A suggested revision for UCDs
dide at discovery.saclay.cea.fr
Thu Oct 23 10:26:42 PDT 2003
Big and impressive work.
I have read and try to absorb the document and most of the following
mails. I try to digest all. It is not so obvious for a frenchy
like me and I hope that my comments below are appropriate.
I apologize for any misunderstanding or confusion, by jumping in
the discussion, but a really want to make a few points.
If a clearly understand the concern of Pedro Osuna, the actual discussion
illustrate what he stressed in strasbourg and his further mail.
UCD1 and a data model for context evaluation is sufficient,
because DM is not available some structure is needed to be introduced
and so UCD2 came. But unless we put DM in UCD (eventually using VOTable
structure) it will be incomplete. Even UCDTree (which I really like)
needed additional structure/template or external reference as stressed
in the document 1.9.9b: p16 last paragraph, p18 last paragraph.
I like several aspect of this proposal, and I agree of some of the comments
of Norman, Ed and Jonathan, but as it is very tedious for me to write
in english, I will not do an extensive comment list of what is hapenning
in the discussion but go directly to the pb I had.
Instead of specific words or context meaning to distinguish
between 3 kind of thing; concept, modifier property and attribute property
I would be more comfortable with a syntax distinction and a specific
separator (i.e. #).
The complexity of the three trees with specific words seems very strange
and is confusing me. My main concern is related to Attributes.
- I did not understand the meaning of local. You can always correlated
a property with an identical or similar property, it depend of the purpose
of your correlation, and without presuposition you cannot exclude
apriori some intressting thing that can be extract from data.
Taking your example: you can extract data where temperature jitter of
the data acquisition system is above a certain threshold and try to correlate
these with phot,flux;error or any kind of measurement error available.
Perhaps not used very often, but not forbidden I hope?
- It seems to me that there are some dangerous redundancy in the
basic word of the attribute tree. For example value, vector, instance,
multiplet are common data properties distinguish by their use or
the context in which they appear. IMO there is no differences
between vector and multiplet and each time you use multiplet
you could replace it by vector. No?
Like jonathan, if you put vectors in value, I did not see the
need for instance.
The relation of measurements, errors and all this kind
of things with the precedding is even more complicated.
A value (I mean a real scalar) can be a unique measurement value,
a statistical property of a measurement serie (like mean, mode,
error, std.dev., skewness...), a parameter and perhaps other
meaning I am not able to think of now.
Trying to be brief, (but unfortunatly uncomplete) I feel that
the pb came from the fact that diff data property are mixed
together (I believe that the same pb occurs in DM group).
Mainly data structure, data meaning (could be extended to purpose) and
data representation (without speaking of format and location).
Data structure would include ; value (scalar), vector (avoiding multiplet)
tree, and extended easly with matrix or composite structure.
For me instance is A structure, but wich one? A free formatted structure,
a VOTable tree formatted? It is not clear.
data meaning is related to measurement, error and all staistical properties.
data representation mixed both things; a measurement serie can be represented
by a vector (even in a VOTable cell), or by one or several statistical prop.
(mean, mean+std.dev.) structured in very diff way (mean and error in separate
cols, mean+mode+median in one col ...)
I feel that refurbishing is needed here to clarify UCD usage more
than UCD existance. I agree with earlier comments that stressed that
the big advantage of UCD1 is the fact that they are not due to
apriori/re-invented structure but are the illustration of the existing data.
- I did not see the filter needs. It seems to me that it try to catch
a part of the data history, but it seems so restrive that it will be
very soon unappropriate I bet.
- concept seems only due to the uncompltness of the words available
in the concept tree root.
In your example (p18) a word correlation (which would certainly be needed for VO)
would better match the needs.
I stop here because it's late and I would become confused,
if not yet done.
Thanks for the food for thought,
DIDELON e-mail : pdidelon_at_cea.fr
CEA SACLAY - Service d'Astrophysique W3 : http://www-dapnia.cea.fr/Sap/
91191 Gif-Sur-Yvette Cedex Phone : 33 (0)1 69 08 58 89
More information about the ucd