Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-9954

Optimize PSF convolution to use sparse matrices

    Details

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

      Description

      Currently the PSF convolution operator is a collection of 2D sparse arrays, which are different for each band, that converted are into full arrays in order to use Peter Melchior's proximal operator algorithm. Since the current simulated data is PSF matched, the processing time can be significantly reduced (by several orders of magnitude) by using the PSF operator as a sparse matrix in each band.

      This fix is temporary so that we can profile and test PSF convolution and other features of the new deblender but it is expected that this section of the code will be re-written in C++ in the final deblender before it is merged into the stack in a future sprint.

        Attachments

          Activity

          Hide
          fred3m Fred Moolekamp added a comment -

          Peter Melchior was able to implement this by evaluating each band separately using the sparse PSF convolution operator created in DM-9170. I updated meas_deblender to use the new API and updated testNMFProximal. to run the deblender using PSF convolution, sparsity, and monotonicty, and display the time to deblend a region with 4 peaks (1min 8s).

          Show
          fred3m Fred Moolekamp added a comment - Peter Melchior was able to implement this by evaluating each band separately using the sparse PSF convolution operator created in DM-9170 . I updated meas_deblender to use the new API and updated testNMFProximal. to run the deblender using PSF convolution, sparsity, and monotonicty, and display the time to deblend a region with 4 peaks (1min 8s).
          Hide
          fred3m Fred Moolekamp added a comment -

          Peter Melchior since you made the modification, do you mind reviewing this ticket? I know that you're busy this week, so it's ok to wait until next week.

          Show
          fred3m Fred Moolekamp added a comment - Peter Melchior since you made the modification, do you mind reviewing this ticket? I know that you're busy this week, so it's ok to wait until next week.
          Hide
          pmelchior Peter Melchior added a comment -

          Well, I recognize my code...

          Show
          pmelchior Peter Melchior added a comment - Well, I recognize my code...
          Hide
          fred3m Fred Moolekamp added a comment -

          Show
          fred3m Fred Moolekamp added a comment -

            People

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

              Dates

              • Created:
                Updated:
                Resolved:

                Summary Panel