[Observation] relation to Dataset

[Douglas Tody]

On Thu, 21 Nov 2013, CresitelloDittmar, Mark wrote:

> Doug,
>
> You're right.. I didn't word that right.
> In the model, multiple datasets may have the same obs_id.
> I was trying to go the other way.. and I may still have this wrong, but it
> says (pg 16)
>
> "The form of the obs_id string is up to the data provider so long as it
> uniquely identifies an observation within the archive."
>
> which I'm taking to mean that a dataset may be associated with only 1
> observation.  That's what was referring to.

These are my words you quote from the spec; sorry if it wasn't more
clear.  Obs_ID *does* uniquely identify an *observation*; each
observation requires a unique Obs_ID.  But multiple table rows in ObsTAP
may share the same Obs_ID, indicating that they are from the same
observation.

In ObsCore/ObsTAP, records may describe either observations, or data
products (datasets).  A pure observation record is not a dataset, it
merely describes an observation (no acref, no data product to retrieve).
If the record describes a data product it is a dataset, a science data
product of some sort.  A data product could be something well defined
like an image, or it could be a "package" file containing all the
standard data products for an observation.  It is possible to combine
any of these.  Once we have data links it will also be possible to
merely describe an observation and link to external data products, or
also expose selected data products.

So ObsCore can describe observations as well as data products (datasets)
associated with an observation.  Without Provenance or something
comparable it is hard to tell how a given dataset is derived from other
data products.

 	- Doug


> Mark
>
>
>
>
> On Thu, Nov 21, 2013 at 6:31 PM, Douglas Tody <dtody at nrao.edu> wrote:
>
>> On Thu, 21 Nov 2013, CresitelloDittmar, Mark wrote:
>>
>>
>>> I don't think that a Dataset should have a bi-directional relation to the
>>> full Observation(s) as I noted at the head of this thread, but should
>>>   a) have an association back to components of the Observation (
>>> ObsConfig, Proposal ) which become part of the Dataset 'provenance'.
>>>       (which is what I think Arnold was saying in the other thread).
>>>   b) have metadata identifying the relevant Observation(s) comprising
>>> Dataset (DataID.ObservationID), as Francois notes.
>>>       but this gets tricky because ObsCore expects a singular (well
>>> unique) obs_id for each Dataset.
>>>
>>
>> I was with you up to here.  Obs_ID does *not* have to be unique for each
>> dataset - the pubDID is what has to be unique.  Multiple datasets may
>> share the same obs_id; this is an essential feature of ObsCore.
>>
>>
>