SED serialization

[Doug Tody]

Hi Ivo -

Perhaps it is better to look at it the other way around?  A simple
1D spectrum is a special case of a multi-segment SED with associated
metadata.  SED is clearly the more general case.  With this approach,
we still have the general case of a SED as described below, plus the
special case of a 1D spectrum (the Spectrum model), since it is so common.
The most powerful applications should be able to deal with either, but
many simpler applications may prefer to work with Spectrum, which also
covers most spectral survey data.

The SED model being proposed is not merely a multi-segment spectrum,
as it also needs general metadata describing the Dataset as a whole
(DataID, Curation, Target, Char, etc.), plus probably a table giving a
uniform view of the SED as a whole, with each segment reduced to a few
values with uniform units.  Then follows the multi-segment part as in
our original SED model, where an individual segment may be essentially
a complete observation in itself.  Since for SED the segments can come
from independent observations with different instruments, in the general
case it really is more than just a mere multi-segment spectrum.

	- Doug


On Tue, 13 Feb 2007, Ivo Busko wrote:

> Doug Tody wrote:
>
> <snip>
>
> > Not greatly different, but the main point is that a SED may be worth
> > considering as a new type of astronomical data (a primary type) and not
> > just a fancy Spectrum.  That is, a SED could be a top level Dataset type
> > (like Image Spectrum, TimeSeries) which is multi-segment, with an overall
> > Characterization and Identification, and homogeneous units to summarize
> > all the segments.  The individual segments can be instances of Spectrum,
> > TimeSeries, or individual photometry points.  Auxiliary metadata might
> > be required to describe how the SED was computed, especially in the case
> > of dynamically computed SEDs, where one might want to include images
> > cutouts for photometry from multiband imagers and so forth.
> >
> > Spectral associations are slightly different, being informal
> > associations of primary datasets rather than a single physical dataset.
> > A multi-segment Spectrum could be considered an Association or a special
> > case of a SED; both could be valid.
> >
> > 	- Doug
>
> Doug,
>
> I still see a benefit in having a SED to be a special case of Spectrum.
> Being a subclass of Spectrum means that any application code that can
> handle a Spectrum instance, will be able to handle a SED instance
> transparently. It might not be able to provide everything that a
> SED-aware application is, but it won't crash or refuse to handle the SED
> data.
>
> Maybe I am influenced by the design choices we made when writing the
> Specview application, but the "SED as a special type of Spectrum" design
> paradigm showed to be very useful in that case.
>
> -Ivo
>
>
>