Coords/Measurement model RFC comment response

Laurent MICHEL laurent.michel at astro.unistra.fr
Tue Feb 25 10:59:10 CET 2020


Dear all,

Thank you Mark for this RFC answser.

- The new design of the space coordinates with classes better matching 
entities usually used by astronmers, is a big improvement.

- It is very valuable for the model to clearly state its limitations 
(error modeling, use cases) in the spec.

- I agree with you that the model must be validated against real data 
but that validation must not be bind to one specific annotation mechanism.

I propose to put a link to your message in the RFC pages instead of 
copying your anwsers one by one, which would be a waste of time.

As you are proposing significant changes in the model design (I speak 
for both Meas and Coord), I think we have to formally extend the RFC 
cycle on a new Wiki page with document pointers and data samples fitting 
the new model features.
Let me know whether you agree or not.

Laurent


Le 20/02/2020 à 20:11, CresitelloDittmar, Mark a écrit :
> All,
> 
> My apologies for the delay in getting things moving on these models.  I 
> have had very little time to put toward VO work since the last interop.
> 
> Below are responses to, and actions to take, based on the RFC comments 
> for the Measurement and Coordinate models.  I will be officially 
> implementing these, generating new example files, and working with the 
> Transform team to ensure their extended use case is satisfied.  
> Currently, the updates for the coords, meas, and transform models have 
> been modeled.  I needed to adjust the plan from the Netherlands slightly 
> to satisfy the integration with the Cube model.  These went to the 
> Transform folks for review, and I have a followup issue from them to 
> work still.  I'm fairly confident that this will have NO effect on the 
> coords or meas models, but will resolve the details prior to 
> re-submitting the documents.  I am currently updating the model documents.
> 
> These notes will be added to the appropriate RFC pages.
> 
> Mark
> 
> Starting from the item having the greatest impact:
> *Coords*:
> 1) AST implementation project expansion in scope to "enable the 
> description of a complete WCS ( axes, frames, mappings between ) 
> independent of the coordinates in that space".  With the current model, 
> the CoordSpace and CoordFrame are only related through the Coordinate 
> itself.
> *ACTION:* Modify model to restore an element which combines CoordSpace 
> and CoordFrame, which is, presumably then referenced by the Coordinate.
> *NOTE:* this action triggers subsequent changes in the model which are 
> in place, at least partially, to accommodate the separation.  For 
> example, with the singular access point to both CoordSpace and 
> CoordFrame information, there is no need for the Coordinate to 
> separately reference a CoordSpace axis and the Frame.  I will post an 
> outline of the logic tree for the change set to the twiki.
> *ACTION:* This will also require bringing an object from the Cube model, 
> or something similar, to either Coords or Transforms.  This object 
> associates 2 instances of this combined CoordSpace/Frame with a Mapping 
> defining the transform between them.  This is similar to objects already 
> in the cube model to define virtual axes and columns.
> 
>    2) SpaceFrame.planetaryEphem multiplicity is 1, should be [0..1]
> *ACTION:* change multiplicity to [0..1]
> 
>    3) URL's for vocabularies - should use 'http' not 'https'
> *ACTION:* make correction in document
> 
>    4) Appx C.3 - bad LaTex for TCG and TCB records.
> *ACTION:* make correction in document
> 
> *Measurement:
> *
>     *) Request for additional 'catalogue data' use case
> *RESPONSE:* The current model is based on long established use cases 
> related to Cube data;  this request is covered by the Source model 
> effort.  I will adjust the text in "Context and Scope" to clarify that 
> the scope of this version is limited to the described use cases, while 
> the larger scope is retained to inform design decisions.
> 
>     *) requirement meas.003: After reading the standard, I think I 
> understood what that means, although I'm not sure what the reason for 
> the requirement is (let alone which use case it is derived from). Let me 
> try: "Each error instance must only be referenced by a single 
> measurement." Is that what you mean? If so, why?
> *ACTION:*clarify requirement text - "A value must not be associated with 
> more than one each of systematic and statistical/random errors."
> 
>     *) Error model (several comments in this arena)
>         + Error.statError and Error.ranError - unclear what the 
> distinction is, and if it is necessary
> *ACTION:* drop <Error.ranError>  (Markus professed a mild preference to 
> keep 'stat' since Char/ObsCore have 'stat' based elements)
> *ACTION:* associated issue, correct multiplicity bug in model ( 
> Measure.error[0..1] not "*" )
>         + Included elements and their descriptions
>            o Have Bounds*D, Ellipse, Ellipsoid and CovarianceMatrix.
>               - Most other forms can be represented by CovarianceMatrix, 
> so why include?
> - However, if we assume Gaussian, then we also need the "confidence 
