1.Stockfish 1.5 near equal with Rybka 3 !
2.Shredder 12 very close to Rybka 3 !
Question : a modern computer chess tragedy is on way ?
Regards,
Silvian
:)
2.Shredder 12 very close to Rybka 3 !
Question : a modern computer chess tragedy is on way ?
Regards,
Silvian
:)
> 1.Stockfish 1.5 near equal with Rybka 3 !
Pics or it didn't happen.
Define "near equal"
Define "very close"
Define "very close"
This is horse-poo. My results aren't showing Stockfish 1.5 as anywhere near Rybka 3. And DS12 is good, but not THAT good.
> And DS12 is good, but not THAT good.
Is better than Naum 4 or near in strength?
Regards,
Gaмßito.
about the same. we have to wait for the CEGT/CCRL tests to tell us more.
Stockfish 1.5 is not even close to Rybka 3. But it is climbing alarmingly fast with each new version and it has to be considered the #1 threat at the moment. Those guys are really making a run for #2 and they are pushing freeware so far, so they are bound to build quite user-base.
Wait for Rybka 4 before making judgments about tragedies. And by the way, competition is great. It's the best thing for the entire computer chess community. If and when someone finally equals or surpasses Rybka that too will be a great day for computer chess. (Bad for this forum, but great for computer chess.)
Wait for Rybka 4 before making judgments about tragedies. And by the way, competition is great. It's the best thing for the entire computer chess community. If and when someone finally equals or surpasses Rybka that too will be a great day for computer chess. (Bad for this forum, but great for computer chess.)
>Stockfish 1.5 is not even close to Rybka 3
But sometimes its just a little bit close!
[Event "Blitz 5m"]
[Date "2009.10.06"]
[Round "2"]
[White "Rybka 3 32-bit"]
[Black "Stockfish 1.5 JA"]
[Result "0-1"]
[ECO "B88"]
Rybka3.ctg
Fritz12.ctg
1. e4 {B 0} c5
{B 0} 2.
Nf3 {B 0} d6 {B 0} 3. d4 {B 0} cxd4 {B 0} 4. Nxd4 {B 0} Nf6 {B 0} 5. Nc3 {B 0}
Nc6 {B 0} 6. Bc4 {B 0} e6 {B 0} 7. O-O {B 0} Be7 {B 0} 8. Bb3 {B 0} O-O {B 0}
9. Be3 {B 0} a6 {B 0} 10. f4 {B 0} Nxd4 {B 0 Both last book move} 11. Bxd4 {0.
33/13 6} Qc7 {0.04/15 8} 12. Qd3 {0.30/13 22 (a4)} b5 {0.00/16 10 (Bd7)} 13. f5
{0.30/11 2 (a3)} e5 {-0.08/16 7} 14. Nd5 {0.27/13 5 (Bf2)} Nxd5 {-0.40/17 8}
15. Bxd5 {0.27/12 0} Bb7 {-0.36/17 9} 16. Bxb7 {0.13/14 11} Qxb7 {-0.44/17 6}
17. Bf2 {0.16/14 4} Rfd8 {-0.40/16 11 (b4)} 18. Rfd1 {0.19/12 9 (Rad1)} Rd7 {
-0.32/16 12 (Rdc8)} 19. Qe2 {0.33/12 4 (b3)} b4 {-0.44/16 8 (Rc8)} 20. Rac1 {
0.10/11 5} Qb5 {-0.52/15 10} 21. Kf1 {0.15/12 5} Rc7 {-0.64/15 7} 22. Be1 {0.
