Scores from white's POV or from engine's POV?

[Tord Romstad]

The desktop version of my program includes a user preference for the sign of the engine's evaluations: The scores can be displayed either from white's point of view or from the engine's point of view. In the handheld version of my program, I hesitate to make this optional: The interface very easily becomes too cluttered on the tiny screen if there are too many options, and I try to act as if there were a small tax on every new option I introduce.

Therefore, I ask: What sign convention do you prefer for display of engine evaluations?

Tord

[Bill Rogers]

My program gives a positive score for the side to move.
Bill

[hgm]

Virtually all WinBoard engines report score from their own POV, (Although there is no official recommendation in Tim Mann's description of the protocol, which IMO shoud be taken as to imply: "Do as GNU Chess"), and WinBoard would display that as such. Since Winboard_x there are options to flip the black engine's score, but this is only to discipline rogue engines, and not to allow other displaying of the score: if you would use it for the latter purpose it would wreck the score-based adjudication, and the engine that was winning would be adjudicated to lose.

[sje]

I have always used the convention of the score being reported from the side on the move, a.k.a., "active color". So this would be the engine POV at the root.

In all of my programs, pretty much everything is symmetric with respect to color. If there were a White POV score reporting, it would be the only non-symmetric scoring item in the code and would stand out like a sore thumb.

[Stan Arts]

Viewing the engine as a virtual player, it seems right to have it's score from it's own view, helps add personality.

For analysis you'd probably prefer the white black thing. So perhaps you could go on that, depending if it will be used most for play or analysis.

Stan

[Edmund]

I think the engine should return scores from their own pov. It is the most natural result considering most engines are alpha beta searchers.

Anyway I think + - for white black respectively is easier to analyse and should therefor be printed by the gui. This is especially true, if you replay the game afterwards, where you don't like the sign of the score jumping all the time.

Nevertheless one needs a specification for engine and gui to communicate with each other properly. One way would be for the gui to at initialisation present the engine with a testposition for finding its way of reporting scores.

Reporting Mating scores is another issue.

[Tord Romstad]

Viewing the engine as a virtual player, it seems right to have it's score from it's own view, helps add personality.

For analysis you'd probably prefer the white black thing. So perhaps you could go on that, depending if it will be used most for play or analysis.

Users are not a homogeneous group, of course -- some of them will mostly be analysing, some of them will mostly be playing, and some of them will do both equally much. I have no way of knowing what most users prefer, that's why I made the poll.

I personally much prefer to always see all scores from the engine's point of view, but I have no problems accepting a democratic decision in this case: I can always compile a special version for myself which displays scores the way I like.

Tord

[Tord Romstad]

I think the engine should return scores from their own pov. It is the most natural result considering most engines are alpha beta searchers.

Anyway I think + - for white black respectively is easier to analyse and should therefor be printed by the gui. This is especially true, if you replay the game afterwards, where you don't like the sign of the score jumping all the time.

Nevertheless one needs a specification for engine and gui to communicate with each other properly.

I don't understand why the topic of what how the engine should output the scores comes up in threads like this -- I can't see how it has any relevance whatsoever in this discussion. It's a completely internal design decision which has no visible consequence to the user. Nobody will ever even see what signs of the scores from my engine have, unless they look at my source code.

Which reminds me, I'm a bit disappointed that only programmers respond in this thread. It is not our opinions which count. I hope not everyone who has voted is a programmer.

One way would be for the gui to at initialisation present the engine with a testposition for finding its way of reporting scores.

That would be immensely silly. A protocol which does not specify the sign of the score is badly misdesigned. What convention the protocol specifies doesn't matter, but there must be a convention.

In my case, of course, there's no need to do any guesswork or write a formal protocol specification, as I write both the engine and the GUI.

Tord

[bob]

I have always used the convention of the score being reported from the side on the move, a.k.a., "active color". So this would be the engine POV at the root.

In all of my programs, pretty much everything is symmetric with respect to color. If there were a White POV score reporting, it would be the only non-symmetric scoring item in the code and would stand out like a sore thumb.

You don't have to do this in the search. I have a simple function that displays POV scores by converting to +=good for white at the point of display. So unless you dig into the physical code where I call a small function to actually format/display the score, you don't see any asymmetry at all anywhere...

[bob]

Viewing the engine as a virtual player, it seems right to have it's score from it's own view, helps add personality.

For analysis you'd probably prefer the white black thing. So perhaps you could go on that, depending if it will be used most for play or analysis.

Stan

Now there's an idea. And to take it one step further, when pondering, reverse this again so that you display it from the opponent's point of view.

Then _nobody_ could understand the scores...