index key bug
related to a bug in LGs book making process obviously.

I tested the new position against the book and it is really
stored as a book move.

Look in the message from this old link:
http://wbforum.volker-pittlik.name/viewtopic.php?t=236&highlight=goliath+bug

[Anonymous]

I should have said more precisely that it is an index key bug
related to a bug in LGs book making process obviously.

I tested the new position against the book and it is really
stored as a book move.

Look in the message from this old link:
http://wbforum.volker-pittlik.name/viewtopic.php?t=236&highlight=goliath+bug

In this game, Evolution instantly played the terrible move 40. Ra2 ??. This doesn't look like a book move to me.

[Event "Evolution gauntlet"]
[Site "?"]
[Date "2005.05.16"]
[Round "1"]
[White "Evolution"]
[Black "Crafty Cito"]
[Result "0-1"]

1. c4 e5 2. Nc3 Nc6 3. g3 Nf6 4. Nf3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be7 8. d3 O-O 9. a3 Be6 10. b4 a5 11. b5 Nd4 12. Nd2 c6 13. bxc6 Nxc6 14. Bb2 Qd7 15. Na4 Ra6 16. Nxb6 Rxb6 17. Bc3 Qc7 18. Qa4 Ra6 19. Rfb1 f6 20. Nc4 Rb8 21. Rb2 Bc5 22. e3 Be7 23. Be4 Kh8 24. Rc1 Na7 25. Qd1 b5 26. Qh5 Bg8 27. Rcb1 a4 28. Bb4 f5 29. Qxf5 Re6 30. Ba5 Qc5 31. Bb4 Qc7 32. Nd2 Rf8 33. Qg4 Bxb4 34. Rxb4 Ref6 35. Nf3 g6 36. Nh4 Be6 37. Qg5 Rxf2 38. R4b2 R2f6 39. g4 Kg8 40. Ra2 ??

Regards
Dave

[Anonymous]

In this game, Evolution instantly played the terrible move 40. Ra2 ??. This doesn't look like a book move to me.

[Event "Evolution gauntlet"]
[Site "?"]
[Date "2005.05.16"]
[Round "1"]
[White "Evolution"]
[Black "Crafty Cito"]
[Result "0-1"]

1. c4 e5 2. Nc3 Nc6 3. g3 Nf6 4. Nf3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be7 8. d3 O-O 9. a3 Be6 10. b4 a5 11. b5 Nd4 12. Nd2 c6 13. bxc6 Nxc6 14. Bb2 Qd7 15. Na4 Ra6 16. Nxb6 Rxb6 17. Bc3 Qc7 18. Qa4 Ra6 19. Rfb1 f6 20. Nc4 Rb8 21. Rb2 Bc5 22. e3 Be7 23. Be4 Kh8 24. Rc1 Na7 25. Qd1 b5 26. Qh5 Bg8 27. Rcb1 a4 28. Bb4 f5 29. Qxf5 Re6 30. Ba5 Qc5 31. Bb4 Qc7 32. Nd2 Rf8 33. Qg4 Bxb4 34. Rxb4 Ref6 35. Nf3 g6 36. Nh4 Be6 37. Qg5 Rxf2 38. R4b2 R2f6 39. g4 Kg8 40. Ra2 ??

Regards
Dave[/quote]

I was wrong! I temporarily renamed the Evolution book, and tried this position again. And Evolution doesn't play Ra2, instead it analyzes normally, choosing Rg2. Very strange, a book move on move 40.

