Page 1 of 3

Piece values

Posted: Wed Apr 18, 2018 12:08 pm
by soheil_hooshdaran
Hi.
What's the basis for assigning numerical values 1-3-5-9 to pieces?

Re: Piece values

Posted: Wed Apr 18, 2018 1:07 pm
by Roger de Coverly
soheil_hooshdaran wrote:
Wed Apr 18, 2018 12:08 pm
What's the basis for assigning numerical values 1-3-5-9 to pieces?
Lost in the mists of time I should imagine.

It seems to work as a broad rule of thumb for a basic evaluation. The earliest writings on chess theory were by Philidor and then later by Staunton. Did these authors propose theories of piece value?

Re: Piece values

Posted: Wed Apr 18, 2018 2:37 pm
by Nick Burrows

Re: Piece values

Posted: Wed Apr 18, 2018 3:08 pm
by Michael Farthing
Roger de Coverly wrote:
Wed Apr 18, 2018 1:07 pm
soheil_hooshdaran wrote:
Wed Apr 18, 2018 12:08 pm
What's the basis for assigning numerical values 1-3-5-9 to pieces?
Lost in the mists of time I should imagine.

It seems to work as a broad rule of thumb for a basic evaluation. The earliest writings on chess theory were by Philidor and then later by Staunton. Did these authors propose theories of piece value?
Staunton: The Chess Players Handbook Chapter 4

Code: Select all

Pawn    1.00
Knight  3.05
Bishop  3.50
Rook    5.48
Queen   9.94

Re: Piece values

Posted: Wed Apr 18, 2018 7:41 pm
by soheil_hooshdaran
"Chess:from beginner to GM" gives it the following way:
1-3-4-8

Re: Piece values

Posted: Wed Mar 13, 2019 6:17 pm
by Francis Fields
There use to be a chess website out there that I can no longer find. Someone had analysed games with material imbalances and calculated the following:

pawn 1.00, knight 3.25, bishop 3.5, rook 5.0, queen 9.75.

I have seen an old chess book that says bishop plus one and a half pawns equals a rook. There is another chess book that says queen is worth 9 or 10 points.

Re: Piece values

Posted: Fri Mar 22, 2019 8:56 am
by Mark Ashley
I was listening to the perpetual podcast with Matthew Sadler and Natasha Regan as guests which included a discussion about Alpha Zero. It was interesting to hear that AZ wasnt programmed with a numerical value for the pieces, raising the question as to whether or how that affected the way it played the game.

Re: Piece values

Posted: Fri Mar 22, 2019 9:13 am
by Roger de Coverly
Mark Ashley wrote:
Fri Mar 22, 2019 8:56 am
It was interesting to hear that AZ wasnt programmed with a numerical value for the pieces, raising the question as to whether or how that affected the way it played the game.
It wasn't programmed with any rules at all. What it came up with is entirely self taught. It doesn't evaluate by notional pawn values either, giving assessments as percentage winning chances.

Actually a human could play with no piece values. You observe that at the start of the game, both sides have eight pawns, one queen, two rooks, two knights, a light squared bishop and a dark squared bishop. Whether or not notional values are assigned, the material is equal until you get unbalanced exchanges, even a bishop for a knight. Material values are just a first approximation to evaluating the relative worth of such exchanges.

Re: Piece values

Posted: Wed Jun 26, 2019 4:36 pm
by Francis Fields
A bishop is worth more than a knight in most positions and used to be called the minor exchange.

Re: Piece values

Posted: Thu Jun 27, 2019 11:10 am
by Joey Stewart
Mark Ashley wrote:
Fri Mar 22, 2019 8:56 am
I was listening to the perpetual podcast with Matthew Sadler and Natasha Regan as guests which included a discussion about Alpha Zero. It was interesting to hear that AZ wasnt programmed with a numerical value for the pieces, raising the question as to whether or how that affected the way it played the game.
If only it could actually give back some of its newfound learnings about how the game should be played then it might be worth something more than just another case of "oh look how much better computers are than humans.... at a very specific function... performed on the best hardware money can buy"

Re: Piece values

Posted: Thu Jun 27, 2019 11:54 am
by NickFaulks
Joey Stewart wrote:
Thu Jun 27, 2019 11:10 am
If only it could actually give back some of its newfound learnings about how the game should be played
Isn't that precisely what the interpretative work of Sadler and Regan is doing?

Re: Piece values

Posted: Thu Jun 27, 2019 2:35 pm
by MartinCarpenter
They're doing high level interpretations of the results, which are obviously rather useful! All you can do anyway :)

There's absolutely no way to go back and work what internal representations Alpha Zero has created to use for things like piece values, king safety etc. That's just how neural nets are. A little bit like trying to work out how your own subconcious works.

They are trying to work out ways to produce neural nets you can interrogate but that's a good way off.

Re: Piece values

Posted: Thu Jun 27, 2019 7:18 pm
by John McKenna
We all know that the interrogation will produce the following answer -

8 (pawns) + 6 (2 knights) + 7 (2 bishops) + 10 (2 rooks) + 9 (queen) + 2 (king) = 42!

Re: Piece values

Posted: Thu Jun 27, 2019 9:40 pm
by Kevin Thurlow
"42"

And there may sadly be many who do not realise the significance of that...

Re: Piece values

Posted: Fri Jun 28, 2019 9:57 am
by John McKenna
Yes indeed, pity those who don't know it's the answer to "life, the universe and everything" (including, of course, why the chess king is to be valued at 2 notional pawns!)

https://www.independent.co.uk/news/yes- ... 51201.html