TalkChess.com

In today's search engines, how many extra ISO rating points do today's opening books buy you? I.e. the same engine with vs. without the OB. Does the benefit decrease as the search engine itself becomes more powerful, say with a faster processor?

I think it is significant. But book vs. no book is probably not a good way to measure it. Without books engines frequently depart from known theory early in the game.

The real advantage shows when a poor or shallow book is matched against a very good book. The engine with the weaker book will play into inferior lines and get busted.

Sedat Canbaz ran a series of tournaments just to test opening books fairly recently (https://sites.google.com/site/computers ... ournaments). There were significant differences depending on the book used.

--Jon

Thx for the link. In Sedat's tournament https://sites.google.com/site/computerschess/scct-cs-6,
19 different books are compared all using a fixed engine, namely Stockfish 7 beta 1 x64.

One of the books is "No Book", which is the engine playing without a book.
It finished 6th out of 19, with a score of 3421.4. Number 1 was "SCCT Champions at 3438.0. 19th place score was 3339.2.

I am astonished that No Book would perform so well, finishing ahead of 13 out of 19 books. Or am I misinterpreting something?

jack512 wrote:Thx for the link. In Sedat's tournament https://sites.google.com/site/computerschess/scct-cs-6,
19 different books are compared all using a fixed engine, namely Stockfish 7 beta 1 x64.

One of the books is "No Book", which is the engine playing without a book.
It finished 6th out of 19, with a score of 3421.4. Number 1 was "SCCT Champions at 3438.0. 19th place score was 3339.2.

I am astonished that No Book would perform so well, finishing ahead of 13 out of 19 books. Or am I misinterpreting something?

The test you refer to was only for books with max. 8 moves.

some time ago , the italian g-sei site has done a BOOK TOURNAMENT. The used engine was Stockfish and the result of the first tournament was similar to the result achieved by sedat, NO-BOOK is better than a not good book.

full result ( in italian) ca be found at http://www.g-sei.org/il-primo-torneo-de ... aperture2/

reported below the result of the tournament:
01 RpC-Needle 17.0/24
02 Fauzi 3.4 15.0/24
03 None 14.5/24
04 Diffusion 1.0.5A2 12.5/24
05 NathanaelRussell 2014 12.0/24
06 Aaricia 2012 12.0/24
07 Perfect 2014t 12.0/24
08 Performance 2005 11.5/24
09 SS-ForChiron 1.5 11.5/24
10 Superbook 0.2 11.0/24
11 HS-Grandbook 1.8 10.0/24
12 Ref2550 Lite 8.5/24

I suppose the explanation must be the mismatch between the goals of the player whose games went into the opening book, versus what the engine is able to see of its own goals when it comes out of the book.

jack512 wrote:In Sedat's tournament https://sites.google.com/site/computerschess/scct-cs-6,
19 different books are compared all using a fixed engine, namely Stockfish 7 beta 1 x64.

One of the books is "No Book", which is the engine playing without a book.
It finished 6th out of 19, with a score of 3421.4. Number 1 was "SCCT Champions at 3438.0. 19th place score was 3339.2.

I am astonished that No Book would perform so well, finishing ahead of 13 out of 19 books. Or am I misinterpreting something?

With so few games these numbers don't mean much. E.g., the difference between #1 and #6 is a mere 17 Elo. To have a decently high LOS in this case you'd need somewhere between 1000 and 2000 games (Kai or Ferd would give you an exact minimum); only 50 were played. The Italian tmt had even fewer games and the results there are even harder to take seriously.

These tournaments are mostly done for the sake of entertainment, methinks.

jack512 wrote:Thx for the link. In Sedat's tournament https://sites.google.com/site/computerschess/scct-cs-6,
19 different books are compared all using a fixed engine, namely Stockfish 7 beta 1 x64.

One of the books is "No Book", which is the engine playing without a book.
It finished 6th out of 19, with a score of 3421.4. Number 1 was "SCCT Champions at 3438.0. 19th place score was 3339.2.

I am astonished that No Book would perform so well, finishing ahead of 13 out of 19 books. Or am I misinterpreting something?

"the number of games..."

Ed got over 100 Elo benefit in a test.

In CSS forum there was really stunning result with Perfect2015: