Static Exchange Evaluation

[arjunbasandrai]

I am trying to implement SEE in my chess engine - "Shuffle" (https://github.com/ArjunBasandrai/shuffle-chess-engine/)

However, I am having a bit of trouble understanding what the algorithm should return.

Assuming that the middlegame material scores are - (P=82, N=337, B=365, R=477, Q=1025, K=12000, p=-82, n=-337, b=-365, r=-477, q=-1025, k=-12000)

In the position (8/R7/5k1P/5p2/p1p2P2/Br1pPK1p/1P1R4/r7 b - -)

What should be the SEE of the move :-

• b3a3

• b3b2

My implementation is giving the result for b3a3 as -112 and that for b3b2 as -395. Are these values correct?
From what I have understood the value for b3a3 should be -30 and that b3b2 should be -395 only. Am I right?

[lithander]

I have compiled this little set of test positions for SEE. Some of them I found, some I added myself.
https://github.com/lithander/Leorik/blo ... st/see.epd

[arjunbasandrai]

Thanks for providing the test suite!

One more doubt I have is regarding where exactly to use pruning using SEE in q-search and the negamax? Any resources that I can take a look at?

[arjunbasandrai]

I have compiled this little set of test positions for SEE. Some of them I found, some I added myself.
https://github.com/lithander/Leorik/blo ... st/see.epd

I tested my new implementation on the Arasan Test Suite. That and yours also has a FEN (2r5/1P4pk/p2p1b1p/5b1n/BB3p2/2R2p2/P1P2P2/4RK2 w - -) and the SEE value of it is supposed to be the value of a Rook. Should it not be Bishop value?

[abulmo2]

I have compiled this little set of test positions for SEE. Some of them I found, some I added myself.
https://github.com/lithander/Leorik/blo ... st/see.epd

I tested my new implementation on the Arasan Test Suite. That and yours also has a FEN (2r5/1P4pk/p2p1b1p/5b1n/BB3p2/2R2p2/P1P2P2/4RK2 w - -) and the SEE value of it is supposed to be the value of a Rook. Should it not be Bishop value?

[d]2r5/1P4pk/p2p1b1p/5b1n/BB3p2/2R2p2/P1P2P2/4RK2 w - - Considering the move c3c8 (or Rxc8), the value can be a bishop if you ignore pawn promotion or a rook if you take into account that the pawn on b7 can promote to a queen. Taking or not into account pawn promotion during a SEE sequence is a mater of taste. On the above position the real value of the exchange is a rook, because the bishop is not suppose to capture to avoid the pawn promotion. But there are cases where considering pawn promotions would be a mistake, or too complicated to implement correctly.

[arjunbasandrai]

I have compiled this little set of test positions for SEE. Some of them I found, some I added myself.
https://github.com/lithander/Leorik/blo ... st/see.epd

I tested my new implementation on the Arasan Test Suite. That and yours also has a FEN (2r5/1P4pk/p2p1b1p/5b1n/BB3p2/2R2p2/P1P2P2/4RK2 w - -) and the SEE value of it is supposed to be the value of a Rook. Should it not be Bishop value?

[d]2r5/1P4pk/p2p1b1p/5b1n/BB3p2/2R2p2/P1P2P2/4RK2 w - - Considering the move c3c8 (or Rxc8), the value can be a bishop if you ignore pawn promotion or a rook if you take into account that the pawn on b7 can promote to a queen. Taking or not into account pawn promotion during a SEE sequence is a mater of taste. On the above position the real value of the exchange is a rook, because the bishop is not suppose to capture to avoid the pawn promotion. But there are cases where considering pawn promotions would be a mistake, or too complicated to implement correctly.

Thanks for clarifying! That makes much more sense.

I do have one last doubt. Where and how should I use the SEE function for pruning. In negamax should I use it before LMR and after razoring, futility pruning, LMP? What should be the cutoff SEE value to prune a branch?