The Superfinal

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
CRoberson
Posts: 2006
Joined: Mon Mar 13, 2006 1:31 am
Location: North Carolina, USA
Contact:

Re: Book losses

Post by CRoberson » Mon May 26, 2014 5:21 pm

kranium wrote:Avoid all this mess...
let the engines play the openings!

We take away the endings via TBs...
We take away the openings..
Why?

These powerful engines are capable of playing the whole game and beating any human, no matter what opening

It's pretentious IMO that humans claim to know better than 3200 ELO engines

Some may say it will be boring...no way!
I think it would fascinating to see what openings lines they chose...
from my experience we'd see a good variety, Ruy Lopez, Queens Gambit, Sicilian, etc.

Some may say 'too many duplicate games',
but several LTC matches have been run with 100s of games and 0 duplicates

I say let the engines play!
That has chances. It would be a game changer for some. Some programs have little opening knowledge and more midgame and endgame knowledge. They don't encode the opening knowledge, because they can use a book and it speeds up the program. However, some of us have opening knowledge in the engines. We would gain some (at least temporary) advantage.

You know, the more I think about it, the more I like it. A no book tournament. Norman, I think this might be the best idea you have ever had.

CRoberson
Posts: 2006
Joined: Mon Mar 13, 2006 1:31 am
Location: North Carolina, USA
Contact:

Re: Book losses

Post by CRoberson » Mon May 26, 2014 5:43 pm

Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.

arjuntemurnikar
Posts: 204
Joined: Tue Oct 15, 2013 8:22 pm
Location: Singapore

Re: Book losses

Post by arjuntemurnikar » Mon May 26, 2014 5:52 pm

CRoberson wrote:
kranium wrote:Avoid all this mess...
let the engines play the openings!

We take away the endings via TBs...
We take away the openings..
Why?

These powerful engines are capable of playing the whole game and beating any human, no matter what opening

It's pretentious IMO that humans claim to know better than 3200 ELO engines

Some may say it will be boring...no way!
I think it would fascinating to see what openings lines they chose...
from my experience we'd see a good variety, Ruy Lopez, Queens Gambit, Sicilian, etc.

Some may say 'too many duplicate games',
but several LTC matches have been run with 100s of games and 0 duplicates

I say let the engines play!
That has chances. It would be a game changer for some. Some programs have little opening knowledge and more midgame and endgame knowledge. They don't encode the opening knowledge, because they can use a book and it speeds up the program. However, some of us have opening knowledge in the engines. We would gain some (at least temporary) advantage.

You know, the more I think about it, the more I like it. A no book tournament. Norman, I think this might be the best idea you have ever had.
Although this may be the fairest option, it doesn't bode well for a tournament that is being watched by spectators. There will be duplicate openings at least up till 6-10 ply... I have tried this before. Many engines have preferences in openings. For example SF will keep playing the Ruy Lopez and Komodo always starts with 1. d4. This will lead to lesser variety (though no duplicate full games because chess is just so vast).

A spectator driven tournament has slightly different objectives than say a private tournament. A spectator-driven tournament needs some sort of attractive quality to keep people hooked. If games are 90% boring draws or 90% d4 openings, people will get bored and stop watching. A private tournament that is not watched live can afford more boring draws in search of more fairness and more objective result.

A spectator driven tournament is event-oriented. A non-spectator-driven tournament is result-oriented. Once you understand this critical difference, you will realize that Nelson and co. did the best possible job with the openings (given limited resources).

CRoberson
Posts: 2006
Joined: Mon Mar 13, 2006 1:31 am
Location: North Carolina, USA
Contact:

Re: Book losses

Post by CRoberson » Mon May 26, 2014 6:10 pm

arjuntemurnikar wrote:
CRoberson wrote:
kranium wrote:Avoid all this mess...
let the engines play the openings!

We take away the endings via TBs...
We take away the openings..
Why?

These powerful engines are capable of playing the whole game and beating any human, no matter what opening

It's pretentious IMO that humans claim to know better than 3200 ELO engines

Some may say it will be boring...no way!
I think it would fascinating to see what openings lines they chose...
from my experience we'd see a good variety, Ruy Lopez, Queens Gambit, Sicilian, etc.

Some may say 'too many duplicate games',
but several LTC matches have been run with 100s of games and 0 duplicates

I say let the engines play!
That has chances. It would be a game changer for some. Some programs have little opening knowledge and more midgame and endgame knowledge. They don't encode the opening knowledge, because they can use a book and it speeds up the program. However, some of us have opening knowledge in the engines. We would gain some (at least temporary) advantage.

You know, the more I think about it, the more I like it. A no book tournament. Norman, I think this might be the best idea you have ever had.
Although this may be the fairest option, it doesn't bode well for a tournament that is being watched by spectators. There will be duplicate openings at least up till 6-10 ply... I have tried this before. Many engines have preferences in openings. For example SF will keep playing the Ruy Lopez and Komodo always starts with 1. d4. This will lead to lesser variety (though no duplicate full games because chess is just so vast).

A spectator driven tournament has slightly different objectives than say a private tournament. A spectator-driven tournament needs some sort of attractive quality to keep people hooked. If games are 90% boring draws or 90% d4 openings, people will get bored and stop watching. A private tournament that is not watched live can afford more boring draws in search of more fairness and more objective result.

A spectator driven tournament is event-oriented. A non-spectator-driven tournament is result-oriented. Once you understand this critical difference, you will realize that Nelson and co. did the best possible job with the openings (given limited resources).
Given their goal, I fully agree they did the best they could do with the resources at hand. So, make the books 1 move: you get variety and a the other problem is easier to solve.