16/12 7 (f6)} Qa4 {-0.44/16 5 (g6)} 23. b3 {0.14/12 3} Qa5 {-0.44/16 7} 24. c4
{0.15/12 5 (a4)} bxc3 {0.00/17 6} 25. Rd5 {0.12/13 3 (Rd3)} Qb4 {-0.08/17 5}
26. Rd3 {0.12/11 0} Rac8 {0.00/17 7} 27. Rcxc3 {0.13/13 2} Rxc3 {0.00/16 1} 28.
Bxc3 {0.13/12 0} Qa3 {0.00/17 7 (Qb6)} 29. Be1 {0.14/12 5} Qc1 {0.00/17 7 (Rc1)
} 30. Rd1 {0.19/12 4} Qc6 {0.04/16 5 (Qa3)} 31. Qd3 {0.18/11 3} g6 {0.00/14 5
(h6)} 32. a4 {0.18/11 4 (f6)} gxf5 {-0.04/15 11 (Kg7)} 33. exf5 {0.16/11 3} Kg7
{0.00/14 6} 34. Bb4 {0.16/11 10 (f6+)} f6 {-0.24/15 6} 35. Ba3 {0.17/11 5
(Qg3+)} a5 {-0.28/14 4} 36. Rd2 {0.17/12 6 (Qg3+)} Qb7 {-0.32/14 3 (Kf7)} 37.
Qd5 {0.08/11 4 (Rd1)} Qa6+ {-0.32/15 3} 38. Qb5 {0.07/13 2} Qa7 {-0.32/16 3}
39. Qe2 {0.03/12 2 (Qd3)} Rc3 {-0.56/15 2} 40. Rd3 {0.00/12 2} Rxd3 {-0.48/15 4
} 41. Qxd3 {0.00/14 2} Qd7 {-0.48/17 5 (Qb7)} 42. Qg3+ {0.00/14 3 (Bc5)} Kf8 {
-0.44/16 2 (Kh8)} 43. Qh3 {0.00/14 2} Qb7 {-0.52/16 3} 44. Bc1 {0.00/14 2} Bd8
{-0.56/15 3 (Qd5)} 45. Bh6+ {-0.03/11 2 (Qh6+)} Ke7 {-0.88/15 3} 46. Qd3 {-0.
10/11 3 (Ke2)} d5 {-1.09/15 4} 47. Bd2 {-0.18/11 3 (Qg3)} Kd6 {-1.13/14 4} 48.
h3 {-0.20/11 3 (Bh6)} h5 {-1.29/14 2} 49. Qg3 {-0.34/11 4} Qc8 {-1.17/15 2} 50.
Qg6 {-0.28/11 7 (Qd3)} Qc2 {-1.57/15 3} 51. Bxa5 {-0.44/11 1} Qd3+ {-1.81/13 2
(Bxa5)} 52. Kg1 {-0.54/13 10} Qd4+ {-1.25/14 2 (Bxa5)} 53. Kh1 {-0.25/13 1}
Bxa5 {-1.49/15 2} 54. Qxf6+ {-0.25/11 0} Kc7 {-1.25/15 2} 55. Qe7+ {-0.22/12 1}
Kc8 {-1.45/15 4 (Kc6)} 56. f6 {-0.08/13 2 (Qe8+)} Bb6 {-1.01/14 2 (e4)} 57.
Qe6+ {0.00/15 2} Kb7 {-1.13/15 4} 58. Qd7+ {0.00/15 2} Bc7 {-0.92/15 2} 59.
Qb5+ {0.00/15 1 (f7)} Ka7 {-1.21/15 2} 60. Qc6 {0.00/15 16} Qb6 {-1.73/15 1}
61. Qxb6+ {-0.80/11 3} Kxb6 {-1.17/15 2 (Bxb6)} 62. Kg1 {-1.54/13 13 (g4)} h4 {
-2.58/15 3 (Kc6)} 63. Kf2 {-1.20/11 1} e4 {-3.91/15 1 (Kc6)} 64. f7 {-1.90/11
1 (Ke3)} Bd6 {-7.95/15 1} 65. Ke3 {-3.58/14 10} Bc5+ {-9.93/15 1 (Kc6)} 66. Kf4
{-3.58/14 1} Ka5 {-14.50/15 2 (Kc6)} 67. g4 {-5.34/9 9} hxg3 {-15.95/15 3} 68.
