TalkChess.com

I was wondering what your experiences were with these:

Rook pair penalty (small penalty for having two rooks)
Knight pair penalty (small penalty for having two knights)
Having a pawn bonus (small bonus for having at least one pawn)

I see several program use these and was wondering if anyone had tested how good they might be?

Mark

I have seen some benefit from:

1) Bonus for having at least one orthogonal mover.
2) Bonus for having at least one diagonal mover.
3) Bonus for having at least one pawn on each flank.

In empirical determination of material value of piece combinations, I have never seen any evidence for any such pair penalties.

Apart from the Bishop pair bonus, the only significant non-additive term I could identify was the elephantiasis correction, which reduces the value of super-pieces (like Q) in proportion to opponent lower-valued material.

BTW, Knight-pair is easy to test even with normal Chess engines: Just delete all B and N from the opening setup, and put on the evacuated squares: 1) B for one side and N for the other or 2) B+N for one side and N+N for the other. Play a few hundred games from each of these setups, to see if adding the Knight (creating a pair for only one side) changes the odds.

If the engine does not sufficiently randomize, shuffle the back-rank pieces, so that each game starts from a different setup. Or use many different engines to play the same imblanced position. (Each in selfplay.)

I like the idea, but wouldn't also invoke the bishop pair bonus/piece difference and/or Bishop/knight pair (mating material) for one side?

Because there is never more than 1 Bishop, the B-pair bonus cannot be involved.

I seriously doubt there exists somethng like a measurable BN bonus. In all games my engines ever played, I have never seen even a single one end in KBNK. Mating potential hardly contributes to piece value, also for individual pieces. E.g. a curtailed Rook limited to a maximum of 2 steps in any direction has almost exactly the same opening value as a similarly limited Bishop, despite the fact that the former can mate, and the latter not, against a bare King.

mjlef wrote:I was wondering what your experiences were with these:

Rook pair penalty (small penalty for having two rooks)
Knight pair penalty (small penalty for having two knights)
Having a pawn bonus (small bonus for having at least one pawn)

I see several program use these and was wondering if anyone had tested how good they might be?

Mark

I have had zero success with any of those, except for trying to keep a pawn or more. This can be pretty important as going to KRN vs KR from a KRNP vs KRP leaves you with no winning chances. But the other things (pair of rooks, etc) have simply not produced anything useful in my testing. They don't hurt, but they also do not help at all for me.

mjlef wrote:I was wondering what your experiences were with these:

Rook pair penalty (small penalty for having two rooks)
Knight pair penalty (small penalty for having two knights)
Having a pawn bonus (small bonus for having at least one pawn)

I see several program use these and was wondering if anyone had tested how good they might be?

Mark

I have tested a number of these types of evaluation and I never liked the result. Positions quickly degenerate to endgames.

The same goes for material balance evaluation. They tend to result in quick equal exchanges.

mjlef wrote:I was wondering what your experiences were with these:

Rook pair penalty (small penalty for having two rooks)
Knight pair penalty (small penalty for having two knights)
Having a pawn bonus (small bonus for having at least one pawn)

I see several program use these and was wondering if anyone had tested how good they might be?

Mark

What a coincidence... I am testing this as we speak (type?). I am testing many other combinations too at the same time. So far, after ~360 games it looks no significantly different.

Miguel

CThinker wrote:

mjlef wrote:I was wondering what your experiences were with these:

Rook pair penalty (small penalty for having two rooks)
Knight pair penalty (small penalty for having two knights)
Having a pawn bonus (small bonus for having at least one pawn)

I see several program use these and was wondering if anyone had tested how good they might be?

Mark

I have tested a number of these types of evaluation and I never liked the result. Positions quickly degenerate to endgames.

What do you mean? this type of evaluations are not good because positions with imbalances turn rapidly into endgames? I do not think I get what you meant.

Miguel

The same goes for material balance evaluation. They tend to result in quick equal exchanges.