[QUANTITY] Review of Quantity suggestions

[DIDELON Pierre]

Hi Jonathan and all data modelers not yet in hollidays,

The synthesis document you posted clarify the diff. possibilities 
and options in Quantity modelling. But in addition to the structural 
possible options, it implicitly covers different purposes and usages,
so it is confusing me a lot. I try in the following to explain my point
of view and clarify possible options. I hope that it is not too wordy.
If your are not interested by argumentation but only with conclusion
go directly to the end of the message dealing with "needs".


my point of view concerning Quantity :

As I already stressed in a previous mail exchange with Ray
(http://www.ivoa.net/forum/dm/0087.htm), I hope that Quantity 
is not a tentative to model astronomical data in general but a basic
concept to qualify numbers and values.
In my mind there were a clear difference between data (observational
or anyelse kind of) and quantity but obviously it is not so clear
when models contain things like format, size, dim or others informations
on data structure.

I think that we must not mix several domains which are distinct
albeit dealing with data.
- structure
- storage representation & format
- digitalization implied by measurement and related to resolution, errors...
- measurement meaning and analysis purpose related to observables, errors...
- data representation in a ref. system linked to transformations
- visualisation

A mail of gerard lemson (http://www.ivoa.net/forum/dm/0070.htm)
try to clarify some aspects of this point.

Values, units and errors :

The common minimum  between all views concern the structural
relations between values, units and errors, with optionally others
(resolution...), and oftenly seems to be an attempt to frooze one schema.

I am not sure that we can frooze the relation between value and error
in only one general structure. It is certainly not obvious because
different data can have different relation to errors.
Do interferometry data, like visibility maps, have the same relation 
to error, than photometric data or others?

Errors can be related to one parameter (magnitude) but also to parameter
pair (RA,DEC) or more.
Moreover errors can be factorized, and not directly link to
only one value, at the level of vectors (columns,spectra...)
cube or even a complete dataset, for example resolution related to
one overall survey.

The relation between values and errors depends on the data system coor,
the observing mode and perhaps even on the analysis purposes, 
so it seems to me unrealistic to model everything with one very
restricted froozen model. Generic containers like VOTable or even HDX, 
or their constitutiv elements seems more appropriate to handle this.
The advantage of the VOTable elements Param and Field is that they
share almost completly the same structure, and so can be deduced from
a more general parent class, which can be used to describe then
individual numbers as well as arrays/columns and eventually cube or more.

For me the problem is not to fix or frooze relations, between values and 
errors (but also others parameters like resolution), but to formalize
and define a way to link them inside a data structure.

Needs :

I think that, accordingly with Clive (http://www.ivoa.net/forum/dm/0086.htm)
we must focus on precise domain and needs for quantity. 

For which purposes do we need Quantity, and what domain do we want to model;
- Value; add info to fully define their usage and transformations,
- Observation mode and conditions; formalize metadata to facilitate data 
retrieving and selection
- global data structure; hopefully not.

Essential domains in VO interoperability are to my point of view,
- data conversion, essential for data comparison (->units)
- data quality, essential for data selection, this domain deals with 
all parameters which can be used to characterise data quality like 
errors, range, resolution, S/N...

So a basic simple reusable bloc, eventually derived from Param/Field,
would be a usefull tool to handle these informations uniformly.

Finally I want to stress that we perhaps needs some use cases,
to clarify our minds (at least mine) and the quantity purpose.

Cordially,
Pierre
-------------------------------------------------------------------------------
DIDELON                               e-mail : pdidelon at cea.fr
CEA SACLAY - Service d'Astrophysique  W3 : http://www-dapnia.cea.fr/Sap/
91191 Gif-Sur-Yvette Cedex            Phone : 33 (0)1 69 08 58 89
-------------------------------------------------------------------------------