Kxg3 {-3.58/10 0} d4 {-19.71/14 3} 69. f8=Q {-11.71/9 7 (b4+)} Bxf8 {-25.69/12
0} 70. Kf2 {-13.27/9 6} d3 {-89.09/15 2} 71. Ke3 {-13.31/8 5} Bh6+ {-93.31/15 1
} 72. Kf2 {-13.89/8 5 (Kxe4)} e3+ {-93.26/13 1} 73. Kf1 {-17.42/9 4 (Kf3)} Bf8
{-#10/12 0 (Bg5)} 74. Kg2 {-#14/8 3} e2 {-#9/10 0 (d2)} 75. Kf3 {-#13/7 1 (Kf2)
} e1=Q {-#7/8 0} 76. b4+ {-#9/5 0 (Kg4)} Kxb4 {-#6/8 0 (Bxb4)} 77. Kg4 {-#7/3
0 (Kf4)} d2 {-#5/6 0} 78. Kf5 {-#6/3 0} d1=Q {-#4/6 0} 79. Kf6 {-#5/3 0 (Kg6)}
Qh5 {-#2/6 0 (Qf3+)} 80. h4 {-#2/3 0} Qee5# {-#1/6 0} 0-1
Stockfish 1.5 is nowhere near as strong as some people believe, but we are pretty sure it is stronger than the previous version, and of course we will keep improving it.
Whether Stockfish or some other engine surpasses Rybka some time in the near or far future is, of course, neither good nor bad for the computer chess community, this forum, Vas, or us in the Stockfish team. What many people don't seem to get is that it doesn't matter at all which program is the strongest at any given time. The important things are that the human knowledge of chess programming techniques keeps increasing, that there are enough resources for new programmers to learn from, that the programs still in active development keep improving, and perhaps that some of the things we discover will some day be useful in some domain outside computer chess.
Whether program X is stronger and/or improves faster than program Y, however, is completely irrelevant.
Whether Stockfish or some other engine surpasses Rybka some time in the near or far future is, of course, neither good nor bad for the computer chess community, this forum, Vas, or us in the Stockfish team. What many people don't seem to get is that it doesn't matter at all which program is the strongest at any given time. The important things are that the human knowledge of chess programming techniques keeps increasing, that there are enough resources for new programmers to learn from, that the programs still in active development keep improving, and perhaps that some of the things we discover will some day be useful in some domain outside computer chess.
Whether program X is stronger and/or improves faster than program Y, however, is completely irrelevant.
>Stockfish 1.5 is nowhere near as strong as some people believe, but we are pretty sure it is stronger than the previous version, and of >course we will keep improving it.
Indeed, i was not impressed much with Stockfish's 1.5 play. Note that it's the first version of Stockfish i tried. Note also that i played only blitz games so in longer time controls my impression may be different. Note also that i have seen just 40-50 games of it, so after more games my opinions may be completely different.
Reasons i'm not impressed are its extremely unstable evaluation and its not so good evaluation i think it has.
Its style until now is also non impressing for me. It doesn't force too much things, although it is not a waiting engine, it doesn't make speculative moves, it doesn't make sacrifices but it wins by opponent's mistakes. It's very good at taking advantage of them.
But i see some very good results and combining this with the previous things i've said, i see a huge potential if you correct all these(and IF of course my observations are correct and its evaluation is not so good as i believe) and actually i believe it would rather be easy if you worked full time on it to make it stronger than Rybka 3 in a short period of time.
Hi George!
Stockfish's evaluation is neither better nor worse than most other programs of comparable strength. Style is a subjective issue, but most people's opinion is the opposite of yours. It is certainly designed to be speculative, and the positional terms of the evaluation function are huge. If you don't see sacrifices, you must have been unlucky with the games you have seen. The evaluation instabilities you have seen are, in fact, largely caused by speculative evaluation, combined with the somewhat unorthodox way aspiration windows are used in Stockfish.
I'm sorry you are not happy with Stockfish, but I suppose you get what you pay for, as always. If you have used some commercial chess program in the past, you are not likely to be satisfied with a free engine like Stockfish.
Working full time on computer chess is, of course, totally out of the question, even if I were paid to do it. :)
Stockfish's evaluation is neither better nor worse than most other programs of comparable strength. Style is a subjective issue, but most people's opinion is the opposite of yours. It is certainly designed to be speculative, and the positional terms of the evaluation function are huge. If you don't see sacrifices, you must have been unlucky with the games you have seen. The evaluation instabilities you have seen are, in fact, largely caused by speculative evaluation, combined with the somewhat unorthodox way aspiration windows are used in Stockfish.
