Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka3 only 30 ELO better than Fritz 11
- - By Gr00vy (**) Date 2008-08-11 08:47
This was an interesting couple of days... Resulting in Fritz 11 beating Rybka in the last five games.  I don't know what was going on.  Under the Chessbass UI using the Fritz 11 book for Fritz 11, and they Rybka II book with LEARNING ON and 5 minute blitz...  These are the results after 300 games...

Test Match, Blitz:5'  0
1   Rybka 3 1-cpu 32-bit  +126/-100/=74 54.33%  163.0/300
2   Fritz 11              +100/-126/=74 45.67%  137.0/300

Two very interesting things.  First is that performance started about 20% for Fritz and worked its way back up to 45.67 percent.  This is highly unexpected, and not shown in ANY of the results that have been shown here.  As a matter of fact, in the near religious fervor of avoiding "learning" nobody apparently tests it.   However, it appears to be worth at least 100 ELO.  That is phenomenal.   I wouldn't think that the books would be worth that much.    In the end Fritz 11 was winning both the white and black sides of the Sicilian.  This also means that for whatever reason Rybka is NOT learning, or not learning as well.   I have no reason why that would be true.

I am not sure that this is a bug, or this is expected behavior.   I would like to see more info.

This is running as single processor using the 32 bit version of each program.  I would have expected a MUCH higher performance by Rybka and I am a bit surprised by the results that I got.

Sample games at the end.  Now it doesn't seem that these are book wins.  It seems like that Fritz played the endgames better.  Unfortunately, I am probably not qualified enough to make that judgement, other than that Fritz was Kicking Rybka's butt in each of these games.  I am going for 400 games next.

[Event "Test Match, Blitz:5'"]
[Site "JOHNNYPC"]
[Date "2008.08.10"]
[Round "40"]
[White "Rybka 3 1-cpu 32-bit"]
[Black "Fritz 11"]
[Result "0-1"]
[ECO "B90"]
[Annotator "0.58;0.40"]
[PlyCount "122"]
[TimeControl "300"]

