TCEC 9 super final: the opening selection

[Jeroen]

More information about the opening selection of the TCEC 9 super final can be found here:

http://blogchess2016.blogspot.nl/2016/1 ... udini.html

Enjoy the games!

[Dirt]

The openings have been very successful so far. Congratulations!

[Guenther]

More information about the opening selection of the TCEC 9 super final can be found here:

http://blogchess2016.blogspot.nl/2016/1 ... udini.html

Enjoy the games!

It seems to me the process of selecting openings for such a match becomes more and more a heracles task. The programs become stronger and stronger and an urgent question at least to me is, how can one avoid the slightest bias here?
I mean it is likely that A gives a different eval than B and plays opening X better/worse than opening Y. Moreover there are only a handful of programs in the same league as SF8 and H5, which could be considered for analysing those positions.

A theoretical example: Let's say we have a position which shows at the end +0.55 for SF8 and +0.30 H5, shouldn't we then select another one with exchanged scores? (This even neglects that the meaning of the score
is probably not the same for both - at least SF in the past always showed higher absolutely evals in the opening/early middlegame.)

What were the limits you set for the final positions score and was it required that both or more programs had to agree to this limit and how far into the future analysis went (normally) after your final position?

Of course I will accept if you want to keep your secrets for this topic.

Edit: What draw percentage was your expection before the start? :)

[pohl4711]

More information about the opening selection of the TCEC 9 super final can be found here:

http://blogchess2016.blogspot.nl/2016/1 ... udini.html

Enjoy the games!

Your openings for the 9.SuperFinal are really great!
There can be no doubt about, that if only top-engines play against each other, short and balanced openings leads the modern computerchess to "draw-death" and make watching it more and more boring.
So, your conclusion and solution was absolutely right.
I created my SALC openingbook and my SALC Testset (opening-lines and positions where white and black castled to opposite sides + both queens still on the board) with the same intention...but SALC positions and lines can be found automatically in big databases, which made it possible to filter 22800 games out of 1.200.000 human games and create an opening-book.

Everybody can download it here:
http://spcc.beepworld.de/downloads--links.htm

[MikeB]

....opening-lines and positions where white and black castled to opposite sides + both queens still on the board...

Nice approach and thanks for the link...

[Lyudmil Tsvetkov]

I appreciate everyone's work and, if people like it, it is perfectly fine with me. Just 2 short notices:

1) Imagine a boxing match with 12 rounds, where the first 4 or 5 rounds would be skipped, the match starting from the 6th round. Would that be more interesting and attract bigger audience? Certainly not, quite the opposite. It is the same with a book 18-moves long. Let's suppose, for the sake of argument, that an average TCEC game will last 60 moves. You are actually skipping more than 1/4 of the game, but, as the opening stage is the most interesting part, and where the engines make the most mistakes, you are actually skipping a bigger part, starting from round 5th.

2) Is is obvious that a shorter book means a longer game and more opportunities for one of the engines to make a mistake, and therefore for a winning outcome. I guess your 100-games run was an outlier, or something of the kind, statistically related. In my bullet matches between SF 7 and Komodo 10, about the same strength at this TC, I get winning percentage way above 50% with a short neutral 2-move book, and a winning percentage way below 50% with a longer book, 8-moves and longer. And I have run many 100 games matches of this sort, results have always been consistent: the shorter book always boosts the winning percentage.

Now, I do not in any way say your book is bad, not at all, the book has some very interesting lines, keeping more pieces on the board and relatively closed center is a good decision certainly, but why not let the engines play it all from the beginning?

[corres]

[quote="Jeroen"]More information about the opening selection of the TCEC 9 super final can be found here:

[url]http://blogchess2016.blogspot.nl/2016/1 ... udini.html[/url]

Enjoy the games![/quote]

These sound good.
But where do I found your opening positions?
After the beginning of the Final there is no sense to make this mystery.

[corres]

[quote="Lyudmil Tsvetkov"]
I appreciate everyone's work and, if people like it, it is perfectly fine with me. Just 2 short notices:

1) Imagine a boxing match with 12 rounds, where the first 4 or 5 rounds would be skipped, the match starting from the 6th round. Would that be more interesting and attract bigger audience? Certainly not, quite the opposite. It is the same with a book 18-moves long. Let's suppose, for the sake of argument, that an average TCEC game will last 60 moves. You are actually skipping more than 1/4 of the game, but, as the opening stage is the most interesting part, and where the engines make the most mistakes, you are actually skipping a bigger part, starting from round 5th.

2) Is is obvious that a shorter book means a longer game and more opportunities for one of the engines to make a mistake, and therefore for a winning outcome. I guess your 100-games run was an outlier, or something of the kind, statistically related. In my bullet matches between SF 7 and Komodo 10, about the same strength at this TC, I get winning percentage way above 50% with a short neutral 2-move book, and a winning percentage way below 50% with a longer book, 8-moves and longer. And I have run many 100 games matches of this sort, results have always been consistent: the shorter book always boosts the winning percentage.

Now, I do not in any way say your book is bad, not at all, the book has some very interesting lines, keeping more pieces on the board and relatively closed center is a good decision certainly, but why not let the engines play it all from the beginning?
[/quote]

Fundamentally, the chess engines maid for playing middle game.
To play openings the openings book are used. As to play endgames the endgame databases are used.
The draw is the natural outcome of a correct chess game.
Among strong engines many draws is a natural phenomena, I think.
Few draw may relate to the too big deference in the power of the participants or to a "biased" opening book or positions.

[Kohflote]

Hi,
May I asked what are the 50 opening lines?

Thank you & regards,
Koh, Kah Huat

[Lyudmil Tsvetkov]

Just a quick notice: from 13 opening positions so far played at TCEC, only 4 have evals for both engines on the first move after the opening in the range of (-40cps) - (+40cps) {arguably the losing boundary in chess for the mg}

so, 2/3 of the positions so far are more or less above or close to the losing edge, and in some cases both engines show scores close and above 60-70cps.

So far the openings played have hugely favoured SF (many open games, many Sicilians, many openings with specific threats, requiring depper search, etc.), hence the current score.

I have to admit the Jeroen made a great job in picking up intriguing openings, full of tension, quite trendy at that, but, and that is a but with capital letters, I do not think that starting the game with a lost position, no matter the alternation of colours, is a good/acceptable thing to do, bearing in mind the strength of modern top engines and their aspiration to go after the ultimate truth.