Skip to content

Latest commit

 

History

History
53 lines (43 loc) · 3.24 KB

lightcurves.md

File metadata and controls

53 lines (43 loc) · 3.24 KB

Lightcurves

A first version of a new LightCurveEstimator was introduced shortly after Erlangen coding sprint. The general approach is described in pig 11.

The PIG has been withdrawn because it is already partly implemented and parts of the discussion are no longer relevant given the changes introduced in the data reduction since Erlangen.

Current status and proposed solutions

Time information on datasets

  • LightCurveEstimator relies on dataset time start and stop information stored as meta.
    • to have a cleaner, treatment we have introduced a GTI table on the various datasets
  • LightCurveEstimator should rather deal with GTI

Combining datasets for varying time bins

  • LightCurveEstimator assumes on dataset per time bin and uses only the binning given by the list of dataset
    • a very common use case is to rebin groups of datasets on nightly, weekly, monthly etc basis
  • I propose to pass a tuple of time bins to LightCurveEstimator and let him combine datasets together
    • We need to define the rule (e.g. GTI fully contained in the time bin, or GTI center within time bin)
    • This behaviour is then similar to FluxPointsEstimator
    • see issue #2548

Code duplication

  • The current LightCurveEstimator suffers from heavy code duplication from FluxPointsEstimator. See the code
    • this is due to a very similar task with a slightly different logic
      • Both extract flux by fitting a scaled model of the source on the dataset
      • But FluxPointsEstimator uses the same datasets for all steps (i.e. all energy bins)
      • LightCurveEstimator uses different datasets for each time bin.
    • the Fit object has to be re-instantiated at each step in the loop
    • inheritance will not work
  • I propose to simplify LightCurveEstimator heavily to rely on FluxPointsEstimator
    • For each time bin, create a FluxPointsEstimator for the current list of dataset with one single energy bin
    • Run flux point estimation and extract info from the table

Light curves with high level interface

We need to be able to run light curve from the HLI. This requires a number of additional features.

  • Define a schema for time bins definition in yaml
    • a tstart - tstop, and bin duration? What format for tstart and tstop?
    • from a file containing a list of Time?
  • Apply time selection on Observations

Other issues - important but lower priority

  • How do we deal with sub-observations statistics, in particular for ON-OFF datasets?
    • Usually, OFF can be extracted from the full observation
      • This implies to perform the ReflectedRegionsBackgroundMaker on an unfiltered EventList
      • This introduces bin-to-bin correlations
      • This makes LightCurve extraction more complex if the time bin size is different from the one used to filter the dataset
    • Can there be a grouping scheme to allow for shared background among datasets using wstat?