![[nl]](/mwf/v12/flags/nl.png "Netherlands")
The book did play a huge roll in this match, I like a engine-engine match and not a book-book match.
If the rematch of Rybka-Zappa is also played at the same time of the upcoming Krammink-Anand match I see a nice option to remove the book factor.
1) Play the same number of games als Krammik-Anand at the same time
2) Openings books and tablebases not allowed
3) Use as starting position the opening played by Krammik-Anand
The engines must play white, white, black, black, Otherwise a engine will always play for the same human.
It will be difficult to decide what the starting position is. One option could be to create a database with all 2600+ games from human players at standard time of the last 5 years. The first move that is not in this database is the last move from the opening. So if a human finds a new opeming move the engine that plays for this human follows it, the response to this new opening move is up to the engine that plays for the other human.
This can also generate some publicity, when chess journalists are comparing the human-human game with the engine-engine game.
If the rematch of Rybka-Zappa is also played at the same time of the upcoming Krammink-Anand match I see a nice option to remove the book factor.
1) Play the same number of games als Krammik-Anand at the same time
2) Openings books and tablebases not allowed
3) Use as starting position the opening played by Krammik-Anand
The engines must play white, white, black, black, Otherwise a engine will always play for the same human.
It will be difficult to decide what the starting position is. One option could be to create a database with all 2600+ games from human players at standard time of the last 5 years. The first move that is not in this database is the last move from the opening. So if a human finds a new opeming move the engine that plays for this human follows it, the response to this new opening move is up to the engine that plays for the other human.
This can also generate some publicity, when chess journalists are comparing the human-human game with the engine-engine game.
![[at]](/mwf/v12/flags/at.png "Austria")
You are suggesting to put Erdo, Jeroen and Dagh out of a job. I don't think they will like that ;)
:-) I'm talking about matches that are connected to other matches like the upcoming Kramnik-Anand match.
If you connect a engine-engine match to the human WC match, then do it good and glue those 2 matches more together.
Imagine if Anand reponse at a pers conference is something like "Rybka did handle that new move from Kramnik much better then I did"
Of course Jeroen does a great job, and if it is true what some say in this forum that some positions after the book part are only objective good but not good for Rybka then that is the problem of Vas and Larry, it's not Jeroens problem :-)
If you connect a engine-engine match to the human WC match, then do it good and glue those 2 matches more together.
Imagine if Anand reponse at a pers conference is something like "Rybka did handle that new move from Kramnik much better then I did"
Of course Jeroen does a great job, and if it is true what some say in this forum that some positions after the book part are only objective good but not good for Rybka then that is the problem of Vas and Larry, it's not Jeroens problem :-)
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
A match without books is in my view meaningless, as the various engine games played by the engine rating sites
produce a highly accurately ranking of the engines.
What makes a high profile engine match interesting, is that it allows any type of chess program on the allowed hardware.
To limit the opening books in a high profile engine match is just inviting the teams to include the opening book implicitly in the code of the program.
This can be done blatantly by just including the book as part of the program, or it can be done less blatantly by tuning the parameters in the opening phase so the program will choose the intended moves in certain critical positions. To play with restricted book in a high profile
match is creating a lot of mess, and is not doing anything good for chess programming.
produce a highly accurately ranking of the engines.
What makes a high profile engine match interesting, is that it allows any type of chess program on the allowed hardware.
To limit the opening books in a high profile engine match is just inviting the teams to include the opening book implicitly in the code of the program.
This can be done blatantly by just including the book as part of the program, or it can be done less blatantly by tuning the parameters in the opening phase so the program will choose the intended moves in certain critical positions. To play with restricted book in a high profile
match is creating a lot of mess, and is not doing anything good for chess programming.
![[us]](/mwf/v12/flags/us.png "United States")
If they can be put out of a job because of using no=books in a match, then they never had a job in the first place....in other words...the need for a strong book will still be important for other things in computer chess....Don't you think?:)
Remove the books out of the equation = FRC :-) At least that's my ideal scenario
But I don't think Zappa play it...