I'm sorry you are not happy with Stockfish, but I suppose you get what you pay for, as always. If you have used some commercial chess program in the past, you are not likely to be satisfied with a free engine like Stockfish.
Working full time on computer chess is, of course, totally out of the question, even if I were paid to do it. :)
Hi Tord,
Would you say that the 'evaluation' algorithm you used within Stockfish 1.5 leads to improved analysis the longer the thinking time is? On the face of it this may appear to be a daft question to many readers as wouldn't all chess engines benefit from longer thinking time. In all the years I have spent analysing positions I have noticed (I have no empirical data - just gut-feel) that some engines benefit more from longer times then others - Zappa is an obvious vote that most people will associate with this.
Your views on this would be very welcome.
Would you say that the 'evaluation' algorithm you used within Stockfish 1.5 leads to improved analysis the longer the thinking time is? On the face of it this may appear to be a daft question to many readers as wouldn't all chess engines benefit from longer thinking time. In all the years I have spent analysing positions I have noticed (I have no empirical data - just gut-feel) that some engines benefit more from longer times then others - Zappa is an obvious vote that most people will associate with this.
Your views on this would be very welcome.
> The important things are that the human knowledge of chess programming techniques keeps increasing, that there are enough resources for new programmers to learn from, that the programs still in active development keep improving, and perhaps that some of the things we discover will some day be useful in some domain outside computer chess.
I agree with all of this. But competition is one key element that drives the process forward, and it is undeniable that whichever programmer has the strongest program at any given time receives attention and honors in accord with his achievement. In the Olympics you only give out one gold medal unless there is a tie. If you prefer to see this in academic terms, where the goal is the furtherment of mankind's knowledge, then that's wonderful too. However I feel pretty sure that commercial imperatives are a stronger incentive in most of life's endeavors.
I don't quite agree. Unlike in human chess, competition isn't really a major element in computer chess. The programs, of course, don't care whether they win, draw or lose. The programmers don't care much about winning per se, but rather about improving. When we see our programs play badly and lose unnecessarily, we get unhappy about our work and try to fix it. The end result looks similar to competition, but the motivation is quite different. All of us would like to improve our programs by 100 Elo points, but most of us would be even happier if all other programs would simultaneously improve by the same amount.
I agree that commercial imperatives are a stronger incentive in most of life's endeavors, but in computer chess almost everyone is an amateur, and the few people who aren't mostly compete on other arenas than pure playing strength.
I agree that commercial imperatives are a stronger incentive in most of life's endeavors, but in computer chess almost everyone is an amateur, and the few people who aren't mostly compete on other arenas than pure playing strength.
Very interesting.
My immediate mental feedback was 'Show me a good loser and I'll show you a loser' - but I respect what you say very much even if I personally find it hard to understand due to my innate nature that leads me to see the world only via crushing the competition either for money or for fun, both equally satisfying.
My immediate mental feedback was 'Show me a good loser and I'll show you a loser' - but I respect what you say very much even if I personally find it hard to understand due to my innate nature that leads me to see the world only via crushing the competition either for money or for fun, both equally satisfying.
You gotta love those VL quotes! :-D
It is interesting to read your sentiments and I can only respond by saying that in this entire planet there are very few people who have both the talent and motivation do what you do with such demonstrated skill. We end-users are on a lower plane of computer chess consciousness, really, and so it is a bit of an honor merely to elicit such replies. You may consider your team a group of amateurs but your team is now fast approaching (and may have already surpassed) that imaginary line where you cross over into the professional realm. That you do not make this formally the case is evidence of idealism and nobility...and is as disruptive as hell, as you blow the doors off weaker commercial programs. :-) I look forward to the next version, and the next, and hope you guys don't sidetrack from your improvement goals.
>All of us would like to improve our programs by 100 Elo points, but most of us would be even happier if all other programs would simultaneously improve by the same amount.
I like this attitude very much.
PS: Gens una sumus!
