Designing an analysis friendly Stockfish?

[snino64]

In multipv to change the position, if black has only one move, the engine returns the data from the previous spin.
For example, in the position
6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1

1 1 [M19] 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Bb5 Kc7 10.Ke6 Kd8 11.Bd7 Kc7 12.Ke7 Kb8 13.Kd8 Kb7 14.Bb5 Kb8 15.Nd6 Ka7 16.Kc7 Ka8 17.Bf1 Ka7 18.Nc8+ Ka8 19.Bg2#
2 1 (M21) 1.Nf3
3 1 (M21) 1.Nh3
4 1 (M21) 1.Ne4

1. Nf7

1 1 [M18] 1.... Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Bb5 Kc7 10.Ke6 Kd8 11.Bd7 Kc7 12.Ke7 Kb8 13.Kd8 Kb7 14.Bb5 Kb8 15.Nd6 Ka7 16.Kc7 Ka8 17.Bf1 Ka7 18.Nc8+ Ka8 19.Bg2#
2 1 (M21) 1.... Nf3
3 1 (M21) 1.... Nh3
4 1 (M21) 1.... Ne4

Scid Problem?

This does not happen with 2.01 stockfish JA

[Jeremy Bernstein]

In multipv to change the position, if black has only one move, the engine returns the data from the previous spin.
For example, in the position
6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1

1 1 [M19] 1.Nf7 Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Bb5 Kc7 10.Ke6 Kd8 11.Bd7 Kc7 12.Ke7 Kb8 13.Kd8 Kb7 14.Bb5 Kb8 15.Nd6 Ka7 16.Kc7 Ka8 17.Bf1 Ka7 18.Nc8+ Ka8 19.Bg2#
2 1 (M21) 1.Nf3
3 1 (M21) 1.Nh3
4 1 (M21) 1.Ne4

1. Nf7

1 1 [M18] 1.... Kf8 2.Bh7 Ke8 3.Ne5 Kd8 4.Be4 Kc7 5.Nc4 Kd7 6.Kf7 Kd8 7.Bc2 Kc7 8.Ba4 Kd8 9.Bb5 Kc7 10.Ke6 Kd8 11.Bd7 Kc7 12.Ke7 Kb8 13.Kd8 Kb7 14.Bb5 Kb8 15.Nd6 Ka7 16.Kc7 Ka8 17.Bf1 Ka7 18.Nc8+ Ka8 19.Bg2#
2 1 (M21) 1.... Nf3
3 1 (M21) 1.... Nh3
4 1 (M21) 1.... Ne4

Scid Problem?

This does not happen with 2.01 stockfish JA

It's not coming from the engine:

Stockfish 2.0.1 PA GTB 64bit by Tord Romstad, Marco Costalba, Joona Kiiski
setoption name MultiPV value 4
position fen 6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1
go infinite
info depth 1
info depth 1 seldepth 37 multipv 1 score mate 19 time 47 nodes 477 nps 10148 tbh
its 433 pv g5f7 g8f8 g6h7 f8e8 f7e5 e8d8 h7e4 d8c7 e5c4 c7d7 f6f7 d7d8 e4c2 d8c7
 c2a4 c7d8 a4b5 d8c7 f7e6 c7d8 b5d7 d8c7 e6e7 c7b8 e7d8 b8b7 d7b5 b7b8 c4d6 b8a7
 d8c7 a7a8 b5f1 a8a7 d6c8 a7a8 f1g2
info depth 1 seldepth 1 multipv 2 score mate 21 time 63 nodes 477 nps 7571 tbhit
s 433 pv g5f3
info depth 1 seldepth 1 multipv 3 score mate 21 time 78 nodes 477 nps 6115 tbhit
s 433 pv g5h3
info depth 1 seldepth 1 multipv 4 score mate 21 time 78 nodes 477 nps 6115 tbhit
s 433 pv g5e4
info nodes 477 nps 5074 time 94
bestmove g5f7 ponder g8f8
position fen 6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1 moves g5f7
go infinite
info depth 1
info depth 1 seldepth 36 multipv 1 score mate -18 time 16 nodes 458 nps 28625 tb
hits 414 pv g8f8 g6h7 f8e8 f7e5 e8d8 h7e4 d8c7 e5c4 c7d7 f6f7 d7d8 e4c2 d8c7 c2a
4 c7d8 a4b5 d8c7 f7e6 c7d8 b5d7 d8c7 e6e7 c7b8 e7d8 b8b7 d7b5 b7b8 c4d6 b8a7 d8c
7 a7a8 b5f1 a8a7 d6c8 a7a8 f1g2
info nodes 458 nps 7269 time 63
bestmove g8f8 ponder g6h7

[Jeremy Bernstein]

Jeremy here is a tour 3"3' hash 1024 ponder off ht=on on i7-980 Gran has some problems on fast time controls when it starts TB access and can lose on time. Not the case here though.

Thanks a lot!

Some of the TB probing code I added doesn't take time management into account. I can imagine a fairly simple fix for the worst case, where it sniffs a mate in, say 100, and then generates long PVs out of TB hits, which obviously require _some_ time. There's no reason to do this when under time pressure. I'll post a Gran2d build shortly.

Jeremy

[Jeremy Bernstein]

Jeremy here is a tour 3"3' hash 1024 ponder off ht=on on i7-980 Gran has some problems on fast time controls when it starts TB access and can lose on time. Not the case here though.

Thanks a lot!

Some of the TB probing code I added doesn't take time management into account. I can imagine a fairly simple fix for the worst case, where it sniffs a mate in, say 100, and then generates long PVs out of TB hits, which obviously require _some_ time. There's no reason to do this when under time pressure. I'll post a Gran2d build shortly.

Jeremy

This version should be better behaved under low-time situations.

Jeremy

[keoki010]

eremy, heck my board has a bad key. Please take a look at the attached pgn. It's for the 3/3 tour I ust held for pamc and gran2c. Some of the time losses and ad udicated endings seem strange.

[Jeremy Bernstein]

eremy, heck my board has a bad key. Please take a look at the attached pgn. It's for the 3/3 tour I ust held for pamc and gran2c. Some of the time losses and ad udicated endings seem strange.

The build I just posted (gran2d) should solve a time loss issue with 5 or fewer pieces on the board. Arena sometimes does strange stuff with adjudications, but I guess it all evens out over a lot of games.

I'm running a 1/1 tournament between stock-Stockfish, gran2c and gran2d right now. Last I checked, they were about even, gran2d and stock ahead by a nose.

Jeremy

[keoki010]

Yes 2d seems a slight bit better than 2c. I'll try to run another tour tonight at 3/3.

[keoki010]

You can dl' the pgn from Rybka forum>computer chess> Designing an analysis friendly Stockfish. Your site won't let me attach something for some reason.http://rybkaforum.net/mwf/rybkaattach/1 ... ament6.pgn

[Jeremy Bernstein]

Bildschirmfoto 2011-02-24 um 07.55.04.png (10.98 KiB) Viewed 1671 times

[keoki010]

Bildschirmfoto 2011-02-24 um 07.55.04.png

2d easily beat 2c in 3/3 however it still had 1 time loss. it was +5 =4-1 for a 10 game match 3/3