Regards
Dave

[Guenther Simon]

[quote = Dave]

In this game, Evolution instantly played the terrible move 40. Ra2 ??. This doesn't look like a book move to me. Smile

[Event "Evolution gauntlet"]
[Site "?"]
[Date "2005.05.16"]
[Round "1"]
[White "Evolution"]
[Black "Crafty Cito"]
[Result "0-1"]

1. c4 e5 2. Nc3 Nc6 3. g3 Nf6 4. Nf3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be7 8. d3 O-O 9. a3 Be6 10. b4 a5 11. b5 Nd4 12. Nd2 c6 13. bxc6 Nxc6 14. Bb2 Qd7 15. Na4 Ra6 16. Nxb6 Rxb6 17. Bc3 Qc7 18. Qa4 Ra6 19. Rfb1 f6 20. Nc4 Rb8 21. Rb2 Bc5 22. e3 Be7 23. Be4 Kh8 24. Rc1 Na7 25. Qd1 b5 26. Qh5 Bg8 27. Rcb1 a4 28. Bb4 f5 29. Qxf5 Re6 30. Ba5 Qc5 31. Bb4 Qc7 32. Nd2 Rf8 33. Qg4 Bxb4 34. Rxb4 Ref6 35. Nf3 g6 36. Nh4 Be6 37. Qg5 Rxf2 38. R4b2 R2f6 39. g4 Kg8 40. Ra2 ??

It is stored in the book...
Below are the reactions of LGEvo a) with book and b) with
book disabled. Something goes horribly wrong in its book
making process and it seems to store rubbish from time to
time.

BTW I am using the large book with option *tournament*.

Regards,
Guenther

Code: Select all

9303 >first : setboard 5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - - 0 1
9363 <first : pong 2
11907 >first : name Guenther Simon
11907 >first : time 30000
otim 30000
11907 >first : go
13459 <first : move b2a2

Code: Select all

7971 >first : setboard 5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - - 0 1
8072 <first : pong 2
13289 >first : name Guenther Simon
13289 >first : time 30000
otim 30000
13289 >first : go
14751 <first : 6 -36 4 37640 d4 Rf1+ Rxf1 Rxf1+ Kg2 exd4 exd4
14811 <first : 6 -77 8 48334 d4 exd4 exd4 Qf4
14971 <first : 7 -110 26 123542 d4 exd4 exd4 Qf4
15152 <first : 7 -127 43 201952 d4 exd4 Nf5 dxe3 Nxe3
15212 <first : 7 -115 48 226822 Nf5 Qc3 Ne7+ Kh8 Nxg6+ hxg6 Bxg6
15352 <first : 7 -89 61 311328 Nf5 Qc3 Nd6 Rf1+ Rxf1 Rxf1+ Kxf1 Qc1+ Kf2 Qxb2+ Ke1 Qxh2 Qd8+ Kg7
15372 <first : 7 -80 68 314472 Rb4 Qc3 Bxg6 Qxa3 Be4+
15883 <first : 8 -27 106 607450 Rb4 Qc3 Bxg6 Qxa3 Be4+ Kh8 Qxe5 Bxg4
16844 <first : 9 -12 215 1260866 Rb4 Qc3 Bxg6 Qxa3 Bf5+ Kh8 Bxe6 Rxe6 Nf5
17655 <first : 10 -42 290 1653408 Rb4 Qc3 Bxg6 Qxa3 Bf5+ Kh8 Bxe6 Rxe6 Nf5
19037 <first : 10 -56 411 2682390 Rb4 Qg7 Ng2 h6 Qh4 Qf7 h3 Ba2
23804 <first : 11 -56 915 5760918 Rb4 Qg7 Ng2 h6 Qh4 Qf7 h3 Ba2
23804 <first : move b2b4

[Anonymous]

[quote = Dave]

In this game, Evolution instantly played the terrible move 40. Ra2 ??. This doesn't look like a book move to me. Smile

[Event "Evolution gauntlet"]
[Site "?"]
[Date "2005.05.16"]
[Round "1"]
[White "Evolution"]
[Black "Crafty Cito"]
[Result "0-1"]