User avatar
kranium
Posts: 1843
Joined: Thu May 29, 2008 8:43 am

Re: Book losses

Post by kranium » Mon May 26, 2014 6:14 pm

CRoberson wrote:Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.

I like this idea, but I would suggest no (ownbook) opening book...
(otherwise the best book makers would benefit)

Let the engines be 100% independent from human interference,
they must simply rely on PSTs, positonal algorithms, eval terms, etc. in the opening stages
(many engines do play differing 1st moves on occasion...)
I've seen top engines play 1.e4, 1. d4, 1.Nf3, etc.

A while back, for curiosity, I extracted the openings played in Rob Nichols My Humble Tournament (no books, no tablebases)
http://www.talkchess.com/forum/viewtopic.php?t=48468

http://www.chesslogik.com/images/Humble%20openings.jpg

there was a great variety

User avatar
michiguel
Posts: 6401
Joined: Thu Mar 09, 2006 7:30 pm
Location: Chicago, Illinois, USA
Contact:

Re: Book losses

Post by michiguel » Mon May 26, 2014 6:33 pm

CRoberson wrote:Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.
I would like that as long as learning ON is allowed.

Miguel

CRoberson
Posts: 2006
Joined: Mon Mar 13, 2006 1:31 am
Location: North Carolina, USA
Contact:

Re: Book losses

Post by CRoberson » Mon May 26, 2014 6:36 pm

michiguel wrote:
CRoberson wrote:Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.
I would like that as long as learning ON is allowed.

Miguel
There are two forms of learning:
1) you adjust the weights of the moves in the book
2) (1) + you can grow the book.

I assume you mean 1. I am good with that.

IGarcia
Posts: 543
Joined: Mon Jul 05, 2010 8:27 pm

Re: Book losses

Post by IGarcia » Mon May 26, 2014 7:05 pm

Nelson Hernandez wrote:I am the responsible party for the TCEC openings. What I would suggest is that you first read my notes on the openings in TCEC Help. That will provide insight into how I approached the problem of selecting hundreds of suitable openings given the parameters and priorities decided by Martin Thoresen.

Then you would do well to consider the limited resources we had in terms of time (two months) and CPU power (two old servers). Our job was not to over-promise and under-deliver but rather the reverse. In other words, given that we needed to complete the task in such a short period of time, we had to size the scope of the job so that we would deliver the best product we could in the time allotted. I think we succeeded.

Was every one of the hundreds of openings a success, in that it produced a fair and entertaining contest? I would say "no" immediately, while stating that a high percentage of TCEC attendees agree that the openings as a whole have been a success, even while they personally may have different ideas and approaches.

It has proved a particularly daunting task to coax decisive outcomes out of the Superfinal without employing biased starting positions given the strength of the finalists. Were we to employ optimal openings exclusively the draw-rate might well climb to 80-90% and if it did I would be criticized on those grounds. (What you may not realize is that I was hired for this position in the first place in order to bring down the draw-rate in a controlled and sensible manner.)

In any event there have been lessons learned from this Season. There are areas of possible improvement, and I hope to incorporate them next time around. But I am under no illusions that I will be able to please everyone!
Nelson:

The openings are fine and of course future improvement are welcome.

If there is some real serious mistake on a opening, then, because it will be played in inverse colors, it is compensated. The mayor problem will be with 2 very strong engines on each side a (1-0, 1-0) result pair (or the opposite).

I wonder why there is too much talk here about the openings, and when the author (you) come here nobody speaks to you, neither reply to you... a classical in this forum!

Ignacio.

User avatar
michiguel
Posts: 6401
Joined: Thu Mar 09, 2006 7:30 pm
Location: Chicago, Illinois, USA
Contact:

Re: Book losses

Post by michiguel » Mon May 26, 2014 7:14 pm

CRoberson wrote:
michiguel wrote:
CRoberson wrote:Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.
I would like that as long as learning ON is allowed.

Miguel
There are two forms of learning:
1) you adjust the weights of the moves in the book
2) (1) + you can grow the book.

I assume you mean 1. I am good with that.
I meant positional learning or any other adjustment of parameters. That would allow to fight against getting in the same losing line over and over again. Philosophically, if we tell the engines they cannot use "pre conceived knowledge" we have to allow them to build their own during the tournament.

Miguel

CRoberson
Posts: 2006
Joined: Mon Mar 13, 2006 1:31 am
Location: North Carolina, USA
Contact:

Re: Book losses

Post by CRoberson » Mon May 26, 2014 7:21 pm

michiguel wrote:
CRoberson wrote:
michiguel wrote:
CRoberson wrote:Ok, I suggest a 1 move book. Just for variety. Both white and black can have their first move booked. That way you can force one response to e4 or allow 3 response to e4. Also, white can limit his opening to d4 or allow e4, d4, or c4, but not Nf3, g3, b4 ...

Then the authors can supply their books.
I would like that as long as learning ON is allowed.

Miguel
There are two forms of learning:
1) you adjust the weights of the moves in the book
2) (1) + you can grow the book.

I assume you mean 1. I am good with that.
I meant positional learning or any other adjustment of parameters. That would allow to fight against getting in the same losing line over and over again. Philosophically, if we tell the engines they cannot use "pre conceived knowledge" we have to allow them to build their own during the tournament.

Miguel
You're suggesting allowing something like TDL (Temporal Difference Learning) to adjust the eval equations or maybe search parameters. That would be cool.

Post Reply