{Intel(R) Core(TM)2 Duo CPU     T8300  @ 2.40GHz 2373 MHz  W=10.8 ply; 48kN/s;
Rybka2.ctg  B=15.5 ply; 1,634kN/s; Fritz11.ctg} 1. e4 {B/0 0} c5 {B/0 0} 2.
Nf3 {B/0 0} d6 {B/0 0} 3. d4 {B/0 0} cxd4 {B/0 0} 4. Nxd4 {B/0 0} Nf6 {B/0 0}
5. Nc3 {B/0 0} a6 {B/0 0} 6. Be3 {B/0 0} e5 {B/0 0} 7. Nb3 {B/0 0} Be6 {B/0 0}
8. f3 {B/0 0} Nbd7 {B/0 0} 9. g4 {B/0 0} Nb6 {B/0 0} 10. g5 {B/0 0} Nh5 {B/0 0}
11. Qd2 {B/0 0} Be7 {B/0 0} 12. O-O-O {B/0 0} O-O {B/0 0} 13. Rg1 {B/0 0} Rc8 {
B/0 0} 14. Kb1 {B/0 0} g6 {B/0 0} 15. Qf2 {B/0 0} Nc4 {B/0 0} 16. Bxc4 {B/0 0}
Bxc4 {B/0 0} 17. Na4 {B/0 0} Be6 {B/0 0} 18. Nb6 {B/0 0} Rc7 {B/0 0} 19. Nd5 {
B/0 0} Bxd5 {B/0 0} 20. Rxd5 {B/0 0} Qe8 {Both last book move B/0 0} 21. Qd2 {
0.58/10 6} f5 {(Qa4) 0.40/15 8} 22. gxf6 {0.53/10 4} Rxf6 {0.34/16 14} 23. Qd1
{(Qf2) 0.45/11 9} Qf8 {(Nf4) 0.28/15 9} 24. Bg5 {0.47/11 5} Re6 {0.31/17 11}
25. h4 {(Bxe7) 0.47/13 10} Bxg5 {0.19/16 7} 26. hxg5 {0.46/14 4} Nf4 {0.14/17 6
} 27. Rd2 {0.50/14 21} Qe7 {(Qd8) 0.15/16 8} 28. Rh2 {(Qf1) 0.51/11 4} Qf8 {
(b5) 0.28/16 21} 29. Qd2 {(c3) 0.51/11 6} Re8 {0.26/14 5} 30. Qe3 {
(a3) 0.46/10 3} d5 {(Ne6) 0.02/14 6} 31. Qb6 {(exd5) 0.27/11 9} Rec8 {
-0.13/14 5} 32. c3 {(Rgh1) 0.21/10 5} dxe4 {-0.12/15 6} 33. fxe4 {0.13/10 2}
Qd8 {(Rd7) -0.12/15 5} 34. Nc1 {(Rd2) 0.11/11 8} Rf7 {(Qe7) -0.29/15 4} 35. Qb3
{0.00/10 3} Qe7 {-0.39/15 5} 36. Rd2 {(Rf2) 0.00/12 5} Kg7 {(Rd8) -0.54/15 4}
37. Rgd1 {0.05/11 3} Rc7 {(Qe6) -0.55/16 7} 38. Qb6 {-0.07/10 5} Qxg5 {
-0.49/10 4} 39. Nb3 {-0.24/10 4} Rc6 {-0.88/15 4} 40. Qe3 {-0.30/10 4} Qe7 {
(b6) -1.08/16 8} 41. Qa7 {(Rd8) -0.27/11 6} Qc7 {(Rc4) -1.04/16 18} 42. Rd8 {
(Qe3) -0.22/10 3} Rb6 {(Rc4) -1.22/16 10} 43. Qa8 {-0.37/10 3} Qc4 {-1.24/16 9}
44. Re1 {-0.48/10 4} Ne6 {-1.36/15 5} 45. Rdd1 {-0.66/11 7} h5 {
(Qc7) -1.59/15 5} 46. Qa7 {-0.92/10 7} Qc7 {-1.70/14 3} 47. Rg1 {
(Rd3) -1.05/10 4} Nc5 {(Rd6) -2.15/14 4} 48. Rg3 {(Rg2) -1.47/9 5} Nxe4 {
-2.88/14 3} 49. Rg2 {(Re3) -1.92/10 7} h4 {(Nf2) -3.11/14 8} 50. Re2 {
(Rh2) -1.91/9 3} Nd6 {(Qc4) -3.72/14 3} 51. Red2 {-2.60/9 2} h3 {-3.92/15 4}
52. Qa8 {(Rd3) -2.91/9 3} e4 {(Rf8) -4.20/14 2} 53. Ka1 {(Re2) -2.76/8 3} a5 {
(e3) -4.83/13 2} 54. Qxa5 {-3.85/8 2} Nc4 {-5.34/15 2} 55. Qg5 {-4.72/10 5}
Nxd2 {-6.12/14 1} 56. Qxd2 {-5.11/11 9} h2 {-6.25/14 1} 57. Qe2 {-4.72/10 15}
Rbf6 {(Qe5) -8.11/14 3} 58. Nc1 {-5.93/9 13} Rf2 {(Qe5) -9.96/14 0} 59. Qe1 {
-11.58/11 11} Qc4 {(Rf1) -15.21/14 1} 60. Qh1 {-11.54/9 1} Qb5 {-17.29/14 0}
61. Nb3 {-24.84/11 9} Rf1 {-22.82/16 0} 0-1

[Event "Test Match, Blitz:5'"]
[Site "JOHNNYPC"]
[Date "2008.08.10"]
[Round "43"]
[White "Fritz 11"]
[Black "Rybka 3 1-cpu 32-bit"]
[Result "1-0"]
[ECO "B90"]
[Annotator "0.76;0.07"]
[PlyCount "121"]
[TimeControl "300"]