I think that the problem with improving other programs is that most chess programmers do not understand most of stockfish.
Everything that can help programmers to understand stockfish faster can be productive for computer chess.
Uri
Everything that can help programmers to understand stockfish faster can be productive for computer chess.
Uri
Uri,
Are you still working on Movei?
Regards
Are you still working on Movei?
Regards
No
I do not.
Recently(after the release of stockfish1.5) I gave some suggestions to the stockfish team and we will see if we get something useful from it.
I believe that movei is using some good ideas that can help stockfish but unfortunately I did not have time to understand stockfish well enough to try to implement them in stockfish.
Uri
I do not.
Recently(after the release of stockfish1.5) I gave some suggestions to the stockfish team and we will see if we get something useful from it.
I believe that movei is using some good ideas that can help stockfish but unfortunately I did not have time to understand stockfish well enough to try to implement them in stockfish.
Uri
Test 50 games:
Stockfish 1.5 JA - Rybka 3 (TP= -43 Elo)
+12 =20 -18
Conditions:
Core2Duo E6600 (2x2.4 GHz)
Windows XP x32 OC
Hashtable size: 64MB each
Time: 1'+1''
Book: Perfect 2009 (Optimised)
Book Learning: OFF
Ponder: OFF
PS: An interesting point is that Rybka started with 4-0 (without counting draws) and 7-1 later.
Stockfish 1.5 JA - Rybka 3 (TP= -43 Elo)
+12 =20 -18
Conditions:
Core2Duo E6600 (2x2.4 GHz)
Windows XP x32 OC
Hashtable size: 64MB each
Time: 1'+1''
Book: Perfect 2009 (Optimised)
Book Learning: OFF
Ponder: OFF
PS: An interesting point is that Rybka started with 4-0 (without counting draws) and 7-1 later.
Wow, look at the name. Now there's two fish in the pond.
Test 50 games:
Stockfish 1.5 JA - Rybka 2.3.2a (TP= +35 Elo)
+18 =19 -13
Conditions: Same
PS: Again a strange start +4 =4 for Stockfish and after an almost level result.
Stockfish 1.5 JA - Rybka 2.3.2a (TP= +35 Elo)
+18 =19 -13
Conditions: Same
PS: Again a strange start +4 =4 for Stockfish and after an almost level result.
Finaly,
Test 50 games:
Stockfish 1.5 JA - Stockfish 1.4 JA (TP= +56 Elo)
+19 =20 -11
Conditions: Same
PS: Until game 25 the score was dead even!
Test 50 games:
Stockfish 1.5 JA - Stockfish 1.4 JA (TP= +56 Elo)
+19 =20 -11
Conditions: Same
PS: Until game 25 the score was dead even!
-43 ELO...wow, incredible. I don't completely trust 1+1 games but this is pretty impressive nonetheless. CEGT and CCRL is going to be really interesting in the next month or so.
Huh oh, if this holds true then we'll need a Stockfish 1.6 Rybka in danger! thread.
And it's open source!
And it's open source!
I wonder now if engine really need to have something with fish in order to be great :)
I guess it will be time for all of us to decamp to the Stockfish forum!
> I guess it will be time for all of us to decamp to the Stockfish forum!
Yes! Do you have a link!?
I hope we will not have to learn Norwegian.
> I don't completely trust 1+1 games but this is pretty impressive nonetheless.
Me neither, I like to run my test matches from 5 minutes upwards.
>> I don't completely trust 1+1 games but this is pretty impressive nonetheless.
>Me neither, I like to run my test matches from 5 minutes upwards.
We have large rating list with Blitz results
Perhaps it could be done to produce a rather large list of 1+1-Results.
Then we could have a look at the differences.
Is the list significantly changed?
Are the strength differences significantly changed?
1+1 is very short, that's true. (in average a third of 5+1 or so?)
Will the differences be larger than those between 1min/move and 3min/move? (Here I don't expect important changes)
Quap
It was already demonstrated a couple of years ago that the differences were not significant. We should also consider that none other than Vas and Larry have used lightning games extensively in the past to test evaluation functions. But yet...
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill