Are there rating lists that test material handicap games?

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.
Post Reply
Uri Blass
Posts: 8530
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Are there rating lists that test material handicap games?

Post by Uri Blass » Fri Aug 16, 2019 5:46 pm

I think that it may be interesting to have the following players in the rating lists

1)top engine without a knight(half of the games without knight g1 and half of the games without knight b1)
2)top engine without a rook
3)top engine without a queen

What is possible to achieve in the CCRL rating list in computer rating lists?
What is your opinion?

I guess the following CCRL numbers are possible in the 40/40 rating list:
1)2800 ccrl rating without a knight
2)2500 ccrl rating without a rook
3)2200 ccrl rating without a queen

Of course playing against humans is a different story and no rating that is close to these rating is possible against humans.

Note that possible does not mean that engines can achieve it today and I do not think that engines today play perfectly for these conditions.

Marcus9
Posts: 21
Joined: Mon May 13, 2019 4:05 pm
Full name: Marco Giorgio

Re: Are there rating lists that test material handicap games?

Post by Marcus9 » Fri Aug 16, 2019 8:46 pm

It would be a beautiful thing, But I think you exaggerated the handicap, I think it is better to evaluate pawn (f7) and knight (b1 or b8)

Uri Blass
Posts: 8530
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Are there rating lists that test material handicap games?

Post by Uri Blass » Fri Aug 16, 2019 10:01 pm

I do not think I exaggerated the handicap.

With pawn handicap chess engines also beat GM's and the idea is to expose the weakness of chess engines relative to humans.
I guess stockfish10 without a pawn f7 is already going to get clearly more than 3000 elo in the CCRL rating list.

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Are there rating lists that test material handicap games?

Post by lkaufman » Sat Aug 17, 2019 5:29 am

Uri Blass wrote:
Fri Aug 16, 2019 5:46 pm
I think that it may be interesting to have the following players in the rating lists

1)top engine without a knight(half of the games without knight g1 and half of the games without knight b1)
2)top engine without a rook
3)top engine without a queen

What is possible to achieve in the CCRL rating list in computer rating lists?
What is your opinion?

I guess the following CCRL numbers are possible in the 40/40 rating list:
1)2800 ccrl rating without a knight
2)2500 ccrl rating without a rook
3)2200 ccrl rating without a queen

Of course playing against humans is a different story and no rating that is close to these rating is possible against humans.

Note that possible does not mean that engines can achieve it today and I do not think that engines today play perfectly for these conditions.
"OLD" Lc0 11248 on a 2080 would almost certainly get the highest 40/40 engine rating of current engines giving knight or perhaps even rook odds. I don't think any engine can give queen odds to any engine remotely close to 2200, engines just have no clue how to make a game of it down a queen, mostly because there isn't any way to do so. Unfortunately newer Lc0 engines are hopeless at giving handicaps, due to training with resign on, so this rather spoils the idea for now.
Komodo rules!

Uri Blass
Posts: 8530
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Are there rating lists that test material handicap games?

Post by Uri Blass » Sat Aug 17, 2019 8:08 am

I think that with a queen handicap against weak engines it may be possible to close the position with pablo's style and hopefully even to win the game because the opponent may sacrifice material in order to avoid the 50 move rule.

User avatar
Ovyron
Posts: 2339
Joined: Tue Jul 03, 2007 2:30 am

Re: Are there rating lists that test material handicap games?

Post by Ovyron » Sat Aug 17, 2019 10:24 am

Something to try:

Random mover with extra queen against normal engine without a queen. If normal engine wins all the games then it means there's a series of playing algorithms that cover all possible elos from this to the point an algorithm with the extra queen can win every game.
Make someone happy today.

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Are there rating lists that test material handicap games?

Post by lkaufman » Mon Aug 19, 2019 1:58 am

Ovyron wrote:
Sat Aug 17, 2019 10:24 am
Something to try:

Random mover with extra queen against normal engine without a queen. If normal engine wins all the games then it means there's a series of playing algorithms that cover all possible elos from this to the point an algorithm with the extra queen can win every game.
It takes roughly a 1000 elo for a human to be competitive with me at queen odds, and it would take an engine quite a bit more than that. I was about even with MacHack giving it queen odds over half a century ago when it was rated about 1530. It was vastly stronger than random moves. I think a random mover would need something like queen, two rooks, and three or even all four minors handicap to beat a GM or a good engine.
Komodo rules!

Post Reply