![[us]](/mwf/v12/flags/us.png "United States")
There seems to be a problem in the "Preserve Analysis" feature and I encounter it in the following situation. I am retroanalyzing a line in the Benoni, going backward from move 18. When I move backward, the score seems to level at around +0.35 for white, and this is the situation with black to move on the 11th move. The score gives +0.35, but when I move back a step for white to move, the move 11. f4! is not recommended, and instead 11. Na4?! is given with a score of +0.21. I have an Athlon 64 3200+ with 2 GB RAM and am using 1 GB hash. Here is the line:
1. d4 Nf6 2. c4 c5 3. d5 d6 4. Nc3 g6 5. e4 Bg7 6. Nf3 O-O 7. h3 Qa5 8. Bd2 Nfd7 9. Be2 Ne5 10. Nh2 Qb6 11. f4 Ned7 12. Qc2 Na6 13. a3 Nc7 14. Nf3 e6 15. O-O Re8 16. Bd3 exd5 17. exd5 Nf6 18. f5
And here is the analysis:
After moving forward again for black's 11th move, I get this:
Analysis by Rybka 2.3.2a 32-bit :
11...Ned7 12.Qc2 Na6 13.a3 Nc7 14.Nf3 e6 15.0-0 exd5 16.cxd5 a6 17.a4 Re8 18.Kh1
² (0.35) Depth: 19 00:00:00 0kN
11...Ned7 12.Qc2 Na6 13.a3 Nc7 14.Nf3 e6 15.0-0 exd5 16.cxd5 a6 17.a4 Re8 18.Kh1
² (0.35) Depth: 20 00:00:00 0kN
But after moving back for white's 11th move, I get this after waiting for two minutes:
Analysis by Rybka 2.3.2a 32-bit :
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 18 00:00:00 0kN
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 19 00:00:33 1904kN
Why doesn't Rybka immediately recommend 11.f4! with a score of +0.35 with the "Preserve Analysis" feature on?
EDIT: The score at least becomes correct after awhile, though only because Rybka apparently found a different move that works just as well:
Analysis by Rybka 2.3.2a 32-bit :
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 18 00:00:00 0kN
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 19 00:00:33 1904kN
11.Qc2 Na6 12.f4 Nb4 13.Qb1 Nd7 14.a3 Na6 15.Qc2 Nc7 16.Nf3 e6 17.0-0 exd5
² (0.35) Depth: 19 00:11:02 30013kN
11.Qc2 Na6 12.f4 Nb4 13.Qb1 Nd7 14.a3 Na6 15.Qc2 Nc7 16.Nf3 e6 17.0-0 exd5
² (0.35) Depth: 20 00:24:30 68571kN
1. d4 Nf6 2. c4 c5 3. d5 d6 4. Nc3 g6 5. e4 Bg7 6. Nf3 O-O 7. h3 Qa5 8. Bd2 Nfd7 9. Be2 Ne5 10. Nh2 Qb6 11. f4 Ned7 12. Qc2 Na6 13. a3 Nc7 14. Nf3 e6 15. O-O Re8 16. Bd3 exd5 17. exd5 Nf6 18. f5
And here is the analysis:
After moving forward again for black's 11th move, I get this:
rnb2rk1/pp2ppbp/1q1p2p1/2pPn3/2P1PP2/2N4P/PP1BB1PN/R2QK2R b KQ f3 0 1
Analysis by Rybka 2.3.2a 32-bit :
11...Ned7 12.Qc2 Na6 13.a3 Nc7 14.Nf3 e6 15.0-0 exd5 16.cxd5 a6 17.a4 Re8 18.Kh1
² (0.35) Depth: 19 00:00:00 0kN
11...Ned7 12.Qc2 Na6 13.a3 Nc7 14.Nf3 e6 15.0-0 exd5 16.cxd5 a6 17.a4 Re8 18.Kh1
² (0.35) Depth: 20 00:00:00 0kN
But after moving back for white's 11th move, I get this after waiting for two minutes:
rnb2rk1/pp2ppbp/1q1p2p1/2pPn3/2P1P3/2N4P/PP1BBPPN/R2QK2R w KQ - 0 1
Analysis by Rybka 2.3.2a 32-bit :
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 18 00:00:00 0kN
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 19 00:00:33 1904kN
Why doesn't Rybka immediately recommend 11.f4! with a score of +0.35 with the "Preserve Analysis" feature on?
