For Ed Schroeder: Rebel's Pawn Shield and Pawn Storm Eval

[mfournier]

Hello Mr. Schroeder,
I've read and studied with great interest your description of Rebel's inner workings. It helped me a lot to improve the evaluation of my little engine.

In particular, I think the king safety evaluation is very good but unfortunately the evaluation of pawn shield, pawn rams and opposite castlings was not described.

I would like to know if you could find a bit of time to write the description of those parts since I would like to know how you evaluate them.

Thanks a lot for your paper, I think it's one of the best readings in computer chess.

Marcel

[Rebel]

Hello Marcel, welcome.

The most simple way is to create 2 piece_square_tables in your EVAL. One that evaluates when the king is on A|B|C and one for the F|G|H files. And in reverse for the black pieces (king).

For a more advanced way check (for instance) the Fruit sources.

Good luck.

[mfournier]

Hello Marcel, welcome.

The most simple way is to create 2 piece_square_tables in your EVAL. One that evaluates when the king is on A|B|C and one for the F|G|H files. And in reverse for the black pieces (king).

For a more advanced way check (for instance) the Fruit sources.

Good luck.

Thanks Ed.

I already took a look at Fruit's evaluation of pawn storms/shield.

I thought Rebel's one was more sophisticated, as the rest of the eval is, but perhaps there is no need for that. From my tests, it seems that even major changes on pawn storm/shield eval have no particular effects on Elo. The main part of king safety evaluation is way more important.

Marcel

[diep]

Hello Marcel, welcome.

The most simple way is to create 2 piece_square_tables in your EVAL. One that evaluates when the king is on A|B|C and one for the F|G|H files. And in reverse for the black pieces (king).

For a more advanced way check (for instance) the Fruit sources.

Good luck.

Thanks Ed.

I already took a look at Fruit's evaluation of pawn storms/shield.

I thought Rebel's one was more sophisticated, as the rest of the eval is, but perhaps there is no need for that. From my tests, it seems that even major changes on pawn storm/shield eval have no particular effects on Elo. The main part of king safety evaluation is way more important.

Marcel

As for kingsafety it seems to be heavily dependant upon search depth. Of course a simple kingsafety that's doing well is always important - no question about it.

If you get 10-14 plies or so (classical plies not the ultra thin stuff we get now) then a huge kingsafety like diep has is total dominating the game.

Basically you have no chance against it.

It took me some time to grasp how Fruit got away with nearly no kingsafety at all and still win that many games. My conclusion then was that kingsafety to big extend is a shortterm advantage. Once you search that 17+ plies (nearly real plies, still not the ultra thin plies) then of course if the beancounters can get away tactical at kingside, they'll go for that passed pawn at c-file and win.

So as it appears, kingsafety mostly gives a short term advantage and benefit of it is less when opponents just search their way out.

Even then i feel building a good kingsafety is critical.