Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-10614

Use multiple components per peak

    Details

    • Type: Story
    • Status: Done
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_deblender
    • Labels:
      None
    • Story Points:
      4
    • Sprint:
      DRP F17-1, DRP F17-4
    • Team:
      Data Release Production

      Description

      It is likely that objects like spiral galaxies with a resolve bulge and disk or extended galaxies with color gradients will prove difficult for a single SED deblender. This ticket will test using multiple SEDs for a single peak on HSC data to see if it is able to deblended more complicated structures.

        Attachments

          Activity

          Hide
          fred3m Fred Moolekamp added a comment - - edited

          Hi Peter,
          Here are the changes for a galaxies with multiple components. It works fairly well for weakly blended objects, but I'm not convinced that it will be effective for heavily blended regions, where it might be easier to use a single component for all but the brightest objects in a region.

          See the pull request in the deblender repo, included the notebooks that demonstrate the old and new behavior.

          Show
          fred3m Fred Moolekamp added a comment - - edited Hi Peter, Here are the changes for a galaxies with multiple components. It works fairly well for weakly blended objects, but I'm not convinced that it will be effective for heavily blended regions, where it might be easier to use a single component for all but the brightest objects in a region. See the pull request in the deblender repo, included the notebooks that demonstrate the old and new behavior.
          Hide
          pmelchior Peter Melchior added a comment -

          Difficulties in highly blended regions are to be expected: there are just more moving parts, i.e. more degeneracies in the fitter without more constraints -> unstable behavior. Practically, it'll mean that we need a logic to decide when it is beneficial to fit multiple components per object or when we should restrict ourselves to one.

          Show
          pmelchior Peter Melchior added a comment - Difficulties in highly blended regions are to be expected: there are just more moving parts, i.e. more degeneracies in the fitter without more constraints -> unstable behavior. Practically, it'll mean that we need a logic to decide when it is beneficial to fit multiple components per object or when we should restrict ourselves to one.
          Hide
          pmelchior Peter Melchior added a comment -

          Multiple component fitting works as expected:

          • If there are components of different color, they are fit separately.
          • If there are more components than needed, it extra ones do no harm (data analysis version of the Hippocratic Oath).
          • The centers of multiple components per object are tied together.

          I have a suggestion re interface: unify the three arguments peaks, components, constraints to a single one: objects (a list where each item has a peak position, number of components, and constraints for each component, and maybe a catalog id). We can pick this up in another ticket.

          Show
          pmelchior Peter Melchior added a comment - Multiple component fitting works as expected: If there are components of different color, they are fit separately. If there are more components than needed, it extra ones do no harm (data analysis version of the Hippocratic Oath). The centers of multiple components per object are tied together. I have a suggestion re interface: unify the three arguments peaks, components, constraints to a single one: objects (a list where each item has a peak position, number of components, and constraints for each component, and maybe a catalog id). We can pick this up in another ticket.
          Hide
          fred3m Fred Moolekamp added a comment -

          Merged changes

          Show
          fred3m Fred Moolekamp added a comment - Merged changes

            People

            • Assignee:
              fred3m Fred Moolekamp
              Reporter:
              fred3m Fred Moolekamp
              Reviewers:
              Peter Melchior
              Watchers:
              Fred Moolekamp, Peter Melchior
            • Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Summary Panel