{Intel(R) Core(TM)2 Duo CPU     T8300  @ 2.40GHz 2373 MHz  W=19.9 ply; 2,138kN/
s; Fritz11.ctg  B=14.3 ply; 46kN/s; Rybka2.ctg} 1. e4 {B/0 0} c5 {B/0 0} 2.
Nf3 {B/0 0} d6 {B/0 0} 3. d4 {B/0 0} cxd4 {B/0 0} 4. Nxd4 {B/0 0} Nf6 {B/0 0}
5. Nc3 {B/0 0} a6 {B/0 0} 6. Be3 {B/0 0} e5 {0.07/11 5} 7. Nb3 {B/0 0} Be6 {
B/0 0} 8. f3 {B/0 0} Be7 {B/0 0} 9. Qd2 {B/0 0} O-O {B/0 0} 10. O-O-O {B/0 0}
Nbd7 {Black last book move B/0 0} 11. g4 {B/0 0} a5 {0.04/10 5} 12. a4 {
White last book move B/0 0} Bxb3 {0.16/12 9} 13. cxb3 {0.76/14 2} Nc5 {
(Rc8) 0.27/12 7} 14. Bc4 {0.70/16 8} Nfd7 {0.44/12 23} 15. Nd5 {(Kb1) 0.88/16 8
} Rc8 {0.39/11 7} 16. Kb1 {1.08/18 10} Nb6 {1.06/11 25} 17. Nxe7+ {1.43/19 9}
Qxe7 {1.06/10 0} 18. Qxd6 {1.13/20 17} Qxd6 {1.14/12 3} 19. Rxd6 {1.13/19 1}
Nxc4 {1.15/13 3} 20. bxc4 {0.96/19 5} Nxa4 {0.87/13 4} 21. Rc1 {1.12/18 10} Nc5
{0.84/13 5} 22. Rb6 {1.04/17 6} Nd3 {(Rfd8) 0.84/14 5} 23. Rd1 {
(Rc3) 1.07/18 11} Nb4 {0.80/12 4} 24. Rxb7 {1.06/18 14} Rxc4 {0.80/10 0} 25.
Re7 {0.92/18 8} Nc6 {(Nc2) 0.83/12 3} 26. Rc7 {1.10/18 4} a4 {0.82/14 11} 27.
Rd5 {(Bb6) 1.01/18 8} Re8 {(Ra8) 0.72/13 4} 28. g5 {(Bd2) 1.02/16 4} g6 {
0.71/12 2} 29. Bb6 {0.96/18 6} Rb8 {0.69/14 4} 30. Ba7 {0.88/18 4} Ra8 {
(Re8) 0.54/14 3} 31. Rdd7 {0.88/18 9} Rf8 {0.51/14 12} 32. h4 {0.73/18 8} Na5 {
0.70/15 6} 33. Bc5 {0.76/19 6} Rb8 {(Ra8) 0.86/15 13} 34. Bd6 {1.15/18 4} Re8 {
0.90/15 5} 35. Re7 {(b4) 1.00/19 4} Rxe7 {0.81/14 3} 36. Rxe7 {1.02/19 1} Rc8 {
0.72/15 2} 37. Bxe5 {1.11/20 4} Nc6 {0.91/16 12} 38. Rc7 {0.76/22 4} Rxc7 {
0.91/15 0} 39. Bxc7 {0.77/22 3} Nd4 {0.91/16 1} 40. f4 {0.81/22 5} h5 {
0.96/16 4} 41. gxh6 {(Kc1) 0.92/21 14} Kh7 {0.59/15 1} 42. Be5 {0.76/20 4} Ne6
{(Nf3) 0.94/15 4} 43. Ka2 {1.14/20 6} Kxh6 {1.07/16 3} 44. Ka3 {1.36/21 3} Nc5
{1.68/17 15} 45. Bd4 {1.42/21 2} Ne6 {(Nxe4) 1.88/16 9} 46. Bf2 {1.92/21 2} Kh5
{2.08/16 10} 47. Kxa4 {1.82/21 2} Nxf4 {(Kg4) 2.08/14 2} 48. b4 {2.01/21 2} Nd3
{2.08/16 2} 49. Bd4 {2.32/20 2} Nxb4 {2.08/16 2} 50. Kxb4 {2.15/15 1} Kxh4 {
2.08/17 2} 51. Kc5 {2.35/17 2} Kg4 {(Kg3) 2.08/17 1} 52. Kd5 {2.35/21 2} Kf4 {
2.08/17 2} 53. Bc3 {(Bb2) 2.36/22 4} Ke3 {2.08/18 2} 54. Bb2 {(Bf6) 3.72/23 5}
Kf3 {(Kf4) 2.27/17 1} 55. Bc1 {5.27/20 1} Kg4 {2.27/16 8} 56. Kd6 {5.52/20 1}
g5 {5.12/12 7} 57. Ke5 {7.50/20 1} Kf3 {(Kh4) 5.09/8 1} 58. Bxg5 {11.15/18 0}
Kg3 {(Kg4) 10.48/10 5} 59. Kd6 {(Kf6) 11.97/17 1} Kf3 {(f5) 13.46/10 5} 60. e5
{15.48/18 0} Kg4 {13.46/9 1} 61. Bf6 {15.79/18 0} 1-0
Parent - - By Guenther RWBC (**) Date 2008-08-11 09:34
Gr00vy wrote:

This was an interesting couple of days... Resulting in Fritz 11 beating Rybka in the last five games.  I don't know what was going on.  Under the Chessbass UI using the Fritz 11 book for Fritz 11, and they Rybka II book with LEARNING ON and 5 minute blitz...  These are the results after 300 games...

