Into the VO: Probabilistic Kernel Density Classification?

[Brian Major]

Hi Grid,

I would like to start a discussion on the feasibility of standardizing the
service presented by François-Xavier Pineau at this morning KDD/GWS session.

For those of you who were not able to attend, FX's slides can be seen here:

http://wiki.ivoa.net/internal/IVOA/InterOpOct2017KDD/201710_IVOA_KDD_fxp.pdf

In a nutshell, he presented a prototype of a REST service that automated
source classification with probabilistic estimates.  The estimates of
matching are done though direct database catalogue queries.  (Please,
correct me if I have anything wrong here.)

It occurred to me and to others who were in attendance that this is a good
candidate for a service that could be offered by the IVOA.  It is much like
cutouts--a software function that is common and useful enough to be offered
as a service, as opposed to requiring users to code such a mechanism
themselves.

But, without any presumptions, here are some questions and possible options:

(1)   Does this sound like a good candidate for standardization?

(2)   If (1), how can it be offered:

    (a)  VO can provide or recommend a library that users can use when they
upload code to be executed in a data centre.

    (b)  It can be offered as an independent REST service, as shown in the
prototype, with possible UWS support.

    (c)  It can be offered as VOSpace standard capability.  (If a directory
supports the capability, the data nodes in the directory are treated as
database tables and the estimation capability could be invoked with the
required parameters.  -- or something like that.)

    (d)  It can be offered as a SODA service

    (e)  Others?

Please, if you have an opinion, respond.  My quick thoughts:

(1)  Yes, it is useful and transferrable to different types of problems
users face.
(2)
    (a)  Too much effort placed on users
    (b)  Not general enough
    (c)  A good fit, but a trailblazer for using the VOSpace node
capabilities support
    (d)  SODA operates on files, not catalogues, so not a good fit.


Regards,
Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ivoa.net/pipermail/grid/attachments/20171028/0076f5d4/attachment.html>