Eelco de Groot wrote:CFish I'm sure has a different solution by now from Ronald (I have not checked)
Cfish has the same solution as Houdini (as far as I understand Houdini's solution).
In analysis mode, contempt is forced to zero unless the user has checked the "Analysis Contempt" option. If that option has been checked, contempt will be from white's point of view (so set it to a negative value to give black positive contempt).
In gameplay mode, contempt is for the side to move. "Analysis Contempt" is ignored.
Komodo has a similar but slightly different solution. Instead of "Analysis Contempt", it has "White Contempt". If my understanding is correct, the only difference is that White Contempt also forces contempt to be from white's point of view in gameplay mode.
A practical disadvantage of this solution is that the user has to set "Analysis Contempt" to negative if he/she wants to analyze with positive contempt from black's point of view.
The only solution seems to be standardization so that the GUI can set the appropriate contempt value. For this reason it would be nice if SF was on board...
Analysis Contempt is just a checkbox (also in Houdini if my understanding is correct). But yes, the user has to set Contempt to a negative value to analyze with positive contempt for black.
Rybka seems to use the solution you are thinking of: a slider for contempt during game play and another slider for contempt in analysis.
I fully agree that none of these solutions is ideal. What is needed is a fixed solution mandated by UCI (or some improved protocol) which can then be supported in an intuitive way by GUIs. Since nobody seems to be working on UCI, the second best solution is agreement among the authors of top engines on one solution. Which one is less important (as long as it makes sense).
Analysis Contempt
---------------------
Off
White
Black
This is certainly the best UCI-based solution by far!
With this solution there is a lot less need for special GUI support.
Why not just have the user set those contempts individually?
Contempt [number][up and down arrows to modify] <-Default 20
Analysis Contempt [number][up and down arrows to modify]<-Default 0
It saves space as you don't need the extra drop down list, and allows the user to use different combinations instead of being limited having Contempt and An Contempt equal.
Analysis Contempt
---------------------
Off
White
Black
This is certainly the best UCI-based solution by far!
With this solution there is a lot less need for special GUI support.
Why not just have the user set those contempts individually?
Contempt [number][up and down arrows to modify] <-Default 20
Analysis Contempt [number][up and down arrows to modify]<-Default 0
It saves space as you don't need the extra drop down list, and allows the user to use different combinations instead of being limited having Contempt and An Contempt equal.
But what would be the behaviour of the analysis contempt value?
In game play things are easy. The engine is playing with white or with black but not with both. Contempt should be from the engine's point of view.
In analysis things are less straightforward. Typically, the user will make moves back and forth while the engine is searching. So the "engine's point of view" is not well defined. The simple solution would be to let contempt be from white's point of view, but for a user who has not studied the manual carefully this will not be very clear. The combo box, on the other hand, seems to be completely self explanatory (but perhaps I am overly optimistic here ).
syzygy wrote:But what would be the behaviour of the analysis contempt value?
0 - > Same as combo box Off
Posititive - > Same as combo box White, for a Contempt of this number
Negative - > Same as combo box Black, for a Contempt of this number
syzygy wrote:for a user who has not studied the manual carefully this will not be very clear. The combo box, on the other hand, seems to be completely self explanatory (but perhaps I am overly optimistic here ).
I see.
Indeed, this was a problem the Komodo's team failed at, with the current Contempt implementation being really confusing for new users, and the "you have to create a Komodo instance in your GUI for each side you want to analyze" being absent in the documentation.
Despite all the time that it took them to release S9, in this aspect it seems a bit rushed
syzygy wrote:But what would be the behaviour of the analysis contempt value?
0 - > Same as combo box Off
Posititive - > Same as combo box White, for a Contempt of this number
Negative - > Same as combo box Black, for a Contempt of this number
Yes, that is very sensible. (I think Rybka has this. And I thank Uri for this confirmation.) But it has to be explained to the user how it works.
syzygy wrote:for a user who has not studied the manual carefully this will not be very clear. The combo box, on the other hand, seems to be completely self explanatory (but perhaps I am overly optimistic here ).
I see.
Indeed, this was a problem the Komodo's team failed at, with the current Contempt implementation being really confusing for new users
Their implementation solves the problem, but for a user who (like me for quite some time) does not even realise that there is a potential problem when using contempt during analysis it must be a bit puzzling at first.
Despite all the time that it took them to release S9, in this aspect it seems a bit rushed
Yes, even Komodo's or Houdini's (or Rybka's) solution would be better than the current implementation. Unfortunately, many developers are convinced that the current sign-switching-on-every-move contempt is great.