Analysis of Characterisation in Domain Model Context

1
Analysis of Characterisation in Domain Model
Context

• With application to (SNAP) simulations

Gerard Lemson DWith feedback from (but dont
blame) Mireille Louys, Francois Bonnarel Claudio
Gheller, Patrizia Manzato, Laurie Shaw, Herve
Wozniak Miguel Cervino, Igor Chilingarian,
Norman Gray, Jaiwon Kim, Franck Le Petit, Ugo
Becciani, Sebastien Derriere Especially do not
blame Pat Dowler
2
Goal

• Understand characterisation ...
• context
• use
• application to (SNAP) simulation data
  modelbeyond space/time/lambda/flux
• ... through feedback from you
• Apply to SNAP
• note that use there probably not typical (pattern
  iso direct reuse?)
• Maybe find uses elsewhere?

3
Motivation

• The thing that is characterised does (did?) not
  occur explicitly inside characterisation model
  (Observation is gone)
• Found characterisation-like features in SNAP data
  model, useful for discovery that do contain this
  thing explicitly
• Carries over directly to full domain model

4
(No Transcript)
5
(No Transcript)
6
The simulation model

• Focuses on experiments, which
• have target objects
• which have observables
• which have typical values (as function of time)
• have representations
• consisting of (simulation dependent) object types
• which have (simulation dependent)
  properties/observables (mass, position,
  wavelength, flux, temperature, entropy etc)
• have input parameters
• have results
• which have collections of measurement
  (simulation) objects (corresponding to the
  representation object types)
• which assign values (and errors) to the
  properties

7
Use values/params for discovery

• The full data (results) can not be used as they
  are in discovery and (SXAP-)queryData
• It is hard to query on input parameters when
  semantics, and consequences not well
  known/understood
• Nevertheless useful info contained in them and
  desired for querying
• Use statistical description characterising the
  results, both a priori and a posteriori

8
In domain

• Domain model analyses the domain
• a priori characterisation
• restricts possible values an observable may have
• summarises effects of input parameters
• similar to Characterisation DM (private comm
  HMcD, ML last year) ??
• a posteriori characterisation
• summarises actual results
• statistics of particular observable in result
  collection of objects

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
Back to simulations

• Logical model
• application targeted
• simpler, less normalised
• 1 characterisation object

13
(No Transcript)
14
Conclusion

• Treat characterisation as a pattern iso reusable
  software/dm component
• Coverage characterisation of values
• not (yet) of errors (is this Accuracy?)
• necessary for discovery and query (of
  simulations)?
• No
• accuracy
• where does this go for simulations
• where in domain?
• resolution (does this belong on target object,
  iso representation)
• sampling precision (a priori?)