Observation: V ) relationships characterization SimDB

[bonnarel at alinda.u-strasbg.fr]

  HI All,

    I go on with the compilation of the Obs DM task group discussion.
I include herein the discussion of the relationships between IVOA Char and
simDB

Regards
François

--------------------------------------------------------------------------------
*Step 1 FB's summmary of a meeting between GL, ML, and FB on this topic
-------------------------------------------------------------------------------
Mireille and I discussed this with Gerard just after Trieste.
RELATIONSHIPS with SimDB
A bit of datamodel theory  to begin - another Posting to detail 
this (Mireille?, Gerard?)
- At top we have Domain or Analysis data model: it is a very general 
model for concepts not intended to be implemented in practice.
  - The "Logical model", derived from domain model is in UML. SimDB and 
IVOA char belong to this category
- The physical model is in practice an xml schema (char one eg)or a 
relational schema...
The discussion with Gerard showed that we could reconcile the 
characterization for theoretical data and the one for observations at 
the Domain level.
At the Logical level we may be much closer tan we are today...
Because SimDB consider characterization for "properties" of objects in 
Object collections we have to consider how we can apply this "object 
collection" concepts in the case of observation.
Actually objects can match pixels and collection of pixels match our 
"datasets"...
Each of the simDB properties match with our "axes".
SimDB wouldn't need  Resolution, sampling and error according to Gerard 
(only Value) but it can be seen has a peculiar case (and actually I 
(FB) am not totally convinced they don’t need it).
SimDB can evoluate the CharacterisationType content towards "reference 
value", "bounds", "support", "distribution", much closer to our IVOA 
char four levels.
Reference value matches the "Location" of our IVOA char and 
distribution will be matching the level 4 or sensitivity
Generally the IVOA characterization data model is dealing with "a 
priori" distributions (something like "extended flat fields") while the 
main interest of simDB is in "a posteriori" distributions. But Actual 
value distributions , and support can be estimated for Observational 
data as well. So the flavor of the characterization (a priori or a 
posteriori) could be usefull as well...
Actually in SimDB there is a mediator class between ObjectCollections 
and Properties. This is called ValueAssignment and contains values, 
resolution sampling and error. We can look at this as the starting 
point of our matching between the two datamodel Char classes. This 
leads to the conclusion that SimDB is probably more "Property first" in 
the sense of IVOA char than Axis first (see IVOA reference document to 
see what this means, the current existing serialisations are all Axis 
first. *). It is probably possible to better match both sides by 
comparing it to “Property first” implementations of IVOA char.
All this shows that SimdB characterization and IVOA characterization 
for observations could be inferred from the same domain model char 
which remain to be defined in detail ....
Nota: An axis first serialisation is something like
     <charac>
             <spatialAxis>
                     <coverage>
                     <resolution>
                     <sampling>

             </spatialAxis>
             <timeAxis>
                    <coverage>
                             


                          Etc 
.
A “property first” IVOA char would be:
                        <char>
                            <coverage>
                                <axis1>


                                 <axis2>
                                        

..
                            </coverage>

                            <resolution>
                                 



--------------------------------------------------------------------------------
*Step 2 Igor's point
--------------------------------------------------------------------------------
STC and Char must be connected as they are. And, moreover, we have to 
find a way to interconnect with SimDB at least to use the same "object 
repository" if I can call it like this (Gerard, I know you won't like 
this statement).
--------------------------------------------------------------------------------
*Step 3 Gerard Lemson's answer
--------------------------------------------------------------------------------
As you address me explicitly, an answer to your suggestion as to what I 
will like or not.

> 2) STC and Char must be connected as they are. And, moreover, we have 
> to find a way to interconnect with SimDB at least to use the same 
> "object repository" if I can call it like this (Gerard, I know you 
> won't like this statement).
>
I can not dislike your statement, as I do not understand it.
I have no beef with SimDB connecting to characterisation.
It already does, though not in the naive way many people in the IVOA 
think about data modelling: use the XML element form the other model in 
yours.
I have not had any time to expand on the work on characterisaiton I 
presented last year in Bejing, apart from the discussions I have had 
with some of you. Those seemed to lead to the concensus that we need a 
beter understanding of what characterisation is in the generic sense, 
not just in its application to observations (the current 
Characterisation DM), before we can insists on particular usage in 
other models.
I am perfectly willing to discuss this further, in Baltimore for example.

As to STC, there is no need, therefore no place for it in SimDB.
But if there was, I doubt I could use it "naively" in its current form.
-------------------------------------------------------------------------------
*Step 4 Arnold
--------------------------------------------------------------------------------
> As to STC, there is no need, therefore no place for it in SimDB.
> But if there was, I doubt I could use it "naively" in its current form.

I don't agree - maybe not surprisingly.
STC has provisions for simulation data in its coordinate metadata 
definition and it would be very helpful if such metadata (which are 
actually a form of characterization) were to be included.
-------------------------------------------------------------------------------
*Step 5 FB comments
-------------------------------------------------------------------------------
    A couple of things
       - STC is not a "form of characterization". It is a model for coordinate
systems, coordinates and coordinates area. However IVOA-char MAKES USE of STC.
That's the point.
       - The relationship between char and SimDB has to be reconsidered
when char (version 2) will contain an extended level 4 described by "moments".
       - We have to survey all the use cases where we actually need to request
simulations by "something like IVOA char" on the properties at the snapshot
level. Igor did it for Galmer already long ago
But, when is that useful in general ? (I am not joking, I really want some
inputs on that)
--------------------------------------------------------------------------------