Test Match, Blitz:5'  0
1   Rybka 3 1-cpu 32-bit  +126/-100/=74 54.33%  163.0/300
2   Fritz 11              +100/-126/=74 45.67%  137.0/300

Two very interesting things.  First is that performance started about 20% for Fritz and worked its way back up to 45.67 percent.  This is highly unexpected, and not shown in ANY of the results that have been shown here.  As a matter of fact, in the near religious fervor of avoiding "learning" nobody apparently tests it.   However, it appears to be worth at least 100 ELO.  That is phenomenal.   I wouldn't think that the books would be worth that much.    In the end Fritz 11 was winning both the white and black sides of the Sicilian.  This also means that for whatever reason Rybka is NOT learning, or not learning as well.   I have no reason why that would be true.

I am not sure that this is a bug, or this is expected behavior.   I would like to see more info.
<snip...>


Book learning is not done by the engines, but by the Chessbase GUI. Actually Rybka (and Fritz) don't learn at all, except you could
name the PH feature of Rybka a kind of learning, but currently it is still buggy and shouldn't be used in game playing.
(The games would show several moves marked with depth 5 after 0 seconds)
Unexperencied users can screw up a lot in setting up games especially in GUIs with too much features.

My first guess would be that you despite having choosen all the right book settings(if you did at all?)simply used
a write protected version of Rybkas ctg book, that way it was impossible for the GUI to write learning entries.
(Of course there is also a small chance for a bug in the CB GUI used, but I won't bet on it at least in this case.)
You can easily check the date/time of change for both used ctg books for verification.

Guenther
Parent - - By Permanent Brain (*****) Date 2008-08-11 09:58 Edited 2008-08-11 10:07
I agree that this result probably is an unclear mix of book learning (on one or both sides) and Rybka's persistant hash if it worked here, and how. I am not sure and I also do not understand the bug reports about the persistant hash feature. I think as long as this isn't fixed, Rybka 3 should be tested without it.

Also, as for book learning in a Fritz GUI, I think if the book setting is "optimized" (= narrow book, a few variations repeated again and again), then book learning can be very effective. If the goal is to provide a fair comparison of engines alone(!), either the same write-protected book should be used for all, or predefined opening sets with switching sides for each variation.

Actually I do not know what the book learning feature is for, at all. It disturbs fair engine vs. engine matches and it is absolutely uninteresting for the practical chess player. Book learning is only useful for internet chess server Elo hunting.

I think that ChessBase made a big mistake by introducing the "optimal" setting, which is much too narrow and focusses on a few presumably "best"(?!) variations of a book, and does not pay respect to the complete bandwidth of opening theory. The results are 80% Sicilian Najdorf in Playchess' engine room...
Parent - - By George Tsavdaris (****) Date 2008-08-11 10:22

>Actually I do not know what the book learning feature is for, at all. It disturbs fair engine vs. engine matches and it is absolutely uninteresting for the practical chess player. Book learning is only useful for internet chess server Elo hunting.


I disagree, since if an engine has to be called superior(e.g. Rybka 3) it should surpass every opponent. Even Fritz 11 + its learning.
-Fritz 11 is an opponent. Rybka would crush it if they played a match.
-But Fritz 11 + learning is another opponent and it seems Rybka 3 32-bit had problems with that opponent.

You are right that this wasn't a match Rybka 3 vs Fritz 11 and the title is misleading since it says Rybka 3 32-bit only 30 elo above Fritz 11 while it should say Fritz 11+learning.

But despite that it is interesting to see if RomiChess for example with its amazing learning method can win against Rybka 3 32-bit on a series of 100 matches.
Parent - - By Guenther RWBC (**) Date 2008-08-11 10:40
George, obviously me and Mike read the original post different. IMO he said _both_
programs used book learning, otherwise all that posts would be quite useless so far ;-)

Guenther
Parent - By George Tsavdaris (****) Date 2008-08-11 10:57
Yes indeed, but what i've said about the books is the most strange thing now for me:

I find it almost impossible that Rybka II.ctg that was used, did not contain 11...b5 after:
1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 e5 7. Nb3 Be6 8. f3 Be7 9. Qd2 O-O 10. O-O-O Nbd7 11. g4
.....that has been played in the second game.

