Design of MANGO errors: feedback expected

[Laurent Michel]

Le 01/04/2025 à 16:23, CresitelloDittmar, Mark a écrit :
> Laurent/Mireille,
> 
> I was literally on my way to submit a comment on this!
> 
> I think your proposal is the best approach at this point.. separate the mango:PropertyError from meas:Uncertainty.
> Unless we, as a working group, have a consensus on how to execute these situations in the models, I think it's better to avoid 
> the conflicts which arise from having these stem from the same base.

OK, I'll do this in the next PR

> 
> re:
> • CON: in our experience, cross model references are a burden for clients and particularly for annoters
>      • CON: The MIVOT annotation of some Meas error classes may be tricky to interpret for the clients (attribute values in arrays).
> 
> This is worth discussing in the group (Running Meeting?)...

Yes of course.

> I read this as "you've changed the modeling to avoid issues with the annotation".

I do not say this. Any pattern can be annotated.

I'm just saying that:
- in my perspective, stackholders will prefer to deal with ellipses defined as
      {majorAxis, minorAxis, angle} (one role per item)
   rather than
       [x, y, z] (one role per item rank)
   this point do no relate with the annotation.
- There is no issue with the annotation, but let's mix model
   components only when necessary to keep the annotation process simple.

Laurent
> 
> Mark
> 
> 
> 
> 
> On Tue, Apr 1, 2025 at 9:27 AM Laurent Michel via dm <dm at ivoa.net <mailto:dm at ivoa.net>> wrote:
> 
>     Hello DM,
> 
>     (cf issue https://github.com/ivoa-std/MANGO/issues/60 <https://github.com/ivoa-std/MANGO/issues/60>)
> 
>     We would like get feedback from the WG about the MANGO error.
> 
>     In the MANGO use cases, we have the cross-match case which requires the errors to come with statistical parameters
>     (distribution and confidence level).
> 
>     For this purpose, Mango has an abstract. datatype  with these parameters (mango:PropertyError) from which some concrete
>     error types (e.g. ellipse …) derive.
>          • Symmertrical error  1D and 2D
>          • ASymmertrical error  1D
>          • Ellipse error
> 
>     These 4 classes have counterparts in Meas, but they have more handy attributes.
>          • Individual roles for each attribute (majorAxis, minorAxis, angle) instead of value arrays [x, x, x].
> 
>     These classes also have specific names in order to prevent conflicts with measure classes.
> 
>     The question is to decide whether the abstract mango:PropertyError should or not derive from the astract meas:Uncertainty?
>          • PRO: any Measurement error type could then be reused by Mango properties
>          • CON: this will make Mango providing different error types form the same purpose (meas:ellipse or mango:PErrorEllipse)
>          • CON: in our experience, cross model references are a burden for clients and particularly for annoters
>          • CON: The MIVOT annotation of some Meas error classes may be tricky to interpret for the clients (attribute values in
>     arrays).
> 
>     We suggest not deriving mango:PropertyError from meas:Uncertainty.
> 
>     We would like to get the group (des)agreement before opening the corresponding GIT PR
> 
>     Mireille and Laurent
> 

--
English version: https: //www.deepl.com/translator
-- 
jesuischarlie/Tunis/Paris/Bruxelles/Berlin

Laurent Michel
SSC XMM-Newton
Tél : +33 (0)3 68 85 24 37
Fax : +33 (0)3 )3 68 85 24 32
Université de Strasbourg <http://www.unistra.fr>
Observatoire Astronomique
11 Rue de l'Université
F - 67200 Strasbourg