1. c4 e5 2. Nc3 Nc6 3. g3 Nf6 4. Nf3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be7 8. d3 O-O 9. a3 Be6 10. b4 a5 11. b5 Nd4 12. Nd2 c6 13. bxc6 Nxc6 14. Bb2 Qd7 15. Na4 Ra6 16. Nxb6 Rxb6 17. Bc3 Qc7 18. Qa4 Ra6 19. Rfb1 f6 20. Nc4 Rb8 21. Rb2 Bc5 22. e3 Be7 23. Be4 Kh8 24. Rc1 Na7 25. Qd1 b5 26. Qh5 Bg8 27. Rcb1 a4 28. Bb4 f5 29. Qxf5 Re6 30. Ba5 Qc5 31. Bb4 Qc7 32. Nd2 Rf8 33. Qg4 Bxb4 34. Rxb4 Ref6 35. Nf3 g6 36. Nh4 Be6 37. Qg5 Rxf2 38. R4b2 R2f6 39. g4 Kg8 40. Ra2 ??

It is stored in the book...
Below are the reactions of LGEvo a) with book and b) with
book disabled. Something goes horribly wrong in its book
making process and it seems to store rubbish from time to
time.

BTW I am using the large book with option *tournament*.

Regards,
Guenther

Code: Select all

9303 >first : setboard 5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - - 0 1
9363 <first : pong 2
11907 >first : name Guenther Simon
11907 >first : time 30000
otim 30000
11907 >first : go
13459 <first : move b2a2

Code: Select all

7971 >first : setboard 5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - - 0 1
8072 <first : pong 2
13289 >first : name Guenther Simon
13289 >first : time 30000
otim 30000
13289 >first : go
14751 <first : 6 -36 4 37640 d4 Rf1+ Rxf1 Rxf1+ Kg2 exd4 exd4
14811 <first : 6 -77 8 48334 d4 exd4 exd4 Qf4
14971 <first : 7 -110 26 123542 d4 exd4 exd4 Qf4
15152 <first : 7 -127 43 201952 d4 exd4 Nf5 dxe3 Nxe3
15212 <first : 7 -115 48 226822 Nf5 Qc3 Ne7+ Kh8 Nxg6+ hxg6 Bxg6
15352 <first : 7 -89 61 311328 Nf5 Qc3 Nd6 Rf1+ Rxf1 Rxf1+ Kxf1 Qc1+ Kf2 Qxb2+ Ke1 Qxh2 Qd8+ Kg7
15372 <first : 7 -80 68 314472 Rb4 Qc3 Bxg6 Qxa3 Be4+
15883 <first : 8 -27 106 607450 Rb4 Qc3 Bxg6 Qxa3 Be4+ Kh8 Qxe5 Bxg4
16844 <first : 9 -12 215 1260866 Rb4 Qc3 Bxg6 Qxa3 Bf5+ Kh8 Bxe6 Rxe6 Nf5
17655 <first : 10 -42 290 1653408 Rb4 Qc3 Bxg6 Qxa3 Bf5+ Kh8 Bxe6 Rxe6 Nf5
19037 <first : 10 -56 411 2682390 Rb4 Qg7 Ng2 h6 Qh4 Qf7 h3 Ba2
23804 <first : 11 -56 915 5760918 Rb4 Qg7 Ng2 h6 Qh4 Qf7 h3 Ba2
23804 <first : move b2b4

I'm also using the Large book with option "tournament". If i let it think longer, it will choose the move Rg2, without the book.

Regards
Dave

[Anonymous]

I should have said more precisely that it is an index key bug
related to a bug in LGs book making process obviously.

I tested the new position against the book and it is really
stored as a book move.

Look in the message from this old link:
http://wbforum.volker-pittlik.name/viewtopic.php?t=236&highlight=goliath+bug

I switched to the smaller book, with option "tournament", and this bug doesn't appear, at least on these two positions ...

1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -
5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - -

Perhaps the bug will still appear with the smaller book on other positins?

Regards
Dave

[Uri Blass]

In this game, Evolution instantly played the terrible move 40. Ra2 ??. This doesn't look like a book move to me.

