Page 1 of 1
Dirty quiescence
Posted: Tue Jan 12, 2016 1:01 pm
by geko
The best move for "8/8/2k1K3/8/8/2P2r2/8/8 b - - 0 1" is f3c3
in my engine at depth 1 is ok and centipawn is 564 (captured rook):
Code: Select all
info score cp 564 depth 1 nodes 18 time 0 pv f3c3
at depth 2 best move is c6b5 and centipawn is 568 because in quescence rook is captured after c6b5 e6e5
Code: Select all
info score cp 568 depth 2 nodes 79 time 8 pv c6b5 e6e5
this result sounds wrong, how could have f3c3 ?
Re: Dirty quiescence
Posted: Tue Jan 12, 2016 4:35 pm
by H.G.Muller
I don't get it. The Rook is never captured, it is a Pawn that is captured. And why do you think Rf3xc3 would be better than Kc6b5? Both moves win the Pawn, and it does see that. I don't see why the score would be 3 cP better for black with the Kings on e5 and b5 rather than on e6 and c6, but this must be something in your piece-square table.
Re: Dirty quiescence
Posted: Tue Jan 12, 2016 4:42 pm
by geko
sorry! he Rook is never captured, it is a Pawn that is captured.
Rf3xc3 capture the pawn immediatly, Kc6b5 capture in two or more steps
Re: Dirty quiescence
Posted: Tue Jan 12, 2016 7:41 pm
by H.G.Muller
geko wrote:Rf3xc3 capture the pawn immediatly, Kc6b5 capture in two or more steps
Yes, so what? If I can capture a Pawn in one step, but when I let the Pawn escape, I can gain a Queen in two steps, would you also go for the Pawn because 1 is less than 2 ? An engine does not go for the fastest gain, it goes for the highest gain. The final position after Rxc3 has score +564, that after Kb5 Ke5 Rxc3 +568. And 568 > 564. So the engine goes for the best move, gaining 4 extra centi-Pawn by not being greedy, and postponing the grabbing of the Pawn.
Even if the final positions were exactly the same, the engine would not care if it captures the Pawn near the end of the PV or immediately. Minimax does not award any score for realizing the gain faster, it is only the final position that counts. If you don't want that, you should use a delayed-loss bonus.