some eval experiments (call it an engine if you want)

https://github.com/nescitus/meander

I do not consider it a full fledged engine, rather an attempt to bring some more complexity to handcrafted evaluation function. Tuning this stuff is pain in the ass, but I hope it will inspire someone to try the following ideas:


- Some pawn and knight piece/square tables are chosen depending on the pawn structure in the center and the castling status.
- Many aspects of pawn evaluation are replaced by pawn x pawn table, evaluating doubled, defended and phalanx pawns as well as relationships between more distant pawns (i.e. white pawns on e5 and h5 = good)

The gain is meagre, 56% score against the latest Rodent, translating into 40 Elo. I might release it officially if I get 100 Elo from eval function alone. The grand plan is:

- tuning the hell out of pawn x pawn table
- adding bishop x pawn and knight x pawn table to catch some more features not accounted for in classic HCE
- adding more piece/square tables for specific central pawn structures.

Frankly speaking, I do not expect all of this to work, but I want to be surprised.

i think some good games to tune on would be GM Hikaru's blitz games, very dynamic and over 30k so lots of data, should make a high quality pawn table