a question about Stockfish Cowardness

[syzygy]

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.

No it does not. You said that "Cowardice and Contempt 0 are quite alike" when in fact Contempt 0 gives symmetric evaluation in H3 and Cowardice makes Stockfish's king safety asymmetric.

Contempt at 2 is also not Cowardice but rather anti-Cowardice.
What Stockfish's Cowardice is for the engine's king safety, is Stockfish's Aggressiveness for the opponent's king safety. So Contempt = 2 for H3 could be emulated by increasing Aggressiveness and/or lowering Cowardice.

Stockfish's Contempt does not correspond to any of H3's Contempt values. Contempt for Stockfish is simply the "value" of a draw.

Without thinking too deeply about this, I would say H3's Contempt = 1 approach seems more sound that Stockfish's Contempt approach. With enough material left against a weaker opponent, it makes sense to reject an opportunity for a draw by repetition and it makes sense to actually be positive about the engine's winning chances even if objectively the engine's position is slightly inferior. With little material left H3 will not overestimate its position against a weaker opponent and take the draw if objectively the position is slightly inferior.

H3's Contempt = 1 (and = 2) also affects playing style if there is no draw opportunity. Stockfish's Contempt only makes a difference if there are draw opportunities (somewhere in the tree). Stockfish's Cowardice and Aggressiveness compensate for this (but are more focussed on king safety than on material, although the two of course are not independent).

[Paul Bedrey]

I nominate Evgenii Manev for computer chess enthusist of the year. I am amazed! I always thought changing engine settings was good for only getting a differing style. How can one setting make that big of a difference and yes it does improve it 100 elo.
Thanks again.

[carldaman]

I nominate Evgenii Manev for computer chess enthusist of the year. I am amazed! I always thought changing engine settings was good for only getting a differing style. How can one setting make that big of a difference and yes it does improve it 100 elo.
Thanks again.

+1

It also seems to improve SF4's play but I haven't run exhaustive tests at all.
Has any one done this with SF4? (The original post mentioned v2.3)
Well done, Mr. Manev!

[Paul Bedrey]

I am running two tests with Stockfish 4. One is short time control 4-2 no ponder, one machine using 4 cores each (44 games so far). The other is longer 30-3 with ponder using two machines, 4 cores (26 games so far). The setting seems to be scoring better than 60% in head to head with Houdini 3 Tactical.

I also am trying it with Stockfish 2.22 vs Houdini Tactical and its dead even after 54 games.

I don't have Komodo 5 but would be curious as to how this setting fairs against it.

[GenoM]

thanks:)

it would be nice if Marco Costalba says here if this is tested by the SF team and if they proved it to be right or wrong

ps:
I know that imroving an engine couldn't be so easy

[tpetzke]

I think it is common practice here in this forum to drive and publish conclusions with a very small number of games. However it is still wrong.

I would not be surprised that when Marco reads you statement based on 73 games, he will just smile and move on.

[GenoM]

I think it is common practice here in this forum to drive and publish conclusions with a very small number of games. However it is still wrong.

I would not be surprised that when Marco reads you statement based on 73 games, he will just smile and move on.

it's not about 73 games; it's about pointing out at this possibility.

I'm not a disciplined tester with a lot of processor time and super-modern hardware -- so i'm sharing a possibility with the community if someone wants to check it out. Is there something wrong with my attitude?

Code: Select all

  15 Strelka 5                       3003   64   61    92   72%  2837   39% 
  17 Stockfish 130818                2991   38   37   218   59%  2927   44% 
  20 Stockfish 130805                2984   39   39   208   59%  2926   43% 
  21 Stockfish 130819                2982   39   38   218   58%  2927   36% 
  23 Stockfish 130719                2977   38   38   245   59%  2896   34% 
  24 Stockfish 2.3 mobility          2963   42   41   232   66%  2809   31% 
  25 Stockfish 130809                2963   35   35   248   54%  2930   46% 
  27 Stockfish 2.3 coward            2961   42   41   236   64%  2818   26% 
  28 Stockfish 130803                2959   38   38   208   55%  2926   47% 
  29 Stockfish 2.3.1                 2957   46   46   155   60%  2854   43% 
  31 Stockfish 3 coward              2951   78   73    78   74%  2741   22% 
  32 Stockfish 120907                2947   44   44   181   57%  2849   39% 
  33 Stockfish 3                     2942   32   32   317   53%  2904   40% 
  34 Stockfish 4                     2941   35   35   260   51%  2933   37% 
  35 Komodo 3                        2941   54   54   131   63%  2790   36% 
  36 Gull 2.1                        2941   38   38   237   55%  2886   38% 
  39 Stockfish 4 mobility            2934   56   56   108   52%  2915   31% 
  44 Stockfish 2.3                   2924   37   37   251   55%  2864   38% 
  56 Stockfish 4 coward              2913   54   55   108   48%  2915   43% 
  57 Robbolito 0085g3                2911   57   55   127   69%  2734   25% 
  60 Sting SF-3                      2903   49   49   139   51%  2865   37% 
  63 Stockfish 2.2.2                 2898   38   38   245   56%  2820   41% 
  64 Stockfish 130825                2898   51   52   110   42%  2947   50% 
  71 Sting SF 2                      2889   54   55   122   51%  2834   31% 
  73 Stockfish beta9                 2884   47   48   142   45%  2901   40% 

[Modern Times]

What are your settings exactly ?

[GenoM]

it's about SF 2.3 actually. Cowardice is set up to the maximum (200)
http://talkchess.com/forum/viewtopic.ph ... 61&t=48734

[Paul Bedrey]

I think it is common practice here in this forum to drive and publish conclusions with a very small number of games. However it is still wrong.

I would not be surprised that when Marco reads you statement based on 73 games, he will just smile and move on.

That's ok I won't be offended. Houdini is top dog and I rarely run more than a 100 game match. Being that Stockfish is hard pressed to get 50% against Houdini I'll be pleased as punch to score 60% against Houdini tactical regardless of its statistical validity.

Besides at a time control of 4m 2s I might be dead before I get 10,000 games.