How much elo is a good king safety function worth?

[Whiskers]

I'm implementing king safety into my engine Willow and am wondering how much ELO gain is feasible to get out of it. Currently I have a basic king safety function (attack tables for the middlegame, the results of which are scaled by how many defending pawns are near the king) and have gotten ~50-60 elo from it in self play which is a good start.

If you have a good king safety implementation in your engine (or even if you don't ), how much of an improvement is it?

[algerbrex]

~46 Elo apparently, though I didn't remember.

Note, however, I don't consider my king safety function good. It's pretty basic and only takes into account attacks around the king. I don't have a concrete engine to cite, but I suspect a good king safety - one takes into account a variety of factors like king attacks, pawn structure, checking moves, etc. - is worth 100 Elo or more.

[emadsen]

I got 63 Elo. But the benefit to your engine may be more or less depending on what other features it has or lacks. Search and evaluation features interact in non-linear ways.

[jdart]

#1 thing to focus on in my opinion is having a bug-free implementation of search and eval.
Many weak engines are weak because they are buggy.
That said, king safety is very important. It doesn't matter as much if you are playing weak opponents,
but if you play an engine that has good king safety eval and yours is weak, you will be smashed.

[jhonnold]

IIRC, ~110 Elo in Berserk

https://github.com/jhonnold/berserk/blo ... val.c#L851

[chrisw]

I'm implementing king safety into my engine Willow and am wondering how much ELO gain is feasible to get out of it. Currently I have a basic king safety function (attack tables for the middlegame, the results of which are scaled by how many defending pawns are near the king) and have gotten ~50-60 elo from it in self play which is a good start.

If you have a good king safety implementation in your engine (or even if you don't ), how much of an improvement is it?

Absolutely no where near as much as if you used a simplest of simple tiny NNUEs instead. +300 Elo minimum.

HCE king safety. Compared to none at all +200. Compared to what you’ve already done, maybe 50-100.

[Whiskers]

I'm implementing king safety into my engine Willow and am wondering how much ELO gain is feasible to get out of it. Currently I have a basic king safety function (attack tables for the middlegame, the results of which are scaled by how many defending pawns are near the king) and have gotten ~50-60 elo from it in self play which is a good start.

If you have a good king safety implementation in your engine (or even if you don't ), how much of an improvement is it?

Absolutely no where near as much as if you used a simplest of simple tiny NNUEs instead. +300 Elo minimum.

HCE king safety. Compared to none at all +200. Compared to what you’ve already done, maybe 50-100.

Right now I definitely do not have the programming knowledge to train my own NNUE net, and using an already trained one like SFs feels like cheating. Maybe eventually I'll get good enough at coding