> level", which is not specified nor a model element.
> - Additionally, the general modeling of CovarianceMatrix is inefficient 
> at best.  Unclear if Matrix or Cell or some other Correlation element 
> might be a better approach.
> *RESPONSE:* The various flavors of error types are basic error 
> representations which occur in our data, and come directly from STC-1.  
> Forcing data providers to convert their holdings to Matrix form is not a 
> reasonable expectation.
> *ACTION:*Drop CovarianceMatrix until we have an implementation case 
> which exercises it.
>         + Distributions:
>             o "we should say *something* about assumed or non-assumed 
> distributions"
>             o "When we assume that errors are Gaussian (we probably need 
> a way to mention it explicitly)"...
> o Confidence Level: assuming Gaussian errors, should make a statement 
> about the confidence level in appropriate elements.
>                 eg: Error ellipse assuming 68% confidence level Gaussian 
> distribution unless otherwise specified.
> *ACTION:* Add text to descriptions indicating that this version of the 
> model forms a fundamental basis from which we can grow to include more 
> complex and varied use cases as they present themselves.  The current 
> model assumes Gaussian distributions with shapes defined at the 68% 
> confidence level.  The exact location and nature of the text updates is TBD.
> *ACTION:*add formal modeling of 'confidence level' to 'Next' page for 
> Measurement model.
> + Ellipse.posAngle - needs clarification on specification ( to clarify 
> agreement with IAU definition of "East of North" ).
> *ACTION:* add text describing spatial case where axis1 == North 
> Celestial Pole, axis2 == East, so position angle goes from axis1 toward 
> axis2, or "East of North"
>         + Alternate Error representations in catalogues ( eg: 2 std 
> deviations + covariance )
> *ACTION:* add to 'Next' page for Measurement model for consideration in 
> next revision.
> *ACTION:* adjust phrasing of requirement [meas:006] to match goals of 
> this version of the model (eg: Gaussian distributions at 1-sigma level)
> 
>      *) Domain specific Measurement types ("Time", "Position", 
> "Polarization" ) vs "GenericMeasure"
> *RESPONSE:* We feel there is a distinct advantage to being able to 
> define that Field X is a "Time" and Field Y is a "Position", rather than 
> having everything be a generic measurement type.
> 
>      *) Frame specific types (GalacticPosition, EquatorialPosition, etc )
> *RESPONSE/ACTION:* These will be removed/deprecated by the proposed 
> Coordinates model changes, in favor of a single Position type with Point 
> coordinate.
> 
>      *) Velocity.coordinate frames == "same SpaceFrame" is awkward, 
> "measurement doesn't really talk about frames"
> *RESPONSE/ACTION: *this constraint is not necessary after the Coordinate 
> model changes as the Point coordinate will have only 1 frame reference.
> 
> *) Specialized Positions have [1] multiplicity on each coordinate, not 
> convenient for describing (x,y) pair with CartesianPosition (for example)
> *RESPONSE/ACTION: *this issue is resolved after the Coordinate model 
> changes as the Point coordinate will have [0..1] multiplicity on each value.
> 
>      *) Gaia catalogue case where positions + proper motions are more 
> integrally related.
> *ACTION:* add to 'Next' page for Measurement model for consideration in 
> next revision, Gaia is a targeted provider in the Source model effort.
> 
>      *) ProperMotion.lon - "it is common, though not universal, practice 
> to quote longitudinal PM pre-multiplied by cos(lat) so that the 
> magnitude of the quantity is not affected by its longitudinal 
> position".  Some statement about this should be made in the model.
> *ACTION:*add descriptive text regarding this 'standard', but not 
> universal definition for position errors in longitude.
> 
>      *) Typos
>          o Sec 4.1.1, 7.2.1, 8.2.1: Multiplicity is quoted as "0..*" 
> rather than "*"
> *ACTION:* make corrections in document
> 
> 
> *Implementation/Serialization:
> *
>    This stems from the specific note in the RFC comments by Markus, and 
> follow-up discussion among the TCG at the interop.
> My point of view, is that the DM group approved a set of implementation 
> requirements which does not limit to any particular serialization 
> standard.**This is important since the data models themselves are 
> typically several layers below the level of interoperable 
> implementations of science threads.  These models have been serialized 
> using real-world data, in multiple formats (VOT, Annotated VOT, XML) 
> each validating against their schema.  This demonstrates the usability 
> of the models.  Interoperabile implementations only requires multiple 
> clients to select a common format.  There is no need to delay the model 
> progress to REC pending the approval of a specific annotation syntax.
> 
> 
> 

-- 
---- Laurent MICHEL              Tel  (33 0) 3 68 85 24 37
      Observatoire de Strasbourg  Fax  (33 0) 3 68 85 24 32
      11 Rue de l'Universite      Mail laurent.michel at astro.unistra.fr
      67000 Strasbourg (France)   Web  http://astro.u-strasbg.fr/~michel
---


More information about the dm mailing list