I tried to mess up with settings of Stockfish. With Cowardness set up to Max (200) Stockfish 2.3 is showing some very strange results (~+100 points compared to default settings). The sample is small, so I would like to ask some of you to try this setting and to share results here.
Thanks.
PS: here is some excerpt from my home-made list to show why i'm calling results of Stockfish 2.3 Coward "strange" All games are blitz with increment played under Arena GUI with book of GM games set to 5-8 moves.
I suppose Cowardice at Max is equivalent to Houdini 3 at Contempt 0 ; similar to Houdini 3 at Contempt 0 having a higher ELO rating than Houdini at default Contempt 1.
Read somewhere that the default settings of most top Engines are deliberately set at slightly aggressive levels...more wins mean more attractive to the buyer. But this is not necessarily the strongest setting.
I suppose Cowardice at Max is equivalent to Houdini 3 at Contempt 0
Quite unlikely, since Stockfish has a separate Contempt Factor parameter. Now I'm curious enough to look at the source code ...
[EDIT] It seems that the Cowardice parameter affects how much weight the engine places on the safety of its own king.
On the contrary, quite likely, as Cowardice and Contempt 0 are quite alike ( they almost mean the same thing, in this context ! ) and Houdini 3 does NOT have a separate Cowardice factor !! So, I think that one can safely assume that the Cowardice factor is DEFINITELY factored in, in Houdini's version of Contempt !
I suppose Cowardice at Max is equivalent to Houdini 3 at Contempt 0
Quite unlikely, since Stockfish has a separate Contempt Factor parameter. Now I'm curious enough to look at the source code ...
[EDIT] It seems that the Cowardice parameter affects how much weight the engine places on the safety of its own king.
On the contrary, quite likely, as Cowardice and Contempt 0 are quite alike ( they almost mean the same thing, in this context ! ) and Houdini 3 does NOT have a separate Cowardice factor !! So, I think that one can safely assume that the Cowardice factor is DEFINITELY factored in, in Houdini's version of Contempt !
I think you're essentially wrong here. Contempt influences an engine's willingness to accept a draw. Not closely related to Stockfish's cowardice, which is about safety of the engine's king. But perhaps an expert can clarify.
I suppose Cowardice at Max is equivalent to Houdini 3 at Contempt 0
Quite unlikely, since Stockfish has a separate Contempt Factor parameter. Now I'm curious enough to look at the source code ...
[EDIT] It seems that the Cowardice parameter affects how much weight the engine places on the safety of its own king.
On the contrary, quite likely, as Cowardice and Contempt 0 are quite alike ( they almost mean the same thing, in this context ! ) and Houdini 3 does NOT have a separate Cowardice factor !! So, I think that one can safely assume that the Cowardice factor is DEFINITELY factored in, in Houdini's version of Contempt !
I think you're essentially wrong here. Contempt influences an engine's willingness to accept a draw. Not closely related to Stockfish's cowardice, which is about safety of the engine's king. But perhaps an expert can clarify.
No expert if you look at my occasional games but I think Anil is not so wrong here even if for the simple fact that with these two parameters you change the style of play for Stockfish and Stockfish takes less risk with high cowardice. That is exactly what a human player would do if he thought the opponent would start making mistakes on his own without going into very sharp battles. Or in the case of a stronger opponent would be satisfied with a draw and play as solidly as possible. If you think your opponent calculates deeper than you, going into a king attack may backfire. Never going for the attack will not win you many games either so a cowardice of 200 I don't know if that is good even against Houdini ...
Aggressiveness doubles as Stockfish' contempt against weaker oppponents especially. It is one of the reasons in my opinion why we have a hard time showing elo gains for Stockfish version of contempt against SF 3 at least. The two terms Aggressiveness and Cowardice do make evaluation asymmetrical for the two sides in the game which should not work maybe, theoretically, but inderectly they also make search asymmetrical. Richard Lang could tell you that asymmetrical search can work at leat it did work extremely well in the time of Chess Genius before nullmove etc. Curiously though I do not know of many other examples of programs that do this (apart from Stockfish now, a little bit, because of these asymmetric evals).
Regards, Eelco
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
syzygy wrote:Just see here:
Contempt = 0: symmetric evaluation.
Contempt = 1: piece value imbalance.
Contempt = 2: piece value imbalance and king safety imbalance.
Bingo ! Proves my point completely that Cowardice is factored in, in Houdini's Contempt.
Thank You, Ronald !
Also, I'm getting very good results with the Cowardice maxed out in the latest Stockfish development version, in my online games ! ( Of course, nothing I can prove by the exacting standards of the folk here ! )
I think you are right about the asymmetric search business...seems to be working positively for Stockfish, at least so far.
All games are blitz with increment played under Arena GUI with book of GM games set to 5-8 moves.
...