But I don't think Zappa play it...
I thought the idea of a chess game is to see who could best win from the starting position, and engines having their own books geared toward their "own" personalities is really the only way for this to happen. That is why I think such ideas of these ridiculous random "general" book matches are absurd. If Rybka wants to play that way, fine--don't require the Zappa team to play that way. As it's turning out, Rybka might actually have an advantage in playing that way...
The rest of you guys are trying to gently chide Herbert. Let me say it straight: this is a preposterous idea.
If you want to establish which engine is stronger, if you want engine-engine matches where books are equal, join CEGT or CCRL--both of which provide an objective environment to measure such things and perform an invaluable service. You aren't going to determine which engine is stronger in a ten-game match unless the score is something like 9-1 or 10-0.
Conversely the format that was used in this match was identical to the one that has been used in computer chess tournaments since the dawn of the hobby. The opening book is an integral part of match/tournament preparation, an extraordinarily arduous endeavor, requiring the best efforts of top minds that are pathologically immune to tedium.
Finally: don't you just love the piquant comments of Dagh and Jeroen when they are tweaked by the likes of Alan? Come on, it's the best part of this forum.
If you want to establish which engine is stronger, if you want engine-engine matches where books are equal, join CEGT or CCRL--both of which provide an objective environment to measure such things and perform an invaluable service. You aren't going to determine which engine is stronger in a ten-game match unless the score is something like 9-1 or 10-0.
Conversely the format that was used in this match was identical to the one that has been used in computer chess tournaments since the dawn of the hobby. The opening book is an integral part of match/tournament preparation, an extraordinarily arduous endeavor, requiring the best efforts of top minds that are pathologically immune to tedium.
Finally: don't you just love the piquant comments of Dagh and Jeroen when they are tweaked by the likes of Alan? Come on, it's the best part of this forum.
There is another option besides those mentioned in the replies so far. Choose the first 5 moves (each) by randomly sampling a standard database of top GM games, playing it twice with reversed colors, but allow the use of opening books. This is different from CCRL/CEGT where normal books are not used. This removes the factor of trying to guess what the other "booker" will choose, and puts the emphasis on the overall quality of both the program and the book. It also gives no incentive to "cheat" by making the engine duplicate the work of the book. The book remains a factor, but changing it during the match becomes almost pointless, especially if the second game of each pair is played right after the first, so no changes can be made between those games.
Intriguing concept. I am sure Dagh and Jeroen would hate it, though.
Dagh would hate it, but I'm not sure about Jeroen, as he is pretty good at doing a thorough opening book.
You would be going from 1 starting position to a huge number of starting positions, but you would still want to have book coverage after move 5. The only way to do this would be collect a huge number of quality games and have an automated process to check each line on a bunch of computers. This would make the work done at University of Warsaw very valuable. I hope it comes to pass! :-)
What work are you talking about that is being done at Warsaw University? I did a quick Google search, but couldn't really find anything...
Work was done to take in .PGN files, pull out all unique positions, convert all positions to FEN ina white to move format, evaluate all positions with Rybka 2.2n2 (this was a while back) for 30 seconds each, and store the results in a database. The evaluations were done on 17 computers in the Warsaw University Computer Lab. This worked fine except when the pesky students were trying to use the machines, didn't like the fact that the computers were slow as molasses, and terminated the Rybka processes. After generating a fair amount of ill will, we managed to finish evaluation of ~1.8 million positions.
This was just a test but the lab will be getting a larger supply of newer, faster machines soon, which will allow us to analyze a larger set of positions. One thing I've been thinking about is possibly to analyze the positions with both Rybka and Zappa and flag the positions where the evals are very different. These would be the types of positions that the engine with the correct eval should be aiming for.
Regards,
Alan
This was just a test but the lab will be getting a larger supply of newer, faster machines soon, which will allow us to analyze a larger set of positions. One thing I've been thinking about is possibly to analyze the positions with both Rybka and Zappa and flag the positions where the evals are very different. These would be the types of positions that the engine with the correct eval should be aiming for.
Regards,
Alan
Alan, of one thing I'm certain... correct evals are a sine qua non for any chess engine. I think your idea about having the same positions evaluated by Rybka and Zappa and then flagging the occasions where they vary greatly and then,presumably,
determining which engine evaluated more accurately position by position is a terrific one. Except for finding a forced mate the search function is dependent on proper evals. Perhaps using positions with known results and then comparing the evals
of each engine at various stages of each game might yield useful results more quickly.
regards,
David
determining which engine evaluated more accurately position by position is a terrific one. Except for finding a forced mate the search function is dependent on proper evals. Perhaps using positions with known results and then comparing the evals
of each engine at various stages of each game might yield useful results more quickly.
regards,
David
As you point out, doing the evals and flagging the ones that differ significantly is the easier part. The tougher part is figuring out which of the following cases occured when the evaluations diverged:
1) Engine A was right, engine B was wrong,
2) Enging A was wrong, Engine B was right, or
3) Both engines were way off base.
Once this difficult challenge was met, the next step would be to find these positions as close to the start position as possible and then try to set up opening lines that led to them.
Regards,
Alan
1) Engine A was right, engine B was wrong,
2) Enging A was wrong, Engine B was right, or
3) Both engines were way off base.
Once this difficult challenge was met, the next step would be to find these positions as close to the start position as possible and then try to set up opening lines that led to them.
Regards,
Alan
I like this idea. Just don't use Junior or else you'll have to flag all of the positions.
I don't own Junior, but I have noticed her evaluations are frequently schizophrenic.
Yeah, but not as schizophrenic as Naum's.
Yes; Naum, while a strong engine overall, is possibly the weakest in opening play out of the top dozen commercial engines, and should probably not be trusted on such things as "evaluation out of book". The same goes for Junior, which is also quite weak in opening play.
Two engines that would probably be very reliable at such evaluations, i.e. at least as reliable as Rybka, are Loop and Ktulu 8.
Two engines that would probably be very reliable at such evaluations, i.e. at least as reliable as Rybka, are Loop and Ktulu 8.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
I'm not sure Rybka is a reliable tool for opening analysis. Since you said Ktulu 8 and Loop 10.32f are the best engines for opening analysis, and that's the stage of the game I analyze the most, I've been comparing their analysis to Rybka's. It turned out Rybka liked the moves selected by Ktulu and Loop (more than her own choices mainly when Rybka's choice was Bd3 with pawn on d2 or Nc6 with pawns on c7 and d5) and so I saved time by NOT using Rybka (at all) to analyze the opening moves.
![[de]](/mwf/v12/flags/de.png "Germany")
At the very early stage of the opening Rybka 2.3.2a and earlier versions sometimes give strange moves, but the new versions (e.g. the version that played in Mainz) are much better in this, mainly thanks to some new algorithms by Larry.
If you look at an opening the analysis of early middlegames is quite important and there Rybka is quite good, so I think Rybka is a good analysis tool (and that's what she's made for). Since the most top players use Rybka for analysis of openings, Rybka seems to be a good choice.
If you look at an opening the analysis of early middlegames is quite important and there Rybka is quite good, so I think Rybka is a good analysis tool (and that's what she's made for). Since the most top players use Rybka for analysis of openings, Rybka seems to be a good choice.
Of course when analyzing late opening and early middle game I switch to Rybka (And Zappa Mexico), I was just talking about very early moves out of book. I'm happy to read that latest Rybka won't have these problems :)
![[za]](/mwf/v12/flags/za.png "South Africa")
I've seen Rybka play against Fritz 10 without a book, Rybka played the opening very badly and killed her very quickly. I don't think this is a good idea.
In my opinion books are part of the fun. So "No" to your proposal.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill