how will Leela fare at the WCCC?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
peter
Posts: 1628
Joined: Sat Feb 16, 2008 6:38 am

Re: how will Leela fare at the WCCC?

Post by peter » Fri Jul 13, 2018 10:39 am

whereagles wrote:
Fri Jul 13, 2018 8:34 am
It's the first real test of NN outside the Tyrel... errr... Google Corporation.

Looking forward to it. My guess is top 4.
That sounds realistic to me too, even if I wouldn't expect more than one of the for sure stronger opponents komodo, Jonny, Gingko and Shredder behind LC0, so nr. 5 ranking would be my tip, if there were games enough and all participants had to use same short indifferent book.

Not even Harvey Williamson would manage to write an own Anti- Leela book in that short time against a hardly known hardware- software entity, and Erdogan Günes and Larry Kaufman don't have to.

Of course they all could use sharp gambit lines (e.g. Jeroen Noomen's ones) but that could hurt against the other engines except Leela.
May books be changed from game to game, means opponent to opponent?
:)
Peter Martan

User avatar
Laskos
Posts: 8442
Joined: Wed Jul 26, 2006 8:21 pm

Re: how will Leela fare at the WCCC?

Post by Laskos » Fri Jul 13, 2018 12:00 pm

peter wrote:
Fri Jul 13, 2018 10:39 am
whereagles wrote:
Fri Jul 13, 2018 8:34 am
It's the first real test of NN outside the Tyrel... errr... Google Corporation.

Looking forward to it. My guess is top 4.
That sounds realistic to me too, even if I wouldn't expect more than one of the for sure stronger opponents komodo, Jonny, Gingko and Shredder behind LC0, so nr. 5 ranking would be my tip, if there were games enough and all participants had to use same short indifferent book.

Not even Harvey Williamson would manage to write an own Anti- Leela book in that short time against a hardly known hardware- software entity, and Erdogan Günes and Larry Kaufman don't have to.

Of course they all could use sharp gambit lines (e.g. Jeroen Noomen's ones) but that could hurt against the other engines except Leela.
May books be changed from game to game, means opponent to opponent?
:)
I am not sure, but I think books can be changed. I remember some long time ago, Junior playing using books adapted for opponent, in order to avoid draws against weaker engines. Their book preparation was one of the reasons they won so many titles back then. I think an experienced book builder can have an anti-Leela book in one day, if he has the info that on fairly easy tactical puzzles Leela is only about 2000 Elo points of AB engines, and positionally Leela is above 3500.

yanquis1972
Posts: 1682
Joined: Tue Jun 02, 2009 10:14 pm

Re: how will Leela fare at the WCCC?

Post by yanquis1972 » Fri Jul 13, 2018 1:00 pm

i would also guess an anti-leela book would be fairly easy. i hadn't even considered this. there's nothing preventing leela from using a book, though, which exemplifies the silliness here; next year it'll surely be a battle of cooked books (assuming leela-team is going no-book, or didn't have the time to create an anti-anti-leela book).

changing from game to game should be done away with... there are so few opponents an engine can, as kai says, essentially win before it's made it's first calculation. i get that it's supposed to be unrestricted, but engines are at such a strength now that having a non-default book -- ideally a provided & minimal-ply one (whatever that means; 2, 8, 16, anywhere between) - certainly feels outdated....TCEC does it better.

User avatar
CMCanavessi
Posts: 761
Joined: Thu Dec 28, 2017 3:06 pm
Location: Argentina

Re: how will Leela fare at the WCCC?

Post by CMCanavessi » Fri Jul 13, 2018 1:21 pm

Given that there are so few games, and that you only play each other engine just once, I would like the idea for WCCC to go completely bookless. Difficult to control though.
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls

jdart
Posts: 3618
Joined: Fri Mar 10, 2006 4:23 am
Location: http://www.arasanchess.org

Re: how will Leela fare at the WCCC?

Post by jdart » Fri Jul 13, 2018 4:14 pm

Personally I find book-enabled tournaments more interesting.

I think you can consider the book part of an overall chessplaying system. It is not trivial at all to construct a good book. And, playing bookless, many engines make errors right away in the opening, sometimes serious ones that are known by prior opening theory to be bad.

--Jon

Werewolf
Posts: 1081
Joined: Thu Sep 18, 2008 8:24 pm

Re: how will Leela fare at the WCCC?

Post by Werewolf » Fri Jul 13, 2018 4:34 pm

jdart wrote:
Fri Jul 13, 2018 4:14 pm
Personally I find book-enabled tournaments more interesting.

I think you can consider the book part of an overall chessplaying system. It is not trivial at all to construct a good book. And, playing bookless, many engines make errors right away in the opening, sometimes serious ones that are known by prior opening theory to be bad.

--Jon
Yes I agree with that. For Leela or Alpha Zero (although this would be a big handicap) it might be good for them to be bookless to stay with the whole "zero" human contribution thing. Totally unfair, but maybe when it's stronger...

carldaman
Posts: 1587
Joined: Sat Jun 02, 2012 12:13 am

Re: how will Leela fare at the WCCC?

Post by carldaman » Fri Jul 13, 2018 4:46 pm

Maybe the engine's (own) book should be capped at no more than 15 moves, otherwise you could potentially have games where 30-40 moves are reeled off directly from book, so the engine basically does not play at all, but only the book does.

JJJ
Posts: 1260
Joined: Sat Apr 19, 2014 11:47 am

Re: how will Leela fare at the WCCC?

Post by JJJ » Fri Jul 13, 2018 4:52 pm

jdart wrote:
Fri Jul 13, 2018 4:14 pm
Personally I find book-enabled tournaments more interesting.

I think you can consider the book part of an overall chessplaying system. It is not trivial at all to construct a good book. And, playing bookless, many engines make errors right away in the opening, sometimes serious ones that are known by prior opening theory to be bad.

--Jon

Most top engines doesn't.

User avatar
CMCanavessi
Posts: 761
Joined: Thu Dec 28, 2017 3:06 pm
Location: Argentina

Re: how will Leela fare at the WCCC?

Post by CMCanavessi » Fri Jul 13, 2018 5:13 pm

jdart wrote:
Fri Jul 13, 2018 4:14 pm
Personally I find book-enabled tournaments more interesting.

I think you can consider the book part of an overall chessplaying system. It is not trivial at all to construct a good book. And, playing bookless, many engines make errors right away in the opening, sometimes serious ones that are known by prior opening theory to be bad.

--Jon
That's true, but how do you make sure that the book was made from the engine itself? You could have the Komodo team bring a book that was made from Houdini, SF and Komodo... and that would be pointless imho. The idea is to have the strongest engine, not the strongest book made from god knows what.
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls

peter
Posts: 1628
Joined: Sat Feb 16, 2008 6:38 am

Re: how will Leela fare at the WCCC?

Post by peter » Fri Jul 13, 2018 5:34 pm

Laskos wrote:
Fri Jul 13, 2018 12:00 pm
I am not sure, but I think books can be changed. I remember some long time ago, Junior playing using books adapted for opponent, in order to avoid draws against weaker engines. Their book preparation was one of the reasons they won so many titles back then. I think an experienced book builder can have an anti-Leela book in one day, if he has the info that on fairly easy tactical puzzles Leela is only about 2000 Elo points of AB engines, and positionally Leela is above 3500.
If you're right about the book to be changed at each game and against each opponent, then an Anti-Leela wouldn*t be a problem at all of course.
Peter Martan

Post Reply