The influence of the length of openings

[lucasart]

What would be more interesting, rather than vary the engines and fix the book conitions, is to fix the engine and vary the book conditions.

So you choose only one engine (might as well take the strongest one H3). And you play H3+bookdepth=4 against H3+bookdepth=30.

Unless the book is of prime quality, and every move in it has been checked (which is rare), I wouldn't be surprised that the bookdepth=30 version performs worse.

The problem is when you play bookdepth=30 against bookdepth=30, the book lines are complaisant and play along to reach the position aimed by the book author. But in reality, along the 30 move path, there may be quite a few suboptimal moves that would be sanctionned by the opponent that is out of book.

But there would be a big time biais, so you musn't use a fisher clock or a tournament clock. The only thing you can use a fixed time or fixed nb of nodes per move. Otherwise the one coming out of book after 30 moves has his clock full (+30 x increment if any) while the other one has already consumed half of it or more.

[michiguel]

What would be more interesting, rather than vary the engines and fix the book conitions, is to fix the engine and vary the book conditions.

So you choose only one engine (might as well take the strongest one H3). And you play H3+bookdepth=4 against H3+bookdepth=30.

Unless the book is of prime quality, and every move in it has been checked (which is rare), I wouldn't be surprised that the bookdepth=30 version performs worse.

The problem is when you play bookdepth=30 against bookdepth=30, the book lines are complaisant and play along to reach the position aimed by the book author. But in reality, along the 30 move path, there may be quite a few suboptimal moves that would be sanctionned by the opponent that is out of book.

But there would be a big time biais, so you musn't use a fisher clock or a tournament clock. The only thing you can use a fixed time or fixed nb of nodes per move. Otherwise the one coming out of book after 30 moves has his clock full (+30 x increment if any) while the other one has already consumed half of it or more.

But that is one of points of having a book, saving time.

Miguel

[Laskos]

What would be more interesting, rather than vary the engines and fix the book conitions, is to fix the engine and vary the book conditions.

So you choose only one engine (might as well take the strongest one H3). And you play H3+bookdepth=4 against H3+bookdepth=30.

Unless the book is of prime quality, and every move in it has been checked (which is rare), I wouldn't be surprised that the bookdepth=30 version performs worse.

The problem is when you play bookdepth=30 against bookdepth=30, the book lines are complaisant and play along to reach the position aimed by the book author. But in reality, along the 30 move path, there may be quite a few suboptimal moves that would be sanctionned by the opponent that is out of book.

But there would be a big time biais, so you musn't use a fisher clock or a tournament clock. The only thing you can use a fixed time or fixed nb of nodes per move. Otherwise the one coming out of book after 30 moves has his clock full (+30 x increment if any) while the other one has already consumed half of it or more.

I used gm2600.bin with H3 at 30 plies and 4 plies, at fixed time per move 100ms (st=0.1 in cutechess-cli)

Code: Select all

    Program                            Score     %     Elo    +   -    Draws

  1 H3 30 plies                  : 1050.5/1999  52.6    9    10  10   53.7 %
  2 H3  4 plies                  :  948.5/1999  47.4   -9    10  10   53.7 %

18 +/- 10 points (2SD) advantage for 30 plies book. It seems that the book is still beneficial at this short TC, and I guess gm2600.bin is not the best book around, as it's built from gm2600.pgn of games of human GMs, which are prone to errors. I don't know how polyglot books work, but it seems that book lines which lead to advantageous positions compensate for weaker moves of GMs.

[IWB]

I used gm2600.bin with H3 at 30 plies and 4 plies, at fixed time per move 100ms (st=0.1 in cutechess-cli)

Code: Select all

    Program                            Score     %     Elo    +   -    Draws

  1 H3 30 plies                  : 1050.5/1999  52.6    9    10  10   53.7 %
  2 H3  4 plies                  :  948.5/1999  47.4   -9    10  10   53.7 %

18 +/- 10 points (2SD) advantage for 30 plies book. It seems that the book is still beneficial at this short TC, and I guess gm2600.bin is not the best book around, as it's built from gm2600.pgn of games of human GMs, which are prone to errors. I don't know how polyglot books work, but it seems that book lines which lead to advantageous positions compensate for weaker moves of GMs.

