Characterization draft Note

Wed Dec 21 13:41:43 PST 2005

Dear Pierre,
   Just before the Christmas break a few remarks and answers to your mail

>From owner-dm at eso.org  Mon Dec  5 14:43:23 2005
>From: Pierre Didelon >pdidelon at cea.fr<
>Organization: CEA
>To: dm at ivoa.net
>Subject: Re: Characterization draft Note
>
>
>
>Francois Bonnarel wrote:
>< Dear data modellers
><    This mail to notice you that Jonathan, Mireille, Alberto, Anita, Igor
>< and me have produced a draft IVOA note now available on
>< http://www.ivoa.net/twiki/bin/view/IVOA/IvoaDataModel#Related_Documents
>< 
><   We will take into account your comments before publishing the Note
>< and are preparing the draft and implementation
>< 
>
>First of all
>before any remarks, comments and complains ;-)
>I really want to congratulate you for this tremendous work
>with a such complete and usefull resulting document. :-)
>I was no longer expecting this remembering all the discussions
>and different ways open, each time a new project
>oe point of view was incorporated in IVOA.
>
  Thanks but the proof of the pudding is the pudding itself.
we are working on implementations now...
>
>:
>Nevertheless, ( as always!? :-( ) I have some comments
>and some questions to clarify my limited understanding ;-)
>
># Some general remarks :
>
>- property/axis symetry
>  It is not clear if the proposed serialization has to be adopted,
>if it is an example or the "actual consensus/agreement".
>
>Until reaching the serialization, the respective points of view between
>axis and property seems to be equivalent.
>In such a way that confuses me a lot, it was hard for me to see
>before this point, how axis information would be handle
>and which is its link with properties. In other words,
>it means that at least the figure 3 needs to be explain a little bit more,
>mainly the CharacterisationAxis (class?) and its relation with AxisFrame.
>CharacterisationAxis is derived_from/uses/... AxisFrame? ...
   Ok, The model and its UML representation is really open to both
organisations : it is a graph who could rely on a matricial data 
structure. The XML is hierachichal in essence so we had to choose one
point of view. This will be enhanced _in the document not only in Fig 3 but also in 3.5
and 4.1. The CharacterizationAxis element in the XML is not the serialisation of
a class but a container for the grouping of the Properties by the
CharacterizationAxis parameter.
>
>I was strengthen to read the  example (with >axisFrame<).
>However 3.5 mention the possibility of  two (at least) different XML Schema.
>I did not feel very comfortable with this.  Does it means that it will
>results in characterisationByPropoerty.xsd and characterisationByAxis.xsd
>different files?
   Yes, but in the document there is only one implementation. We simply did
not try to build the other one. If the properties for the different axes 
are separable a property first serialisation will only be an other kind of
grouping of the same atomic elements. But if there is intertwinning (or
non-separabilty) of the axes the sensitivity function (layer 4 of our 
coverage model) will be f(ra,dec,lambda, t) and it will not be possible
to write it as fp(ra,dec)*fw(lambda)*ft(t). Bounds and locations will
always be defined as extreme values and typical values on each of the
axes, but the support will be also inseparable like the sensiivity and this
makes the real difference. Hence, My personal guess is that a property first 
serialisation will be enhancing level 4 and 3 of the coverage instead of
level 1 and 2. That means it will be for more complex use cases, those plannes
for version 2.

>
>- concerning Accuracy
>I have a pb with the diff level and the correponding links with the frame!
>If Error and ErrorBounds can clearly be related to an AxisFrame,
>and so, easly included in the metadata spicifying an axis
>the ErrorMap seems to be more related to the data themself
>and then his relation with the axis is not so obvious.
>What would be an ErrorMap for a 1D spectra, a 2D images time serie....
>A small example would be welcome here p.22/23
>
   Some programs give associated error maps with the data. They may be
packaged with the data themselves in Fits Extension files (see cfhtls 
for example) or as additional vectorial  columns in FITS table for 
spectra or IFU data (see Euro 3D). Something will be added in the 
document abot that.
   By the way, we have to be carefull: everything in Characterziation is
related to the data, and could be derived from the data. Although it is
not the way it will be generally implemented (for obvious simplification 
reasons) one should start characterizing a data set by the Coverage 
sensitivity function, an ErrorMap, and full description of resolution
and sampling as "mathematical distributions". (that is full layer 4)
   The other layers are progressive simplifications and approximations
of this and in practice we will very often (maybe always!) get only these 
simplified layers.
  
># more specific remarks concerning the text form or details :
>
>- concerning 2.5.2 Resolution and Sampling Precision
>  I think a prelimineray remarks concerning their relations would be nice.
>So a quitely different redaction like the following would perhaps be clearer,
>it is only reformating except some proposed additions enclosed in []
>
>2.5.2 Resolution and Sampling Precision
>  The concepts of resolution and sampling precision (or pixelization) are related.
>If the sampling precision period is small compared to the resolution, the
>knowledge of a single [observed] data value is limited by the resolution [(oversampling)].
>If the sampling precision period is coarse compared to the resolution, knowledge of a single
>data value is limited by the sampling[(undersampling)].
>  Resolution  Ultimately resolution describes
>	...
>	Sampling Sampling (or pixelization or precision or quantization) describes
>the truncation of data values as part of the data acquisition [(]or data processing
>[via re-sampling)]. If the mapping of...
>
   OK will be done!!!
>- in 2.6 The layers of description
>Concerning
>"We generalise the four levels described for Coverage to the Resolution
>and SamplingPrecision concepts."
>It would be nice to have an example with eventually a figure as illustration.
>
   For a n-D spectrum the Resolution and sampling is wavelength dependant
so the concept of variation map is valid. Something will be added in the*
document.
>- in figure 7 p.28 independantStatus doesn't appear,
>even though it is mentionned in 3.2 p.20 and prisent in serialisations,
>XML in 5 and VOTable in 6
>
   OK 
>Some more details, but not very important...
>Enough for today perhaps ;-)
>SY
>PIr Didelon
>
>

François

=====================================================================
Francois   Bonnarel               Observatoire Astronomique de Strasbourg
CDS (Centre de donnees          11, rue de l'Universite
astronomiques de Strasbourg)    F--67000 Strasbourg (France)

Tel: +33-(0)3 90 24 24 11       WWW: http://cdsweb.u-strasbg.fr/people/fb.html
Fax: +33-(0)3 90 24 24 25       E-mail: bonnarel at astro.u-strasbg.fr
---------------------------------------------------------------------