|
< Earlier Kibitzing · PAGE 190 OF 190 ·
Later Kibitzing > |
Jan-03-10
 |
| notyetagm: <DATABASE SEARCH REQUEST> Could someone please search their database for the following skeletal position? 
click for larger viewThe White player is an Englishman and I believe Black is an Asian. Black has just blundered with 1 ...  xf2??, allowing the <DISCOVERED ATTACK> 2  d2-e4!, winning the Exchange for a pawn. The point is that the White b2-rook is somehow defended so 2 ... f2xb2 fails to the <ZWISCEHNSCHACH> 3 e4xf6+ and *then* the recapture on b2. If your search fails please try moving the White b2-rook to c2. Please post your response in my forum.
Thanks |
 |
Jan-04-10
 |
| Red October: http://www.youtube.com/watch?v=k4aR... |
|
Jan-11-10
 |
| RandomVisitor: <Red October>Very interesting. Thanks for that. |
|
Jan-13-10
 |
| David2009: John, I have now posted Eliskases vs T Borges, 1946 in reply to your comments two days ago. In brief, you and Rybka-3 are right. |
|
| Jan-13-10 |
| ETERNITY: slowly surely RandomVisitor |
|
| Jan-18-10 |
| cormier: hi <<RV> i want to wish you succes with the positions element formula. i'm on roggoff for now.....tks |
|
Jan-21-10
 |
| YouRang: Hello <RV>. I have a little favor to ask: I was viewing the F Caruana vs L Dominguez, 2010 game on chessok.com, which provides Rybka 3 analysis. After 52...b3, the game had this position:
click for larger viewThe Rybka 3 on chessok.com evaluated <53.g5> at +2.45 (practically winning for white), but my Rybka 2.2 is calling it almost even at +0.31. When you get a moment, can you run this with your Rybka 3 to see if yours agrees with chessok.com, and if so, explain the winning continuation after <53.g5 b2>? I'm seeing <54.g6+ Kh6 55.g7+ Kh7 56.Qf7 Qxf2+ 57.Qxf2 c1=Q> which is a tablebase draw. |
|
Jan-22-10
|
| RandomVisitor: <YouRang>do you have 5- or 6-piece tablebases? As an experiment, I shut off tablebases and the position evaluates to +2.82. It appears that the chessok Rybka has tablebases turned off or is using something like 4-piece tablebases. My eval is showing 0.00 with 6-piece tablebases active. Hope that helps. |
|
Jan-22-10
|
| YouRang: Thanks <RV>. No, that's the funny thing, I don't have tablebases, and yet Rybka 2.2 seems to understand the drawishness of the position better than Rybka 3. That surprised me a bit. Thanks again for checking. :-) |
|
Jan-22-10
|
| YouRang: Continuing above thought: When I first observed the high eval from Rybka3 on chessok.com, I assumed that it was because chessok.com WAS using tablebases. But then I followed the line above myself only to find that when I got to the final position with 6 pieces, the tablebase site called it a draw. I'd be tempted to report the situation to Rajlich, but then again, maybe he gets reports like this all the time. Rybka 3 is obviously better in general, but one can hardly expect it to be better in every case. |
|
Jan-22-10
|
| RandomVisitor: <YouRang>Vas appears to have removed some knowledge from Rybka 2.2, thinking that people using Rybka 3 would use tablebases and the reduced knowledge might speed things up a bit. Go figure. |
|
Jan-22-10
|
| YouRang: Maybe so. Interestingly, in this case the logic removed from Rybka 2.2 actually helped it -- at least in the sense that Rybka 2.2's eval of +0.31 was much closer to the truth (0.00) than Rybka 3's +2.45. I'm a little surprised that chessok.com doesn't use tablebases. I would think that endgame evals would be MUCH better with tablebases in effect. |
|
Jan-23-10
|
| RandomVisitor: <YouRang>The other possibility here is that Rybka 3 is configured to use tablebases <rarely>. There is a configuration parameter that you can set which controls how often Rybka decides to access the tablebases. I will give chessok an e-mail and let them know what you have discovered and perhaps they can get Vas to recommend a tablebase setting for Rybka 3. |
|
Jan-27-10
|
| YouRang: <RV> Interesting. My suspicion is that chessok.com isn't using tablebases at all, but perhaps you'll learn differently. But like I said, my first assumption was that I was wrong, since Rybka 3 is clearly stronger and I've noticed that Rybka 2.2 (at least the x64 version) is a little buggy. BTW, I was wondering: Are 4/5/6-piece tablebases you use available as an option when you buy Rybka 3, or do you get them as a separate package? Also, are they the full tablebases, or is it the 'bitbase' (that just use 1 bit to note if a position is won or not)? |
|
Jan-28-10
|
| RandomVisitor: <YouRang>The tablebases are available for free, in various places online, but they take forever to download. I just bought a DVD collection of them from one of the online chess book places. They are just regular files, they take up about 39 Gigabytes of storage on your hard drive. |
|
Jan-28-10
|
| YouRang: <RV> Thanks for the information. I can't imagine trying to download that much data, lol. Wait until we get 7-piece tablebases, which I've heard would consume ~100 TB. :-o BTW, The idea that Rybka has an option to minimize use of tablebases suggests that tablebases slow things down. I would have guessed that tablebases would speed things up because it effectively prunes the tree with one very precise snip. |
|
Jan-28-10
 |
| zanshin: <The idea that Rybka has an option to minimize use of tablebases suggests that tablebases slow things down.> <YR> Excuse me while I butt in here. Unless I am missing something, this is a GUI setting - not an engine option. In Aquarium, you can set the engine to use TBs: frequently, normally, rarely or never. The default is rarely. And yes, TB access slows down calculations, so it would make sense to use them never in the opening, all the way to frequently in the endgame. |
|
Jan-28-10
|
| YouRang: <zanshin> Are you sure about that? Even with Rybka 2.2, the option to use tablebases is part of the Rybka configuration screen -- not part of the GUI options. However, you're comment about usage options interests me. I would have thought that the engine would only look at the tablebase when it's about to evaluate a position that has N or fewer pieces, where N is the maximum number of pieces supported by the tablebase. If that were the case, then the engine shouldn't consult the tablebase in the opening anyway, even if you selected the option to use it frequently. Of course, if my ideas about how things *ought* to work were true, then you would always use tablebases because they would make for faster and more accurate evaluations. In fact, there would be no need to tell the engine to use it "rarely". Evidently though, doing a tablebase lookup must have to hit the hard drive, which is expensive. PS. You're always welcome to "butt in" as far as I'm concerned. |
|
Jan-29-10
|
| zanshin: <YouRang: <zanshin> Are you sure about that? Even with Rybka 2.2, the option to use tablebases is part of the Rybka configuration screen -- not part of the GUI options.> <YR> You are right. They are Rybka options changed through the GUI. I had assumed incorrectly that the engines had the same (or very similar) options. I guess I have been using Rybka so much that I had taken Nalimov usage forgranted. My bad! |
|
Jan-29-10
|
| zanshin: <YR> Thanks for the invitation to butt in ;-) but in this case necessary to correct an error. I agree with you though that an engine should be able to decide on its own how frequently TB access is needed. Surely, a piece count is easy enough to do - then using some sort or probability factor, set an option on whether or not TBs are needed. However, I have always assumed TB access is expensive because often they are stored on external drives. It makes me wonder though how other engines decide on TB usage given that it is not an option setting. I remember that the setting of 'never' was included because people were complaining about problems with TB access. Now, I can understand this, but it seems to me the engine would do best by deciding for itself how often TBs would need to be accessed based on the given position. |
|
Jan-31-10
|
| RandomVisitor: <RandomVisitor has kibitzed 6500 times to chessgames >What? It can't be that many. Most of them are just me posting computer analysis. |
|
Feb-04-10
|
| RandomVisitor: Check out the "new" firebird chess engine - it's free and just as good as Rybka... http://www.chesslogik.com/firebird.... |
|
Feb-05-10
|
| zanshin: <RV> If you are advertizing Firebird, then I will take it as a stamp of approval that it is not a Rybka clone ;-) In an engine match I ran (3 mins) per side, it crushed Rybka 3 by a score of 16 to 4 on my Quad. At longer time controls (standard), Rybka won 5.5 to 4.5. I am running more tests this weekend. |
|
Feb-05-10
|
| zanshin: Also, if you haven't already, be sure to read this before it gets removed: http://www.nybooks.com/articles/23592 |
|
Feb-05-10
|
| RandomVisitor: <zanshin>Interesting article, thanks for that. Firebird might be a Rybka clone - but who will ever know? |
|
 |
|
< Earlier Kibitzing · PAGE 190 OF 190 ·
Later Kibitzing > |
