For Marco---possible Stockfish bug

[lech]

By making to SF to solve one specific position, one can justify almost any change. What we need is a logical reasoning: why do you think that algorithmically it's better to allow alpha := beta. Give one (hypothetic) example situation where this could make positive difference.

Becouse there is only one VALUE_DRAW, and two sides.

[lech]

If we have no asserts than we proceed to step 2 that is the real games testing.

I tried to test it. The result was interesting, but I have to weak computer.

[zullil]

Hmm, my Stockfish 2.2.2 can solve this one.

Code: Select all

info depth 37 seldepth 45 score cp 7 lowerbound nodes 105904119 nps 7105274 time 14905 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 46 score cp 15 lowerbound nodes 106588199 nps 6978407 time 15274 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 26 lowerbound nodes 107664649 nps 6801734 time 15829 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 43 lowerbound nodes 108597029 nps 6657901 time 16311 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 68 lowerbound nodes 109584437 nps 6529101 time 16784 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 107 lowerbound nodes 110526895 nps 6456764 time 17118 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 164 lowerbound nodes 112962749 nps 6322070 time 17868 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 249 lowerbound nodes 117600036 nps 6271667 time 18751 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 377 lowerbound nodes 181320601 nps 6324843 time 28668 multipv 1 pv h7h6 f8g8 h6h5 h2e5 f6f1 e5c3 f1b5 e8e7 b5b7 e7e6 b7c6 c3e5 c6d5 e6e7 d5b7 e7e6 b3b4 g8h8 h5g4 h8h2 b7c8 e6e7 g4f3 h2b2 c8b7 e7f8 b4b5 f8g7 b5b6 e5d4 b7d5 d4b6 d5d6 b2b3 f3g4

In console mode, under Linux 64bit.

With default settings? I'll have to try again.

[zullil]

Hmm, my Stockfish 2.2.2 can solve this one.

Code: Select all

info depth 37 seldepth 45 score cp 7 lowerbound nodes 105904119 nps 7105274 time 14905 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 46 score cp 15 lowerbound nodes 106588199 nps 6978407 time 15274 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 26 lowerbound nodes 107664649 nps 6801734 time 15829 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 43 lowerbound nodes 108597029 nps 6657901 time 16311 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 68 lowerbound nodes 109584437 nps 6529101 time 16784 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 49 score cp 107 lowerbound nodes 110526895 nps 6456764 time 17118 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 164 lowerbound nodes 112962749 nps 6322070 time 17868 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 249 lowerbound nodes 117600036 nps 6271667 time 18751 multipv 1 pv h7h6 h2e5 h6g7 e5g3 c3c4 b5c4 e4e5 g3e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 37 currmove h7h6 currmovenumber 1
info depth 37 seldepth 52 score cp 377 lowerbound nodes 181320601 nps 6324843 time 28668 multipv 1 pv h7h6 f8g8 h6h5 h2e5 f6f1 e5c3 f1b5 e8e7 b5b7 e7e6 b7c6 c3e5 c6d5 e6e7 d5b7 e7e6 b3b4 g8h8 h5g4 h8h2 b7c8 e6e7 g4f3 h2b2 c8b7 e7f8 b4b5 f8g7 b5b6 e5d4 b7d5 d4b6 d5d6 b2b3 f3g4

In console mode, under Linux 64bit.

OK, after many fails, my SF succeeded too. SMP search is remarkably chaotic here:

Code: Select all

info depth 51 seldepth 60 score cp 8 lowerbound nodes 478124414 nps 4931964 time 96944 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4 b5c4 e4e5 h2e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 60 score cp 16 lowerbound nodes 842857406 nps 6066340 time 138940 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4 b5c4 e4e5 h2e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 62 score cp 28 lowerbound nodes 932649427 nps 5639535 time 165377 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4 b5c4 e4e5 h2e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 62 score cp 46 lowerbound nodes 1077295393 nps 4914736 time 219197 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4 b5c4 e4e5 h2e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 66 score cp 73 lowerbound nodes 1231451241 nps 5201176 time 236764 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 67 score cp 114 lowerbound nodes 1625386996 nps 4922968 time 330164 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 69 score cp 175 lowerbound nodes 1801198745 nps 4132592 time 435852 multipv 1 pv h7h6 h2e5 h6g7 e5h2 c3c4 b5c4 e4e5 h2e5 b3c4 e5f6 g5f6 d6d5 c4d5
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 69 score cp 267 lowerbound nodes 2105897877 nps 4283414 time 491640 multipv 1 pv h7h6
info depth 51 currmove h7h6 currmovenumber 1
info depth 51 seldepth 74 score cp 404 lowerbound nodes 4786923334 nps 6241164 time 766992 multipv 1 pv h7h6 h2e5

[lech]

[OK, after many fails, my SF succeeded too. SMP search is remarkably chaotic here:
info depth 51 seldepth 74 score cp 404 lowerbound nodes 4786923334 nps 6241164 time 766992 multipv 1 pv h7h6 h2e5

depth 51 - overpowering result

[zullil]

[OK, after many fails, my SF succeeded too. SMP search is remarkably chaotic here:
info depth 51 seldepth 74 score cp 404 lowerbound nodes 4786923334 nps 6241164 time 766992 multipv 1 pv h7h6 h2e5

depth 51 - overpowering result

I'm just proving that the solution can be found. I'm not claiming that it's being found efficiently.

[lech]

[OK, after many fails, my SF succeeded too. SMP search is remarkably chaotic here:
info depth 51 seldepth 74 score cp 404 lowerbound nodes 4786923334 nps 6241164 time 766992 multipv 1 pv h7h6 h2e5

depth 51 - overpowering result

I'm just proving that the solution can be found. I'm not claiming that it's being found efficiently.

I would add this change to Sting SF 2.0. Let there be tested.
Sting will contain an interesting interpretation of the 8 point. Next try to get a real human chess-engine without a loss of ELO.
If it will work well, I hope that it will go to Stockfish.

[Joerg Oster]

Sorry, forgot to mention.

I use different piece values in my own compile and I changed the MinimumSplitDepth for the smp search. But your default SF found it too, though more time needed.

But this position http://talkchess.com/forum/viewtopic.ph ... =&start=20 is only solved with Marek's addition.

@Marco: Marek is most likely right here, isn't he? Look at this thread where Don wrote:

Don't forget that a score >= beta is a failure - a common bug is to say:

if (score > beta) { do something ... }

That's a bug and should be:

if (score >= beta)

But I'm no expert, of course.

[lech]

VALUE_DRAW should be alpha for both sides. In other case a batter move can't appear.

[zamar]

By making to SF to solve one specific position, one can justify almost any change. What we need is a logical reasoning: why do you think that algorithmically it's better to allow alpha := beta. Give one (hypothetic) example situation where this could make positive difference.

Becouse there is only one VALUE_DRAW, and two sides.

So? This shouldn't be any kind of problem for negamax. Please give more details.