Here Rybka played the awful 11...a5? but a just decent book should contain 11...b5 as this is a well known very deep variation. In fact it should cover the moves after 11...b5 up to move 30 or 35.

And RybkaII.ctg was not just a decent book but a good one so i wonder why this book left Rybka in dark.....
Parent - - By Martin Hander (*) Date 2008-08-11 09:46
Game 1:
My Rybka 3 1-cpu 32-bit  thinks on 32.c3 (which throws away all the advantage) only for 0.1s,
then it plays 32.exd5, which results nearly in a winning position.

Game 2:
I would call this a book win, because the black position is very bad after the book ends.

I think, more games have to be examined to find the reason.
Parent - - By pokerpawn (***) Date 2008-08-11 09:52
yes i would test using the fritz11 book for both , now you are not testing the engines but the book...
Parent - - By Guenther RWBC (**) Date 2008-08-11 09:56
No. The decisive point in the original post is about _learning_. It is highly unlikely to get >45% in a match
if you were down to 20% after dozens of games, if learning worked for both ctg books in the GUI.

Guenther
Parent - By Nelson Hernandez (Gold) Date 2008-08-11 10:08
The decisive point is the one unspoken.  Look at the comparative depths the engines are getting immediately after exiting book.
Parent - By George Tsavdaris (****) Date 2008-08-11 09:54
Can you email me all the 300 games of this tournament?
My email is:
gts77avdar@gmail.com   (remove 77)

Thanks......
Parent - - By Nelson Hernandez (Gold) Date 2008-08-11 10:07
Rybka is consistently getting far lower depths than Fritz in both games you show.  The way it is acting it looks like you have permanent brain 'on' for Fritz.

Additionally in the second game white definitely has a book win.  Black has scored 18.6% in my database after 12.a4.  RybkaII.ctg is not particularly strong in B90 variations; Fritz 11 can capitalize in the Najdorf simply because it came out later.
Parent - - By Guenther RWBC (**) Date 2008-08-11 10:12
Rybka3 simply shows lower depths than F11. That's normal.
Why don't you wait for verification that the Rybka ctg book did not learn,
before wildly speculating?

Guenther
Parent - - By Nelson Hernandez (Gold) Date 2008-08-11 10:23
I am not "wildly" speculating.  I am commenting on what I am observing and correlating the behavior to what I have seen in the past.  You find it "normal" that R3 would exit book and think five and six fewer plies than Fritz for several moves?
Parent - By George Tsavdaris (****) Date 2008-08-11 10:32 Edited 2008-08-11 10:35
Depths are almost* fine. I checked it now and it seems his hardware is like the one i used....

*almost means in some(6-7%) positions i get 1-2 ply faster than it reports.
Parent - - By Guenther RWBC (**) Date 2008-08-11 10:35
I don't have Rybka 3 yet and neither F11. I know though that Rybka 3 even reports lower depths
now than Rybka 2.32 and that F11 shows very high depths. If you have both programs why don't
you simply compute the expected depths from your machine?
Well, we don't even know, if the original poster played with ponder on/off.
So far we can only tell that the score had risen for F11 from 20% to 45% over 300 games, which is very unlikely.
From the current informations I can only deduce a book learning error, but if you know the expected depth to time
of Rybka 3 on his machine tell it :)
(I would wonder though why the change in score should have appeared later, because setting up
ponder would be done before the match started)

Guenther
Parent - By Nelson Hernandez (Gold) Date 2008-08-11 10:41
I agree with this much: an authoritative answer won't come without more information and a forensic analysis of the games.
Parent - By George Tsavdaris (****) Date 2008-08-11 10:47

>So far we can only tell that the score had risen for F11 from 20% to 45% over 300 games, which is very unlikely.


If the 20% is after a good sample of games(100 out of the 300 games played) then it is indeed very very unlikely that the score without the influence of learning became 45% after 300 games.
20% going to 45% can only happen(in the majority of times) due to learning.
Parent - - By George Tsavdaris (****) Date 2008-08-11 10:43
Black's (Rybka's) book in game 2 stopped at move 10.

11...a5? that it has been played it utter crap!
I find it very strange that Rybka II.ctg book does not have 11...b5 for a long well-known theoretical battle.