[Event "Evolution gauntlet"]
[Site "?"]
[Date "2005.05.16"]
[Round "1"]
[White "Evolution"]
[Black "Crafty Cito"]
[Result "0-1"]

1. c4 e5 2. Nc3 Nc6 3. g3 Nf6 4. Nf3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be7 8. d3 O-O 9. a3 Be6 10. b4 a5 11. b5 Nd4 12. Nd2 c6 13. bxc6 Nxc6 14. Bb2 Qd7 15. Na4 Ra6 16. Nxb6 Rxb6 17. Bc3 Qc7 18. Qa4 Ra6 19. Rfb1 f6 20. Nc4 Rb8 21. Rb2 Bc5 22. e3 Be7 23. Be4 Kh8 24. Rc1 Na7 25. Qd1 b5 26. Qh5 Bg8 27. Rcb1 a4 28. Bb4 f5 29. Qxf5 Re6 30. Ba5 Qc5 31. Bb4 Qc7 32. Nd2 Rf8 33. Qg4 Bxb4 34. Rxb4 Ref6 35. Nf3 g6 36. Nh4 Be6 37. Qg5 Rxf2 38. R4b2 R2f6 39. g4 Kg8 40. Ra2 ??

Regards
Dave

I was wrong! I temporarily renamed the Evolution book, and tried this position again. And Evolution doesn't play Ra2, instead it analyzes normally, choosing Rg2. Very strange, a book move on move 40.

Regards
Dave[/quote]

Nothing is strange here.

Programs store book positions based on the hash key.

Movei store all the 64 bits of the hash key so it has no problem but if a programmer is too smart and want to save space by storing only 32 bits then the program may play this stupid errors.

Movei can in theory also play a stupid error if there is hash collision with book position but the probability for hash collision is so small that it practically does not happen(it is something like 1/2^40 for a move).

Inspite of it I think that it may be better to save the position and not the hash key in the book and I may change later the format of the book.

The reason is that I think to give the user an option to fix the book or extend the book by analyzing the book position or positions one ply after the book position.

Today Movei cannot find the positions of the book based on the book file.
Finding them based on the pgn that is used to construct the book means building the book again with positions instead of hash keys because not every position in the pgn is in the book.

I do not like to need to build something simililar to the book again and I prefer to use the book so if I decide to have an option to change or extend the book by analyzing the book position then I think that it is better to have the book positions in the book inspite of the fact that it means bigger book.

It is not in the top of priority to do it so do not expect me to do it in the next few months.

Uri

[Guenther Simon]

I switched to the smaller book, with option "tournament", and this bug doesn't appear, at least on these two positions ...

1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -
5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - -

Perhaps the bug will still appear with the smaller book on other positins?

Regards
Dave

I want to add another position:
Here LG Evo plays ...e5?? from 'book'
[diag]1r3rk1/p2bp1bp/5pp1/1ppP1P2/q1P1P3/3Q1NP1/PR4BP/1R4K1 w [/diag]
1r3rk1/p2bp1bp/5pp1/1ppP1P2/q1P1P3/3Q1NP1/PR4BP/1R4K1 w

Regards,
Guenther

P.S. I would suppose the smaller book has also some
of those key 'holes', but at least all 3 positions above
are clean
(may be the probability is much lower? or it is indeed
without that bugs...I will change to it for my next games)

[Anonymous]

I switched to the smaller book, with option "tournament", and this bug doesn't appear, at least on these two positions ...

1r4k1/2q1bppp/3p1n2/1pp1pP2/8/1P1P1N1P/2PBQPP1/R5K1 b - -
5rk1/n1q4p/4brp1/1p2p1Q1/p3B1PN/P2PP3/1R5P/1R4K1 w - -

Perhaps the bug will still appear with the smaller book on other positins?

Regards
Dave

I want to add another position:
Here LG Evo plays ...e5?? from 'book'
[diag]1r3rk1/p2bp1bp/5pp1/1ppP1P2/q1P1P3/3Q1NP1/PR4BP/1R4K1 w [/diag]
1r3rk1/p2bp1bp/5pp1/1ppP1P2/q1P1P3/3Q1NP1/PR4BP/1R4K1 w

Regards,
Guenther

P.S. I would suppose the smaller book has also some
of those key 'holes', but at least all 3 positions above
are clean
(may be the probability is much lower? or it is indeed
without that bugs...I will change to it for my next games)

I checked the games lost by Evolution, and out of 55 total games, i found eight total occurances of this bug. In addition to the position i already posted, here are the others ...

rq3rk1/pb2bpp1/4p2p/3nP3/1p2N1Q1/3B4/PP3PPP/R1B2RK1 w - - 0 1
Evolution played Be3?? giving away a pawn.

4rrk1/1p3pbp/p5p1/8/3q2P1/4P3/P3QBPP/1RR3K1 b - - 0 1
Evolution played Ra8?? giving away it's queen.

5rk1/pp3rp1/2nbbq1p/3p1P2/8/P3B3/1PBQN1PP/4RRK1 b - - 0 1
Evolution played Nd8?? losing a bishop.

4rrk1/1p1n1ppp/1qp2n2/p7/PbBP1B2/1QN4P/1P3PP1/2RR2K1 w - - 0 1
Evolution played Qa3?? giving away it's queen.

r3rbk1/pppq1p1p/5np1/3pRN2/3P1B2/2PQ3P/PP3PP1/R5K1 w - - 0 1
Evolution played Qd2?? losing a knight.

r2k1b1r/1p2pppp/p4n2/1B3b2/1n1P4/2N1B3/PP1K1PPP/R5NR w - - 0 1
Evolution played Ke2?? losing the bishop.

r1bq3k/pp2p1bp/3p2p1/3Qnr2/2P2B2/2N4P/PP2BPP1/R4RK1 w - - 0 1
Evolution played Rfc1?? losing the bishop.

In all cases, Evolution was clearly out of book, and all of these moves were made instantly, without any thinking. Using the smaller book, Evolution doesn't play any of these moves.

I'll continue to watch and see if this bug occurs with the small book.

Regards
Dave

[Roger Brown]

I checked the games lost by Evolution, and out of 55 total games, i found eight total occurances of this bug. In addition to the position i already posted, here are the others ...

rq3rk1/pb2bpp1/4p2p/3nP3/1p2N1Q1/3B4/PP3PPP/R1B2RK1 w - - 0 1
Evolution played Be3?? giving away a pawn.

4rrk1/1p3pbp/p5p1/8/3q2P1/4P3/P3QBPP/1RR3K1 b - - 0 1
Evolution played Ra8?? giving away it's queen.

5rk1/pp3rp1/2nbbq1p/3p1P2/8/P3B3/1PBQN1PP/4RRK1 b - - 0 1
Evolution played Nd8?? losing a bishop.

4rrk1/1p1n1ppp/1qp2n2/p7/PbBP1B2/1QN4P/1P3PP1/2RR2K1 w - - 0 1
Evolution played Qa3?? giving away it's queen.

r3rbk1/pppq1p1p/5np1/3pRN2/3P1B2/2PQ3P/PP3PP1/R5K1 w - - 0 1
Evolution played Qd2?? losing a knight.

r2k1b1r/1p2pppp/p4n2/1B3b2/1n1P4/2N1B3/PP1K1PPP/R5NR w - - 0 1
Evolution played Ke2?? losing the bishop.

r1bq3k/pp2p1bp/3p2p1/3Qnr2/2P2B2/2N4P/PP2BPP1/R4RK1 w - - 0 1
Evolution played Rfc1?? losing the bishop.

In all cases, Evolution was clearly out of book, and all of these moves were made instantly, without any thinking. Using the smaller book, Evolution doesn't play any of these moves.

I'll continue to watch and see if this bug occurs with the small book.

Regards
Dave

Hello Dave and Guenther,

Thanks heaps for your analysis.

This kind of detailed research is appreciated particularly by those who are like me, ignorant.

Oh, I am alone in this category?



Thanks guys.

I guess bigger is not always better...



Later.