< Earlier Kibitzing · PAGE 2 OF 2 ·
|Oct-07-07|| ||psmith: <nimh>
The use of "mistake" for not always choosing Rybka's highest rated move strikes me as a... mistake.
Especially when the difference between moves is slight -- as between "0.08" and "-0.19" -- are you really that confident in the evaluation function?
|Oct-07-07|| ||Gypsy: <psmith> Since it is quite clearly stated that threshold for flagging a move is 0.25 pt in Rybka's evaluation, I find no real problem with the label. Indeed, "inaccuracy" would probably be both more respectful and more fitting label in most cases, buut it is clear to me that no disrespect is meant.|
Contrary, seeing how very few moves altogether are flagged -- and also matching those with the ones pinpointed in Morphy's annotations -- it all gives me ever more respect for the talent and skills of Messrs. La Bourdonnais, McDonnell, and Morphy.
|Oct-08-07|| ||nimh: The lower threshold the better it is to compare older games with newer ones. In the case of, say, threshold 50 I think we would see only 2-3 inaccuracies per 100 moves in games between 2700-rated players.|
<Gypsy> May I ask, what's your rating and have you checked you games with the aid of computer?
|Oct-08-07|| ||psmith: <Gypsy> <nimh>|
I don't think that .25 is a reasonable threshold. This has to do with the fact that in the end you are relying on the evaluation function built into Rybka, and a different evaluation function with the rest of the programming unchanged might well produce just as good (successful) play with a somewhat different style. When you are talking about an evaluation of +.13 versus -.13, you are talking about a difference that might change with a slightly tweaked evaluation function.
It is my understanding that Fritz evaluates positions between .25 and -.25 as "=". (At least, this is what Fritz 5.32 does.) This, to me, indicates that the choice between a +.13 move and a -.13 move is a matter of style more than anything else, as far as the program can tell -- that is, that the difference here is within the program's margin of error, so to speak.
I don't have access to Rybka, so I don't know if it includes qualitative evaluations like , etc. But I think the Fritz rule above is reasonable. And by my lights that means a threshold of .5 would be much more reasonable than .25.
By the way, I haven't played over the board in 20 years. My highest USCF rating was about 2100, in about 1985. I am not sure what that equals in 2007 ratings -- has there been inflation? I haven't tested my own games against a computer but I am sure I would find many mistakes in them if I did. And I would recognize them as real mistakes.
|Oct-08-07|| ||psmith: Let's try this: consider .25-.5 difference to be an inaccuracy, .5-1 to be an error, and 1 and above to be a major error (like dropping a pawn). What do we get?|
De La Bourdonnais
2 major errors
This gives a more informative picture of what went on.
|Oct-08-07|| ||darth pawn: I dunno what Morphy says about McDonnell playing badly in the opening, but before 15. ... f4?! the game was actually in Black's favour ( ) according to Fritz 8.|
Also, before 23. ... Nc8?? (which resulted in mate) Fritz says the position is , but Black could still possibly draw if White wasn't careful. Labourdonnais' style isn't really cautious, so chances are that McDonnell would find a draw there.
|Oct-08-07|| ||Gypsy: <nimh, psmith> My 'true rating' is anybody's guess as I have not played a rated game in decades. (And friendly games, of course, are a different ball of wax.) Referring to a really old scale, my subjective estimation is that I am about the level that used to be a candidate of master -- a dilettante in openings, fair in middle-games, and decent in end-games; good in technical chess, uneven in tactics.|
I do not use computer engines. But I do benefit when others post computer evaluations. Because of cg challenge games, team analysis, and <random visitor> forum, I think I have by now calibrated reasonably well what Rybka means when she reports 0.x pawn difference or so.
Of course, splitting the flagged moves into 'inaccuracies', 'errors', and 'blunders' -- provided that the reports do not, by doing so, become too busy -- is a sound idea.
|Oct-08-07|| ||nimh: As for the weights of mistakes I have by the way already noted the evaluation differences between the best and actual moves.|
|Oct-08-07|| ||psmith: <nimh> Yes, that's how I came up with my more refined division in the example of this game. But my point is that at a fairly high level of refinement it is not at all obvious that Rybka is just plain right all or even most of the time.|
Suppose a player chooses a move that Rybka rates +3.44 and there is another move that Rybka rates +3.77. You will mark the player down as having made a mistake even though there may be no way for his opponent to recover after the move played. That just doesn't seem right to me. Even "inaccuracy" seems too strong in such a case.
|Oct-08-07|| ||nimh: I indeed cannot exclude the possibility that a 0.25 gap in Rybka's opinion may objectively be substantially smaller. However I don't have enough time or processing power, so I'm gonna to stick to current threshold. It shouldn't be a big matter since all games are treated under the same conditions of analysis.|
|Oct-11-07|| ||nimh: I'm forced to admit that psmith is right. I compared different engines in same position, most of them agreed with wat was the best move, but evaluations differed sometimes more than 0.25.|
Conclusion: I'd better switch to a threshold of 0.33.
|Oct-11-07|| ||nimh: Correction, new threshold 0.33.
