How important is an opening database to a computer?

[Chessqueen]

In traditional chess there are thousands of openings with different lines and variations and there is no match for the storage capacity of a computer, therefore none of the best top 5 players can come even close to outperform a computer in the opening stage. Knowing that I also noticed that there are no serious chess match between Computer and Top humans recorded on Chess960, and I believe that the human might fair a little bit better in chess960 than standard chess against a computer relying only on opening knowledge and principle of good openings. It would be interesting to see how Komodo will do against GM Nakamura or MVL with f2 or f7 pawn odd on 5 different random chess960 positions.

[Dann Corbit]

As far as strength goes, it is measured at 121.7 Elo here for a very good book verses no book (wide error margin):
https://sites.google.com/site/computers ... book-cs-19

And 50 Elo here (the only difference between Brainfish and Stockfish is Brainfish has a special opening book):
https://www.sp-cc.de/
The 50 Elo measurement does not take the fact that the regular SF was given a book (just not as good as the brainfish book), so it is harder to say what this means.

I suspect that with TCEC conditions a really, really good book would add 25 Elo or so. That is because the hardware is so powerful that books are less valuable since the engines will be looking ahead somewhere around 18 full moves.

On the other hand, books will always have a big value for chess tournaments to make us want to watch them: Variety.
Nobody wants to watch 200 English games in a row or 200 French or 200 Sicilian games in a row. Without a book, we won't see any variety.

[Dann Corbit]

It would be interesting to see how Komodo will do against GM Nakamura or MVL with f2 or f7 pawn odd on 5 different random chess960 positions.

I think it will turn out the opposite to what everyone supposes. Chess 960 will benefit the computer and not the human, and by a lot.

What I mean by that is we have hundreds of years of world class analysis for standard chess openings, both by machines and by humans.
For Chess 960, we only have a couple decades.

This does not benefit the human, it benefits the computer (when it comes to chess 960).
The computer will flawlessly look ahead to great depth. The human is relying on very little well tested out and analyzed opening theory, so he is on his own.

I strongly believe that this situation strongly favors the computer NOT the human. Humans can have advanced study of well known standard chess openings that are closed or that have sound gambits. Computers will struggle more with these. Those areas give the humans a tiny boost (though it is far too late to imagine a return to human dominance -- we can't even win in Go any more).

I know that Chess 960 was supposed to be a big benefit in the battle with the computers. I don't think it will ever achieve that.

[Chessqueen]

If I understand correctly Alpha Zero does not use an opening book based on previous GM games, but instead it created its own opening book from thousands of games by playing against itself, if that is the case computer are capable of creating better opening books than what is collected from previous GMs games.

[jdart]

There are quite a few opening positions where search may lead you astray at tournament time controls.

Some opening analysis in books comes from correspondence games, where computer assistance is allowed and players have days to make a move.

--Jon

[jp]

If I understand correctly Alpha Zero does not use an opening book based on previous GM games, but instead it created its own opening book from thousands of games by playing against itself, if that is the case computer are capable of creating better opening books than what is collected from previous GMs games.

Not really its own opening book, but its own opening play, yeah.
But no, it did not play better openings than the best GM opening theory. Just better than having no opening book at all.

[Chessqueen]

It would be interesting to see how Komodo will do against GM Nakamura or MVL with f2 or f7 pawn odd on 5 different random chess960 positions.

I think it will turn out the opposite to what everyone supposes. Chess 960 will benefit the computer and not the human, and by a lot.

What I mean by that is we have hundreds of years of world class analysis for standard chess openings, both by machines and by humans.
For Chess 960, we only have a couple decades.

This does not benefit the human, it benefits the computer (when it comes to chess 960).
The computer will flawlessly look ahead to great depth. The human is relying on very little well tested out and analyzed opening theory, so he is on his own.

I strongly believe that this situation strongly favors the computer NOT the human. Humans can have advanced study of well known standard chess openings that are closed or that have sound gambits. Computers will struggle more with these. Those areas give the humans a tiny boost (though it is far too late to imagine a return to human dominance -- we can't even win in Go any more).

I know that Chess 960 was supposed to be a big benefit in the battle with the computers. I don't think it will ever achieve that.

It is true that we have hundreds of years of world class analysis for standard chess openings, both by machines and by humans.
For Chess 960, we only have a couple decades. But at the same time when a GM face a computer with a huge database opening of most of the previous best opening lines ever played, the GM is at a disadvantage because NOT even a GM with photographic memory can remember thousands and thousands of opening lines and variations and that is the reason why in chess960 where there is no opening databases the human GMs at least can rely on general opening principles even if the computer can out calculate and come out with better opening in Chess960. I do not know how other players feel when they are face with a computer that no matter what you play will instantly reply with the best move during the opening stage in standard chess, but when I set my Fritz16 to play at 2050 at standard chess, the computer win by 58 % of the time, and when I play Chess960 at the same setting of 2050 which is my USCF rating I manage to win 52% of the time, and that is after I read the book Play Stronger Chess by Examining Chess960. Honestly the book showed me how to open the game with different Chess960 positions depending on the position of the Bishops, Queen, and Knights, but because of the opening database of of Fritz16 that does NOT forget lines and variations like we humans I do worse in standard chess than in Chess960.

