This release restores the evaluation penalty for "a cornered bishop blocked by a friendly pawn diagonally in front of it" that was removed in 1.9. So, 2.0.1 isn't quite equal to 2.0 in chess960. (In standard chess, Stockfish knows to skip this test.)
UncombedCoconut wrote:This release restores the evaluation penalty for "a cornered bishop blocked by a friendly pawn diagonally in front of it" that was removed in 1.9. So, 2.0.1 isn't quite equal to 2.0 in chess960. (In standard chess, Stockfish knows to skip this test.)
Yes you are right, in Chess960 not only we fix the breakage but restored a piece of code removed in 1.9, anyhow there is NO FUNCTIONAL CHANGE for standard chess. This is important because testers can keep results accumulated so far.
BTW did you spot something "familiar" among the changes ? Thanks for that !
UncombedCoconut wrote:This release restores the evaluation penalty for "a cornered bishop blocked by a friendly pawn diagonally in front of it" that was removed in 1.9. So, 2.0.1 isn't quite equal to 2.0 in chess960. (In standard chess, Stockfish knows to skip this test.)
Well spotted. Because chess960 support was broken in 2.0 anyway, we allowed ourselves to add this in addition to the bug fixes. 2.0.1 is functionally equivalent to 2.0 only in standard chess.
According to my own tests, this evaluation term is actually worth a lot of Elo points in chess960. It will be interesting to see whether Ray's tests confirm this.
UncombedCoconut wrote:This release restores the evaluation penalty for "a cornered bishop blocked by a friendly pawn diagonally in front of it" that was removed in 1.9. So, 2.0.1 isn't quite equal to 2.0 in chess960. (In standard chess, Stockfish knows to skip this test.)
Well spotted. Because chess960 support was broken in 2.0 anyway, we allowed ourselves to add this in addition to the bug fixes. 2.0.1 is functionally equivalent to 2.0 only in standard chess.
According to my own tests, this evaluation term is actually worth a lot of Elo points in chess960. It will be interesting to see whether Ray's tests confirm this.
I wonder if you need a special rule for chess960.
Why not to use the simple rule that a bishop that cannot move is bad(that is also relevant for Bc1 blocked by pawns at b2 and d2)?
Ba1 blocked by pawn b2 is a private case of many cases that can happen in chess.
I think that it may be interesting to give a big bonus for the opponent(something like 2 pawns if the bishop has no moves)
Here is an example for a position that white is losing because the bishop cannot move.
[D]6kb/1p6/pP6/Bp4K1/1P4P1/8/8/8 w - - 0 1
Of course there are cases when the bishop can move later(for example with no black pawn at b7) but I am not sure if the wrong score for these positions is going to have a big influence
Stockfish hello to the team!
I am really surprised how programs with the fantastic quality of
Stockfish, Houdini, Critter, among others, are free, and their
developers fix any bugs yet with incredible speed.
While the paid programs and expensive (Rybka 3, 4, etc.) that make their
Buyers like idiots without corrections.
Congratulations to these amazing men and their wonderful machines!
Here is another example that stockfish underevaluates mobility
based only on material the score is +2(if you use bishop=3 and pawn=1) but it is clear that black has big positional advantage(the white bishop cannot move and the black king is more active and black has a passed pawn)
For some reason stockfish evaluates the position as more than +2 for white.
New game - Stockfish 1.9 JA, 5'/40+5'/40+5'/40
[D]8/7p/8/8/6k1/1p1p4/1P1P4/2B3K1 w - - 0 1
If I simply remove the b2 and b3 pawn there is no big difference in the static evaluation(only 0.12 pawn difference between 2 squares for the bishop and 0 squares)
I think that the difference should be significantly higher(even if stockfish does not know that the bishop is blocked forever).
Maybe difference of 2 pawns is too much but as a chess player difference of only 0.12 pawns between 2 squares for the bishop and 0 squares for the bishop seems to me a joke.
New game - Houdini 1.5 w32
[D]8/7p/8/8/6k1/3p4/3P4/2B3K1 w - - 0 1
I can add that the difference between 0 squares and 10 squares seems to be near 0.8 pawns and it seems that the evaluation is almost linear on the number of squares(I think that the difference in mobility between 0 squares and 2 squares should be higher than the difference between 2 and 10)
third example
New game - Stockfish 2.0.1 JA
[D]8/7p/8/2B5/6k1/1p1p4/1P1P4/6K1 w - - 0 1
