TalkChess.com

[Mike Libanan]

First, sorry I'm just a hobbyist coder, but I have some bad opinion on
modern day eval functions.

I have browsed codes of eval() of some engines and noticed a bunch of
useless code which can't be applied on the current board position.

There are lines of code for checking if(BAD_Bishop) or if(BishopPair) or
if(King_Uncastled), which just makes the evaluation function slow due to
useless knowledge which can't be applied all the time. Like when in
endgame phase without bishops, it would be useless to keep on checking
those codes related to bishops.


GMs just uses specific knowledge and includes this to his assessment of current position on hand.
To make engines assess a position like a GM, I think Evaluation functions
can be implemented similar to the techniques used by modern malwares
like zbot/zeus. Modern/polymorphic malwares just injects some code to a
running process, and I think similar technique can be used by a chess
engine. Using this code injection techniques, chess engines can then
inject only a lean specialised knowledge, or specialised eval function,
which can be used for the current position.


Like during K+B+N vs K endgame, most engines can't announce
checkmate without the use of a TB, but if an engine can grab a file and
inject this knowledge to its own running process, then an engine can
announce mate without using a TB. Similar eval injection can be done on
specific opennings where PST can change depending on the openning, or
knowledge of R endgames.