EDIT: The score at least becomes correct after awhile, though only because Rybka apparently found a different move that works just as well:
Analysis by Rybka 2.3.2a 32-bit :
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 18 00:00:00 0kN
11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2 Qxb3
= (0.23) Depth: 19 00:00:33 1904kN
11.Qc2 Na6 12.f4 Nb4 13.Qb1 Nd7 14.a3 Na6 15.Qc2 Nc7 16.Nf3 e6 17.0-0 exd5
² (0.35) Depth: 19 00:11:02 30013kN
11.Qc2 Na6 12.f4 Nb4 13.Qb1 Nd7 14.a3 Na6 15.Qc2 Nc7 16.Nf3 e6 17.0-0 exd5
² (0.35) Depth: 20 00:24:30 68571kN
Now this is simply infuriating: I go back one more move to look at black's ninth move, and I get the following, with the recommended line being white playing 10.Na4 and a score of +0.21. What I am posting here is after three minutes, where Rybka has just gotten through analyzing all of the moves at depth 19 and is moving to depth 20. This "Preserve Analysis" feature apparently isn't very reliable...
Analysis by Rybka 2.3.2a 32-bit :
10...Qb6 11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2
= (0.21) Depth: 18 00:00:00 0kN
10...Qb6 11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2
= (0.21) Depth: 19 00:00:00 0kN
rnb2rk1/pp2ppbp/3p2p1/q1pPn3/2P1P3/2N4P/PP1BBPPN/R2QK2R b KQ - 0 1
Analysis by Rybka 2.3.2a 32-bit :
10...Qb6 11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2
= (0.21) Depth: 18 00:00:00 0kN
10...Qb6 11.Na4 Qa6 12.Qb3 f5 13.f4 Nf7 14.exf5 Bxf5 15.Ng4 Bd7 16.Nc3 Qb6 17.Kf2
= (0.21) Depth: 19 00:00:00 0kN
Hi,
I was having similar problems with the other position.
At the last analyzed move I got 0.21 when I went back 2 moves I got 0.11 but after some time - about 60' - I got 0.21 back.
It could be the interface problem.
Regards
Hetman
I was having similar problems with the other position.
At the last analyzed move I got 0.21 when I went back 2 moves I got 0.11 but after some time - about 60' - I got 0.21 back.
It could be the interface problem.
Regards
Hetman
![[hu]](/mwf/v12/flags/hu.png "Hungary")
Thanks. This looks like a bug in the hash table. I'm not sure if it's restricted to preserve analysis.
I've saved the position and will look at it when there is more time (ie. after Mexico).
Vas
I've saved the position and will look at it when there is more time (ie. after Mexico).
Vas
![[cr]](/mwf/v12/flags/cr.png "Costa Rica")
Yes, I remember that some time ago, Genorb also exposed a similar problem here. The conclusion in that thread was that it could have been a hash bug.
Regards,
Gambito.
Regards,
Gambito.
![[ee]](/mwf/v12/flags/ee.png "Estonia")
I have similar problem too, and without "preserve analysis" option.
In my recent case there was position where in infinite mode I forced a move that was clearly better according to my chess knowledge, but Rybka did not see it from the starting point where Rybkas evaluations where around 6. As soon as the move was forced, Rybkas evaluation grew above 13 almost instantly (depth 17). After take back Rybka dropped the move and evaluation was again around 6 even if the analysis depth was 4 ply deeper (21).
In my recent case there was position where in infinite mode I forced a move that was clearly better according to my chess knowledge, but Rybka did not see it from the starting point where Rybkas evaluations where around 6. As soon as the move was forced, Rybkas evaluation grew above 13 almost instantly (depth 17). After take back Rybka dropped the move and evaluation was again around 6 even if the analysis depth was 4 ply deeper (21).
Ok, thanks for the comments.
Note that what you describe here isn't necessarily a bug, it's something which can happen as a part of search.
Vas
Note that what you describe here isn't necessarily a bug, it's something which can happen as a part of search.
Vas
Can you, please, explain, how this can happen as a part of search?
I see quite often, that when analysing certain line using backwards method, Rybka drops the best move that was found during long analysis. And if to follow the new choosen line, the evaluation remains much worse than in case the initial line. It seems, that this case Rybka does not trust hash values and does very short search and prunes the best line off.
I see quite often, that when analysing certain line using backwards method, Rybka drops the best move that was found during long analysis. And if to follow the new choosen line, the evaluation remains much worse than in case the initial line. It seems, that this case Rybka does not trust hash values and does very short search and prunes the best line off.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill