Recent SF vs H6.03 match

[Nordlandia]

i7-4790K 4.5GHz
4-cores each
1024 mb hash each
Houdini contempt at -6
SF contempt at +24
EGTB: Probing 5-men with auto adjudication at 6-pieces.
Book used LittleCat_1.8.2.bin
GUI: cutechess gui
TC: 7m+7s

Score of stockfish_20032617_x64_bmi2 vs Houdini 6.03: +12 -2 = 86 [0.550]
Elo difference: 34.9 +/- 24.9, LOS: 99.6 %, Drawratio: 86.0 %

100 of 100 games finished.

Tournament duration: about 1.5 days

«It seems quite apparent that negative contempt make Houdini capable to put up more resistance, against stronger opposition»

Any comment would be appreciated.

[Jouni]

Short 1m+1s test shows big gain from -6 contempt:

Code: Select all

                                
1   Stockfish 11 64 BMI2  +191  +26/=23/-1 75.00%   37.5/50
2   Houdini 6 x64-pext    -191  +1/=23/-26 25.00%   12.5/50

Code: Select all

Houdini con = -6

                                
1   Stockfish 11 64 BMI2   +56  +9/=40/-1 58.00%   29.0/50
2   Houdini 6 x64-pext     -56  +1/=40/-9 42.00%   21.0/50

[lkaufman]

i7-4790K 4.5GHz
4-cores each
1024 mb hash each
Houdini contempt at -6
SF contempt at +24
EGTB: Probing 5-men with auto adjudication at 6-pieces.
Book used LittleCat_1.8.2.bin
GUI: cutechess gui
TC: 7m+7s

Score of stockfish_20032617_x64_bmi2 vs Houdini 6.03: +12 -2 = 86 [0.550]
Elo difference: 34.9 +/- 24.9, LOS: 99.6 %, Drawratio: 86.0 %

100 of 100 games finished.

Tournament duration: about 1.5 days

«It seems quite apparent that negative contempt make Houdini capable to put up more resistance, against stronger opposition»

Any comment would be appreciated.

We have also found that using negative Contempt reduces the elo deficit for Komodo against Stockfish, perhaps even by more than it should based on pure math. I suspect that Komodo is relatively stronger in the endgame, so negative Contempt tends to produce more trades and more endgames.

[Nordlandia]

In case SF Dev against Komodo, which value is good starting value?

-13 or -20 maybe !?