Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-9681

Write function to apportion flux based on NMF template weights

    Details

    • Type: Story
    • Status: Done
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_deblender
    • Labels:
      None

      Description

      The main assumption of the current deblender is that galaxies have an identical profile in each band. This is not strictly true, as the color near the center of galaxies is more white.

      It is worth attempting to use the NMF deblender output as a template, similar to the symmetric templates generated by the SDSS deblender, and re-apportion flux based on the ratio of the templates in each pixel. This may help us recover more accurate colors while still performing better in the wings of blended galaxies than the current deblender.

      This ticket will refactor a version of lsst.meas.deblender.apportionFlux to re-apportion flux for NMF templates.

        Attachments

          Activity

          Hide
          fred3m Fred Moolekamp added a comment -

          This work is included in the branch for DM-9172 in order to test the new validation metrics. See commit 6bfdc3a

          Show
          fred3m Fred Moolekamp added a comment - This work is included in the branch for DM-9172 in order to test the new validation metrics. See commit 6bfdc3a
          Hide
          fred3m Fred Moolekamp added a comment -

          Peter Melchior sorry to send you a bunch of review tickets today but I got a lot of cleanup work done on the plane yesterday so that I can close some of these issues that were nearly complete and move on to creating the PSF convolution operator. This one just implements the algorithm to reapportion the flux from the real data based on the weights given by the individual templates.

          I haven't implemented any code to try and reduce the footprint size, which is likely why both flux measurement methods are usually high for most sources. You can see the ipython notebook here, which shows a comparison of the original deblender, NMF with monotonicity, and re-apportioned flux using the NMF output to weight the sources.

          Show
          fred3m Fred Moolekamp added a comment - Peter Melchior sorry to send you a bunch of review tickets today but I got a lot of cleanup work done on the plane yesterday so that I can close some of these issues that were nearly complete and move on to creating the PSF convolution operator. This one just implements the algorithm to reapportion the flux from the real data based on the weights given by the individual templates. I haven't implemented any code to try and reduce the footprint size, which is likely why both flux measurement methods are usually high for most sources. You can see the ipython notebook here , which shows a comparison of the original deblender, NMF with monotonicity, and re-apportioned flux using the NMF output to weight the sources.
          Hide
          pmelchior Peter Melchior added a comment -

          It's worthwhile clarifying one thing: the flux reapportion has nothing to do with colors in the centers vs outskirts. What it should solve is that the non-negative constraint on the "templates" cuases a bias on the fluxes.

          Show
          pmelchior Peter Melchior added a comment - It's worthwhile clarifying one thing: the flux reapportion has nothing to do with colors in the centers vs outskirts. What it should solve is that the non-negative constraint on the "templates" cuases a bias on the fluxes.
          Hide
          fred3m Fred Moolekamp added a comment -

          Good point.

          Show
          fred3m Fred Moolekamp added a comment - Good point.

            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