PS: In order to give any GM a better shot of coming even out of the opening the computer will have to be given an opening database of all the games opening previously played by that particular GM up to probably move 8 or 10 and of course the computer will out-calculate the GM in the middlegame, and that is why for example GM Nakamura or MVL should start playing with program rated around 2600 CCRL and depending on how they come out of the match either replace the program by increasing it or decreasing its CCRL rating. By doing this or using the GM personal openings and place it on the computer it will benefit the GM to improve his own opening repertoire when faced against another human.

[Chessqueen]

It would be interesting to see how Komodo will do against GM Nakamura or MVL with f2 or f7 pawn odd on 5 different random chess960 positions.

I think it will turn out the opposite to what everyone supposes. Chess 960 will benefit the computer and not the human, and by a lot.

What I mean by that is we have hundreds of years of world class analysis for standard chess openings, both by machines and by humans.
For Chess 960, we only have a couple decades.

This does not benefit the human, it benefits the computer (when it comes to chess 960).
The computer will flawlessly look ahead to great depth. The human is relying on very little well tested out and analyzed opening theory, so he is on his own.

I strongly believe that this situation strongly favors the computer NOT the human. Humans can have advanced study of well known standard chess openings that are closed or that have sound gambits. Computers will struggle more with these. Those areas give the humans a tiny boost (though it is far too late to imagine a return to human dominance -- we can't even win in Go any more).

I know that Chess 960 was supposed to be a big benefit in the battle with the computers. I don't think it will ever achieve that.

It is true that we have hundreds of years of world class analysis for standard chess openings, both by machines and by humans.
For Chess 960, we only have a couple decades. But at the same time when a GM face a computer with a huge database opening of most of the previous best opening lines ever played, the GM is at a disadvantage because NOT even a GM with photographic memory can remember thousands and thousands of opening lines and variations and that is the reason why in chess960 where there is no opening databases the human GMs at least can rely on general opening principles even if the computer can out calculate and come out with better opening in Chess960. I do not know how other players feel when they are face with a computer that no matter what you play will instantly reply with the best move during the opening stage in standard chess, but when I set my Fritz16 to play at 2050 at standard chess, the computer win by 58 % of the time, and when I play Chess960 at the same setting of 2050 which is my USCF rating I manage to win 52% of the time, and that is after I read the book Play Stronger Chess by Examining Chess960. Honestly the book showed me how to open the game with different Chess960 positions depending on the position of the Bishops, Queen, and Knights, but because of the opening database of of Fritz16 that does NOT forget lines and variations like we humans I do worse in standard chess than in Chess960.

PS: In order to give any GM a better shot of coming even out of the opening the computer will have to be given an opening database of all the games opening previously played by that particular GM up to probably move 8 or 10 and of course the computer will out-calculate the GM in the middlegame, and that is why for example GM Nakamura or MVL should start playing with program rated around 2600 CCRL and depending on how they come out of the match either replace the program by increasing it or decreasing its CCRL rating. By doing this or using the GM personal openings and place it on the computer it will benefit the GM to improve his own opening repertoire when faced against another human.

To prove my point if you take the same Deep Fritz that played against Kramnik back in 2006 using the same hardware and same opening database that Deep Fritz used back 12 years ago, GM Kramnik will more likely beat Deept Fritz since there has been several opening lines and variations improvement in the last 12 years, but if you use the same Deep Frit and same hardware used back in 2006 and simply replace all the new openings innovations with a new database up to now, since the program will NOT forget any lines and variations the program will again have an advantage over Kramnik.

I do not know what was the CCRL rating of Deep Fritz back in 2006 using the same hardware used for this match, but GM Kramnik was very close to Deep Fritz back then. Kramnik is not as strong as he was back then in 2006, but we can test the same Deep Fritz with the same hardware or one that can calculate very close to the hardware used back in 2006 and use it against GM MVL, GM Nakamura, or GM Carlsen to compare the strength of Kramnik back in 2006 to GM Carlsen nowadays. This is like having a time machine since we can NOT bring Kramnik to the same strength as he was Younger by 12 years but we have the equipment to compare against the Current Champion or even GM Caruana now.
https://www.youtube.com/watch?v=7q7BEQ6jSD8

[Chessqueen]

Kraminik back in 2006 was only rate around 2750 FIDE and Deep Fritz was rated 3110 by CCRL and t Kramnik almost drew the match, therefore there is a difference of 300 rating discrepancy that are very close to Humans FIDE rating and CCRL. I strongly believe that if Magnus ever decide to play vs a computer DF would be a perfect match that would be very close.

[Uri Blass]

Kraminik back in 2006 was only rate around 2750 FIDE and Deep Fritz was rated 3110 by CCRL and t Kramnik almost drew the match, therefore there is a difference of 300 rating discrepancy that are very close to Humans FIDE rating and CCRL. I strongly believe that if Magnus ever decide to play vs a computer DF would be a perfect match that would be very close.

I do not see 3110 for old Deep Fritz by the CCRL list

http://www.computerchess.org.uk/ccrl/40 ... +opponents

Deep Fritz 10 was stronger than the engine that beat kramnik and
I see
Deep Fritz 10 4CPU 2831 +20 −20 49.1% +5.9 40.1% 795