Stockfish 9 is OUT!

[corres]

[quote="Michel"]

Can you back that up? contempt=20 was shown to be neutral in self play.

[/quote]

The results of self play is rather dubious because self play makes disappear the effect of small difference and contempt = 20 is a same thing.

[syzygy]

I am afraid a debate on contempt is a worthless thing because every kind of contempt decrease the power of a well-tuned engine. Against an engine having the same power of Stockfish, Stockfish 9 with contempt = 20 is about 5 Elo weaker then with contempt = 0. The real difference depends on the power of enemy and the type of opening.

Can you back that up? contempt=20 was shown to be neutral in self play.

Evidence that contempt hurts self play would indeed be very welcome.

[Nordlandia]

Kaufman said default contempt value of 10 is good in K selfplay matches.

How about SF contempt of 10 in selfplay matches?

[corres]

[quote="Nordlandia"]

Kaufman said default contempt value of 10 is good in K selfplay matches.
How about SF contempt of 10 in selfplay matches?

[/quote]

The effect of contempt depend on the given engine, the power of enemies and the opening book used.
For e.g. the most best to play TCEC Stage 1 with contempt = 20-40
and TCEC final with contempt = 0.
It is pity but Stockfish 8 (dev) - in contrast to its main rival - had no contempt during the last TCEC.

[tmokonen]

They are available now.

https://stockfish.s3.amazonaws.com/stockfish-9-mac.zip

[Uri Blass]

Can you back that up? contempt=20 was shown to be neutral in self play.

The results of self play is rather dubious because self play makes disappear the effect of small difference and contempt = 20 is a same thing.

I do not understand.

If there is no proof that contempt=20 is counterproductive in self-play then
common sense tell me to use it against opponents with equal strength.

Maybe contempt is not counterproductive in self play because of a bug but until you fix that bug the default choice should be to use contempt not only against weaker opponents.

[AdminX]

Nice!

Code: Select all

10 Minutes/Game  2018

                                   1                    2                    3                    
1   Stockfish 9 64 BMI2 [3]        ******************** 1½½½½1½½½½½½1½1½1½½½ ½1½½½½1½½111½½0½½½½½  24.5/40
2   Houdini 6.03 Pro x64-pext [3]  0½½½½0½½½½½½0½0½0½½½ ******************** ½½½½½1½½1½½11½½½0½1½  19.5/40
3   Komodo 11.2.2 64-bit [3]       ½0½½½½0½½000½½1½½½½½ ½½½½½0½½0½½00½½½1½0½ ********************  16.0/40

[corres]

If you make a minimal modification on an engine (like adding some contempt) and run an engine-engine match between the unmodified and modified engines the change in result of match maybe different then you manage engine-engine matches between the unmodified engine and other strong engine, and between modified engine and this other engine, respectively.
This is the main issue if one uses self play only to develop his engine.

[Ovyron]

You are just misinterpreting a single position.

No, I'm misinterpreting all positions where black has the edge and Houdini seems to switch Contempt for the black side.

I guess you'd need to get the engine to test and see what I claim, otherwise, you're just comparing reality with your imagined suppositions and assumptions of how the engine would behave if what the manual claims is true.

Either it's a glitch or an outdated manual, in either case, Houdini is detecting the side with the advantage and applying Contempt to this side, and implementing this for Stockfish would be not trivial. Implementing what the manual says it does would be trivial, but how relevant is it if the manual is wrong?

Coming from the Video Game field, this is a common thing.

In the Pokémon games, there's a thing called a "Critical Hit", that is an event that happens rarely that allows your Pokémon to inflict higher damage to the enemy.

They implemented the move "Focus Energy", that was supposed to increase the chance of Critical Hits happening 4 times. Instead, after using it the chance of Critical Hits happening would be divided by 4, making this an useless move.

But what was the manual saying? What were the guides you could get from the game claiming that Focus Energy did? They all said it would cause you to inflict Critical Hits 4 times more often. This was the official behavior supposed to happen, but players would see in reality Critical Hits happening 4 times less often.

You'd imagine that if they implemented this move, that they'd test it carefully, and that the game should behave like the manual claims, but the reality was opposite from the intended behavior.

Houdini is showing this behavior everywhere, and it's very common to see 0.60s scores going down to -0.09 scores if all its variations fail, and what remains detects that it's white fighting for the draw. This wasn't just observed in a single position.