Does anyone that has a Rybka II.ctg can confirm that it does not contain 11...b5 after:
1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 e5 7. Nb3 Be6 8. f3 Be7 9. Qd2 O-O 10. O-O-O Nbd7 11. g4
Parent - - By Guenther RWBC (**) Date 2008-08-11 10:57 Edited 2008-08-11 11:06
From the original unmodified RybkaII.ctg:

                                       Prob(to play)

11...b5(green)                          100
------------------------------
11...Qc7(below all red)                   0
11...Nb6                                  0
11...Rc8                                  0
11...a5                                   0


*This means book settings were set wrong at least.
We still don't know if the book was set to learn...

Guenther

Edit: (especially for George)
*It is likely that the book settings were wrong, but we still cannot exclude the possibility
that the book settings were ok _and_ book learning was set too and the book learned
11...a5 from previous not sucessful games.
Well it is a difficult and complex world ;-)

(Actually from my long time experience in >90% of the cases user reported unusual/surprising
games/results the setup was screwed up.)
Parent - - By George Tsavdaris (****) Date 2008-08-11 10:59 Edited 2008-08-11 11:12
Thanks, was too lazy to download it just to see this.

So indeed book settings were badly set.

[edit]

>*It is likely that the book settings were wrong, but we still cannot exclude the possibility
>that the book settings were ok _and_ book learning was set too and the book learned
>11...a5 from previous not sucessful games.


I'm not sure what you are suggesting here.
11...a5 has been played by Rybka 3 and not from the book.

Also even if 11...a5 was played from the book(which is not the case as i've said) what you were suggesting? That book learned from the games so far that 11...a5 is good so it played it? There is a problem with this: 11...a5? is very bad so there is no chance that the score for black with 11...a5 would be good.

>Well it is a difficult and complex world


Not really, if the original poster provides the games.... :-)
Parent - - By pokerpawn (***) Date 2008-08-11 11:20
anyone have a link where i can download RybkaII.ctg i've looked on the site once but only found one in another format (arena)
Parent - By dragon49 (****) Date 2008-08-11 12:07

>anyone have a link where i can download RybkaII.ctg i've looked on the site once but only found one in another format (arena)>


There is a link here:

http://www.rybkachess.com/index.php?auswahl=Download+demo>
Parent - - By Guenther RWBC (**) Date 2008-08-11 11:30 Edited 2008-08-11 11:32
[edit]

    >*It is likely that the book settings were wrong, but we still cannot exclude the possibility
    >that the book settings were ok _and_ book learning was set too and the book learned
    >11...a5 from previous not sucessful games.

I'm not sure what you are suggesting here.
11...a5 has been played by Rybka 3 and not from the book.

Also even if 11...a5 was played from the book(which is not the case as i've said) what you were suggesting? That book learned from the games so far that 11...a5 is good so it played it? There is a problem with this: 11...a5? is very bad so there is no chance that the score for black with 11...a5 would be good.


------

I should have told it different. If b5 would have failed already, Rybka would have tried other alternatives and I guess
because all other moves have zero probability it would have started thinking for that move instead of replying from
book. That's at least how I think it would work in the CB GUI. But as you said that's very unlikely.

BTW how is the most efficient way to quote here?

Guenther
Parent - - By Vempele (Silver) Date 2008-08-11 11:35

> BTW how is the most efficient way to quote here?


Most people use the quote button:
Parent - - By Guenther RWBC (**) Date 2008-08-11 11:42 Edited 2008-08-11 11:48
Lol never noticed it. You must admit that it is well hidden and not very obvious, especially
for people with quite weak eyesight. I wouldn't even have guessed that the grey things in
it were 'quotatation marks' until you put my nose on it. Thanks :)

Guenther

Edit:

BTW the forum help file doesn't know yet that button:

Quoting

mwForum uses email-style quoting. To quote someone, simply copy&paste a line of text from the original post and prefix it with a > sign. It will then get highlighted in a different color. Please don't quote more text than necessary to establish context. Some forums may also have automatic quoting enabled, in that case please also trim down the quoted text to the minimum required for context.
Parent - By Vempele (Silver) Date 2008-08-11 12:54
There's a topic about it on the FAQ board.
Parent - By Gr00vy (**) Date 2008-08-11 16:07
Ponder/permanent-brain is off.
Parent - - By M ANSARI (*****) Date 2008-08-11 11:59
You don't get an ELO rating for an engine by using book learning ... the idea is to see how good the engine plays in various different positions ... that is why to check ELO all the independent testing groups use neutral books ... and some do repeated opposite positions where both sides get the same position once as black and once as white.  I am sure if we used an optimized Rybka book from one of the Playchess server's strong players or R3.ctg the results would also be tilted artificially for R3.  As for book learning ... now that is something I never really used ... generally if I want to improve my book I will use the "learn from database" and use a database of all my games.
Parent - By X3 (**) Date 2008-08-11 12:42
Well, SSDF use native (or recommended book) books. We play on 2 comps using auto232, so learning works for both engines.

X3
Parent - - By Gr00vy (**) Date 2008-08-11 16:03
I am rerunning the test.   I have reinstalled the RybkaII.CTG.  The current test is at 58 games with 28% result.  This is similar as to the first test.  Fritz did not get better results until about 100 games or so in... 

As to the question of Learning.  Why wouldn't it be a valid test?  I am wanting to see the results of best v. best, at least on my machine.   I am surprised by the quality of the result. The most interesting thing, is how learning affects mostly the weaker engine.   Though I would not have suspected such dramatic results.  Though I would guess that what we are seeing is the result of discovering bad book lines.  Having a controlled book, doesn't really tell me about the best quality engine I can have on my machine, nor tell me whether I should be purchasing the Rybka3 book.   But a good book is part and parcel of a good engine.  How we know if it is working is to test.

And for now, the test is with SUPPLIED books.  I agree that a test with the *same* books is interesting as well.  But supplied books should show each engine off at their best.
Parent - By Nelson Hernandez (Gold) Date 2008-08-11 16:25
Just for kicks, you might considering flipping books after your current test is done.  Might reveal something new.
Parent - - By Guenther RWBC (**) Date 2008-08-11 17:23

> I am rerunning the test.   I have reinstalled the RybkaII.CTG.  The current test is at 58 games with 28% result.  This is similar as to the first test.  Fritz did not get better results until about 100 games or so in... 
>
> As to the question of Learning.  Why wouldn't it be a valid test?  I am wanting to see the results of best v. best, at least on my machine.   I am surprised by the quality of the result. The most interesting thing, is how learning affects mostly the weaker engine.   Though I would not have suspected such dramatic results.  Though I would guess that what we are seeing is the result of discovering bad book lines.


Haven't you read the thread at all? We still don't know if you had learning on at all for Rybka.
We only know that you at least had screwed up the book settings for Rybkas book...
Have you ensured that Rybkas book isn't write protected? Did you check the book files
after your first match, as I have suggested?
From your post it seems you don't understand what we were talking about. Repeating
the same errors again instead of checking what happened is not very effective and we could save
our time for better things...

Guenther
Parent - By George Tsavdaris (****) Date 2008-08-11 17:49
Not to mention that he didn't provide the games, something that would show us almost for sure of what happened and if was something wrong....
Parent - - By Fulcrum2000 (****) Date 2008-08-11 17:55
Do you have persistent hash enabled?. If so, disable persistent hash and try again. The results of R3 will remarkably improve! :)
Parent - - By George Tsavdaris (****) Date 2008-08-11 17:57
Where do you see depth 5 moves?
I don't see any move that is on ply 5....
Parent - - By Fulcrum2000 (****) Date 2008-08-11 18:12
I think it's not always a depth 5 move, it depends on the hash write depth value also. But I think we need Vas to confirm this.
Parent - By Uly (Gold) Date 2008-08-11 19:18

> I think it's not always a depth 5 move, it depends on the hash write depth value also. But I think we need Vas to confirm this.


I can confirm that usage of Persistent Hash always shows a depth of 5, so in these games Persistent Hash was not present.
Parent - - By Banned for Life (Gold) Date 2008-08-11 19:57
A persistent hash move always shows up as depth 5, even if its really depth 20...

Regards,
Alan
Parent - - By Uly (Gold) Date 2008-08-11 20:04
I guess you wouldn't want the bug fixed because your workaround gives you an advantage against those that don't know how to use Persistent Hash properly? ;)
Parent - By Fulcrum2000 (****) Date 2008-08-11 20:10
I would be even more fun the other way around. Playing the PH move instantly, showing d=20 but internally only use d=5 ;)
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka3 only 30 ELO better than Fritz 11

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill