value of king tropism in eval function

[silentshark]

The title says it all - does this have some value, or is it just too crude to help much these days?

I've been playing around with it, and found (surprisingly) that kind king tropism seems to hurt strength, unless it is kept very "light."

Views?

[Jan Brouwer]

The title says it all - does this have some value, or is it just too crude to help much these days?

I've been playing around with it, and found (surprisingly) that kind king tropism seems to hurt strength, unless it is kept very "light."

Views?

King tropism is the distance between king and attacking pieces to measure king safety (chess programming wiki), right ?
It is probably to indiscriminating, and if evaluation strength is limited by the biggest error that can occur, as is often suggested, then I can image that only small values work.
Perhaps for "slow" pieces (knight), the weight can be a bit larger.

[hgm]

Just an observation: in Xiangqi the value of tropism of Knights and Pawns is really dramatic. A program with light tropism gets totally slaughtered by the same program with heavy tropism for these pieces. It is not centralization of Knights and Pawns that counts, but distance to the enemy King.

[Ferdy]

The title says it all - does this have some value, or is it just too crude to help much these days?

I've been playing around with it, and found (surprisingly) that kind king tropism seems to hurt strength, unless it is kept very "light."

Views?

I think the idea is to teach the program to get interested in hunting opp king, giving high score individually will not work as you have found out.
Try combination tropism - queen and pawn, queen and knight, queen and bishop, queen and rook. Pawn and knight probably will not work, same with knight and bishop, sometimes rook and bishop and rook and knight will not work also, then try triples, rook, knight and bishop. Also effective is king, pawn, rook or king, pawn, bishop and knight attacking opp king and forcing it to the edge of the board. Combination of efforts directed in a given location with a perfect time is probably the secret to an interesting and stronger program .

[silentshark]

The title says it all - does this have some value, or is it just too crude to help much these days?

I've been playing around with it, and found (surprisingly) that kind king tropism seems to hurt strength, unless it is kept very "light."

Views?

I think the idea is to teach the program to get interested in hunting opp king, giving high score individually will not work as you have found out.
Try combination tropism - queen and pawn, queen and knight, queen and bishop, queen and rook. Pawn and knight probably will not work, same with knight and bishop, sometimes rook and bishop and rook and knight will not work also, then try triples, rook, knight and bishop. Also effective is king, pawn, rook or king, pawn, bishop and knight attacking opp king and forcing it to the edge of the board. Combination of efforts directed in a given location with a perfect time is probably the secret to an interesting and stronger program .

an interesting pointer! thanks