De La Bourdonnais 2 mistakes:
14.Bc2 -0.19 (14.Re1 0.16)
15.Qe2 -0.30 (15.Re1 0.08)
McDonnell 4 mistakes:
15...f4 0.62 (15...Nb4 -0.30)
17...Bf7 1.25 (17...Bf5 0.73)
20...Bc4 2.59 (20...Rc8 1.59)
23...Nc8 #6 (23...Bf6 3.19)
|Feb-26-08|| ||wolfmaster: Cool pawn mate!|
|Jun-05-08|| ||Ulhumbrus: 15 Qe2 prepares to occupy the central post e4 and the opened diagonal b1-a7 in the event of ...f4|
|Jun-18-08|| ||al wazir: I dunno. 23...Bg5 looks better for black than either the line played or Morphy's suggestion:|
A) 24. Qh8+ Kf7 25. Qh7+ Kf6 26. Rxe8 Rfxe8 27. Kxf1. White has ++++ vs. ++.
B) 24. Qxf8+ Qxf8 25. Nxf8 Rxf8 26. Rxf1, and white has ++ vs. .
But not C) 24. Qxg5 Qxe1 or D) 24. Rxe8 Bxh6.
|Jun-18-08|| ||mezzieh: Excuse me gentlemen, but I have a problem. I remember Botvinnik (White) played quite a good game against guy who also left his Knight on b6 useless and never managed to activate it. Can anyone post here a link to this game?|
Thanks in advance.
|Jun-18-08|| ||kevin86: I think it is rather comical to have all the major forces chase black's king into the opening and then mating the king with a little pawn move.|
|Jun-18-08|| ||Lutwidge: If you're looking for games with abandoned knights on b6, my guess is searching for "Alekhine Defense" might bring up a few. :)|
|Jun-18-08|| ||Marmot PFL: La Bourdonnais, the last of the great French chess players.|
|Jun-18-08|| ||PinnedPiece: I think chess masters must be able to actually see the squares on the board under attack around the king, as a different color. |
One last square to color, let it be done ...g4 so!
|Jul-15-08|| ||patzer2: White commences a decisive assault on the weak Black castled position with 21. Qh6!|
Morphy's analysis is amazingly accurate, even by today's standards when games are checked and double checked with computers for accuracy. After the followup 22. Bxg6! White will win, though, as Morphy notes, with the stronger defense 23...Bf6 Black can prolong the game a bit longer.
One of many winning possibilities after 23...Bf6 is 24. Rxe8 Rfxe8 25. Kxf1 Nc4 26. Ne2 Rad8 27. b3 Nb6 28. Ngf4 Bg7 29. Qg5 Rd7 30. Nh5 Kf8 31. Qc5+ Rde7 32. Nhf4 Nd5 33. Nxd5 cxd5 34. g4 b6 35. Qd6 Kg8 36. Qxd5+ Kh8 37. Ng3 Re1+ 38. Kg2 Rf8 39. Nf5 Rf6 40. g5 Rf8 41. Qd7 Rg8 42. Qf7 Bh6 43. g6 Rxg6+ 44. Qxg6 Bf8 45. Qf7 with mate to soon follow. Notice how the advance of the King-side pawns in this variation facilitates White's attack, just as Morphy indicated.
|Oct-01-09|| ||Ulhumbrus: <mezzieh: Excuse me gentlemen, but I have a problem. I remember Botvinnik (White) played quite a good game against guy who also left his Knight on b6 useless and never managed to activate it. Can anyone post here a link to this game?
Thanks in advance.> Botvinnik played a great many games, and a N out of play on b6 may occur in many games. Can you tell us anything more that you can remember about the game, such as the opponent, the opening, the pawn structure or the type of ending?|
|Oct-01-09|| ||laskereshevsky: .... this is the game mentioned in the above post:
Botvinnik vs M Yudovich Sr., 1933
|Feb-01-12|| ||Knight13: <6... Be7: See notes on previous games at the same opening. > For those who don't know, Morphy was referring to the better move 6... Bd6.|
|May-13-12|| ||LoveThatJoker: Guess-the-Move Final Score:
La Bourdonnais vs McDonnell, 1834.
YOU ARE PLAYING THE ROLE OF LA BOURDONNAIS.
Your score: 63 (par = 51)
< Earlier Kibitzing · PAGE 2 OF 2 ·