issue with check extension

[adams161]

pulsar played an IM last night and it sacked a pawn with a bishop, it could win a second pawn but not much more. I ran it through analysis and saw that pulsar didnt anticipate its opponents response. it thought that it would be bishop takes pawn and then pawn takes bishop. playing a different rook move first to force pulsars queen to move first is what happened.

if it went down the way pulsar saw it, it would get a nice fork. but that wasnt reasonable.

I tried disabling hashing then disabling null move and it still played the move. i checked teh move generator and no issues ( though i didnt expect any they have been well tested). it would see that move was bad if you left it to search longer ( longer than it had in game time), but early on it liked the sack.

finnaly i turned off my check extension and pulsar never considered that bad move.

I had it extending all checks if the search had reached ply 3 in a given iteration. so search() then search() then search() check extending enabled.

i compromised and left the check extension on but bumped it up another level of search before it would turn on and it also doesnt consider the move. any idea why a check extension early in teh search tree might make an engine try a move that is bad and it wouldnt try without the extension. i'd put up the fen but i'd need someone to give me the command for that.

bear in mind i ran this with hashing turned off and same issue. if it had the check extension it would try a move that was faulty.

Mike

[sje]

A diagram would be useful here.

[adams161]

the fen is rnb1kb1r/1p3pp1/p4n1p/3p4/3N2PP/q1N1BP2/P1PQ1K2/1R3B1R b kq - 0 13. the bad move is Bxg4. nothing i can try turns that into a diagram though i recall i made diagrams with some html like code years ago in the old talkchess forum.

Mike

[kiroje]

put [D] in front of the fen

like this

[D] rnb1kb1r/1p3pp1/p4n1p/3p4/3N2PP/q1N1BP2/P1PQ1K2/1R3B1R b kq - 0 13

[MattieShoes]

Hmm. How deep is it searching? Is it doing a normal Q-search?

[bob]

I did a quick check with crafty, and it won't play Bxg4. I then played black and played this move, and it only took 4 plies to see that PxB is no good and that Rb3 is almost +2 (good for white).

I'd guess you have a serious bug in your actual search here...

[sje]

[D] rn2kb1r/1p3pp1/p4n1p/3p4/3N2bP/q1N1BP2/P1PQ1K2/1R3B1R w kq - 0 14

Code: Select all

[+0.744/1/0.015/40/0] 14 fxg4 Nxg4+ 15 Kg2
[-1.170/2/0.304/272/0] 14 fxg4
[-0.750/2/0.304/340/0] 14 Rxb7 Bd7
[-0.745/3/0.325/861/0] 14 Rxb7 Bd7 15 Be2
[-0.765/4/0.516/6,857/0] 14 Rxb7 Bc8 15 Rc7 Bd7
[-0.760/5/0.904/31,036/0] 14 Rxb7 Bc8 15 Rc7 Bd7 16 Be2
[-0.867/6/1.358/211,443/0] 14 Rxb7 Bc8 15 Rc7 Bd7 16 Be2 Bd6
[-0.535/6/1.378/229,834/0] 14 fxg4 Qxc3 15 Qxc3 Ne4+ 16 Kg2 Nxc3 17 Rxb7 Nxa2
[-0.373/6/1.626/369,946/0] 14 Rb3 Qe7 15 fxg4 Nxg4+ 16 Kf3 Nxe3 17 Qxe3
[-0.267/7/1.977/577,911/0] 14 Rb3 Qe7 15 fxg4 Nxg4+ 16 Kf3 Qd7 17 Kg2 Be7
[-0.010/8/3.934/2,267,769/0] 14 Rb3 Qe7 15 fxg4 Nxg4+ 16 Ke2 f6 17 Nxd5 Qe4 18 Nc7+ Kf7

[adams161]

the funny thing is it never considers BxG4 at all on any ply search when i move check extensions to not turn on early. i'll have to look into it more.
Mike

[adams161]

with check extensions on it got to ply 6 in 3 0 with them off it can get deper maybe ply 7. it saw the move Bxg4 as good with check extensions on, on ply 6 and bad on ply 7, and wtih check extending pushed up to later in search or off it never considered the move and behaviore was similar to crafty it immediatly saw Rb3.

Mike