... Fixed time per move ... I over read that. Sorry. Forget my post.

Bye
Ingo

[Laskos]

I used gm2600.bin with H3 at 30 plies and 4 plies, at fixed time per move 100ms (st=0.1 in cutechess-cli)

Code: Select all

    Program                            Score     %     Elo    +   -    Draws

  1 H3 30 plies                  : 1050.5/1999  52.6    9    10  10   53.7 %
  2 H3  4 plies                  :  948.5/1999  47.4   -9    10  10   53.7 %

18 +/- 10 points (2SD) advantage for 30 plies book. It seems that the book is still beneficial at this short TC, and I guess gm2600.bin is not the best book around, as it's built from gm2600.pgn of games of human GMs, which are prone to errors. I don't know how polyglot books work, but it seems that book lines which lead to advantageous positions compensate for weaker moves of GMs.

With performance.bin at 500ms fixed time per move (5 times larger TC) I got

Code: Select all

    Program                             Score     %      Elo    +   -    Draws

  1 Houdini 3  30 plies            : 559.5/1049  53.3     12   13  13   63.8 %
  2 Houdini 3   4 plies            : 489.5/1049  46.7    -12   13  13   63.8 %

24 +/- 13 Elo points advantage for 30 ply deep book against 4 ply book with H3, at this longer TC. It seems that engines don't play openings very well, if even with bad moves in the book, the book prevails. And from the first post, all 4 engines play similarly badly the openings.

[ThatsIt]

[...snip...]
An engine that I really hated for that was Spike 1.2: you cannot disable the book and it's hardcoded inside the executable!
[...snip...]

There is a file named Spike.cnfg in the engine folder of Spike 1.2.
Open it and find:
<bibliothek enabled="yes" filename=/>

change into:
<bibliothek enabled="no" filename="dummy.pgn"/>

You have to create "dummy.pgn" as an empty (text)file in Spikes folder.

Best wishes,
G.S.

[hgm]

Otherwise the one coming out of book after 30 moves has his clock full (+30 x increment if any) while the other one has already consumed half of it or more.

But when you play a d=4 book against a d=30 book, the engine using the d=30 book will not come out of book after 30 moves at all. Typically it will come out of book after 5 or 6 moves. A d=30 book does not cover a full perft(30) set of variations.

[Laskos]

What would be more interesting, rather than vary the engines and fix the book conitions, is to fix the engine and vary the book conditions.

So you choose only one engine (might as well take the strongest one H3). And you play H3+bookdepth=4 against H3+bookdepth=30.

Unless the book is of prime quality, and every move in it has been checked (which is rare), I wouldn't be surprised that the bookdepth=30 version performs worse.

The problem is when you play bookdepth=30 against bookdepth=30, the book lines are complaisant and play along to reach the position aimed by the book author. But in reality, along the 30 move path, there may be quite a few suboptimal moves that would be sanctionned by the opponent that is out of book.

But there would be a big time biais, so you musn't use a fisher clock or a tournament clock. The only thing you can use a fixed time or fixed nb of nodes per move. Otherwise the one coming out of book after 30 moves has his clock full (+30 x increment if any) while the other one has already consumed half of it or more.

And here is what damage can provoke a good opening book (RpC to 30 plies) to 4 plies H3 opening average book. Engines still do not play openings well, and the opening book is still a significant component in their overall strength. TC is 500ms/move, so the engine with longer opening doesn't get advantage.

Code: Select all

    Program                            Score        %      Elo    +   -    Draws

  1 H3 RpC-SF-1.4.bin 30 plies      : 1603.5/2924  54.8     17    9   9   50.6 %
  2 H3 Performance.bin 4 plies      : 1320.5/2924  45.2    -17    9   9   50.6 %

34 +/- 9 points advantage with a good book.