Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-10310

Implement translation updates

    Details

    • Type: Story
    • Status: Done
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_deblender
    • Labels:
      None
    • Story Points:
      10
    • Sprint:
      DRP F17-3
    • Team:
      Data Release Production

      Description

      Implement the translation operator and algorithm designed in DM-10189.

        Attachments

          Issue Links

            Activity

            Hide
            pmelchior Peter Melchior added a comment -

            The operator is there, but we need to determine how to update it with the internal state parameters

            Show
            pmelchior Peter Melchior added a comment - The operator is there, but we need to determine how to update it with the internal state parameters
            Hide
            fred3m Fred Moolekamp added a comment -

            The changes have been implemented in the deblender repository in the "translations" branch. See all of the examples in the deblender repo to see how translations improve the efficiency and residuals in the deblender, and single_sim_hist.ipynb for an example with the LSST stack.

            There are still a few issues that need to be addressed in future tickets:

            1. The behavior for peaks that drift needs to be understood and a better solution needs to be implemented (see single_sim_hist.ipynb
            2. The convergence checks need to be improved, since the algorithm practically converges very quickly, even though it can be difficult to get the actual result to converge
            Show
            fred3m Fred Moolekamp added a comment - The changes have been implemented in the deblender repository in the "translations" branch. See all of the examples in the deblender repo to see how translations improve the efficiency and residuals in the deblender, and single_sim_hist.ipynb for an example with the LSST stack. There are still a few issues that need to be addressed in future tickets: The behavior for peaks that drift needs to be understood and a better solution needs to be implemented (see single_sim_hist.ipynb The convergence checks need to be improved, since the algorithm practically converges very quickly, even though it can be difficult to get the actual result to converge
            Hide
            fred3m Fred Moolekamp added a comment -

            Hi Peter, I know you're busy on Thursday but when you get a chance can you have a look at the pull request on the deblender repo and review this ticket?

            Thanks,
            -Fred

            Show
            fred3m Fred Moolekamp added a comment - Hi Peter, I know you're busy on Thursday but when you get a chance can you have a look at the pull request on the deblender repo and review this ticket? Thanks, -Fred
            Hide
            fred3m Fred Moolekamp added a comment -

            Created DM-11553 to improve the behavior of drifting peaks and DM-11554 to implement an improved check for convergence of Symmetry-like constraints.

            Show
            fred3m Fred Moolekamp added a comment - Created DM-11553 to improve the behavior of drifting peaks and DM-11554 to implement an improved check for convergence of Symmetry-like constraints.
            Hide
            pmelchior Peter Melchior added a comment -

            These work very well. In fact, I'm quite amazed that this solution for the centering updates is capable of finding solutions that are hard to identify by eye (and I've tried...). My interpretation is that considering several bands and all peaks at the same time enables robust updates.

            In fact, these work so well that finally we can use our default, cheap "M" monotonicity operator again and have convergent solutions.

            Show
            pmelchior Peter Melchior added a comment - These work very well. In fact, I'm quite amazed that this solution for the centering updates is capable of finding solutions that are hard to identify by eye (and I've tried...). My interpretation is that considering several bands and all peaks at the same time enables robust updates. In fact, these work so well that finally we can use our default, cheap "M" monotonicity operator again and have convergent solutions.

              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