[Passband] a useful self-contained model?

[Martin Hill]

David Berry wrote:

> Martin,
>        Just got back to work after a while away (hope you all had a
> good time in Boston!) and noticed this. I like the "divide and conquer"
> notion - define general purpose independent component models and then
> aggregate them into more specific-purpose higher level models.

I guess we need to be careful about terms: I don't think of Passband as a 
'general purpose' model, rather as a very specific model of how to describe 
passbands and only passbands.

> So a
> passband would be used as a component in a more specific higher level
> model. To be consistent, though, should not the same approach be
> used to design the passband model itself? 

Yes I agree; if there are smaller components that we can model separately, then 
we mark these as placeholders, define what we need to know about them from the 
passbands point of view (ie define an interface for them), and model them later.

> That is, a passband could be
> seen as a specific use of general purpose component models.
>
> A passband is basically a recipe for transforming an input scalar
> value (a spectral axis value) into an output scalar value (transmission).
> Now we have the general purpose Mapping model which can be used to
> describe such numerical transformations, so why not use a Mapping as a
> component of a passband? Compound passbands are then simply catered for by
> using a SeriesMap/CompositeMap as the Mapping.
(snip)

Ah, that's different :-) My feeling is that the DM group models what we need to 
know, and *only* what we *need* to know.  If implementors decide that they want 
to create a passband out of Frames and Mappings that's fine; but if someone 
wants to implement it using a graph, or a formulae, or a simple central fravergy 
+ width, then that should be fine too.

I want to be sure that the model describes our data and how it interacts; when 
we start building lots of 'generalised' submodels in, then we can not only end 
up obscuring our model, but also forcing our model to fit the generalised 
components rather than our actual data.

If you would like to model a Passband built from Frames & Mappings (but 
conforming to the Passband interface) I would be very interested.

The Passband model page is here: 
http://www.ivoa.net/twiki/bin/view/IVOA/DataModelPassband

but I haven't updated it yet following everyone's comments.  Soon, real soon, 
honest...

Cheers,

Martin

-- 
Martin Hill
www.mchill.net
+44 7901 55 24 66