Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-8112

Assess failures of stack code and new code.

    XMLWordPrintable

    Details

    • Story Points:
      10
    • Sprint:
      Alert Production F16 - 11, Alert Production F16 - 11b, Alert Production F16 - 11c, Alert Production S17 - 12, Alert Production S17 - 1, Alert Production S17 - 2
    • Team:
      Alert Production

      Description

      Understand specifically why failures occur for certain data sets in both algorithms and if new implementation succeeds explain why.

      Deliverable: Short documentation of each failure. (e.g. what was run, how it failed, field properties) Proposed fix if possible.

        Attachments

          Activity

          Hide
          cmorrison Chris Morrison [X] (Inactive) added a comment -

          Began comparing data between the current implementation of the pattern matcher and the pure python code.

          Show
          cmorrison Chris Morrison [X] (Inactive) added a comment - Began comparing data between the current implementation of the pattern matcher and the pure python code.
          Hide
          cmorrison Chris Morrison [X] (Inactive) added a comment - - edited

          Current failures both within the stack code and my own implementation of matchOptimsiticB appear to largely come from inflexible starting tolerances and default config parameters. For dense stellar fields, a common is problem that using only the 50 brightest stars is not sufficient to find a pattern match between the reference source catalogs. Also a one size fits all coordinate distortion tolerance for dense and sparse fields appears to be insufficient to fit all field densities. This failure is also seen when the reference catalog is padded around the CCD to accommodate the maximum shift allowed. The padding causes the current stack matcher to fail on fields it once succeeded in finding a successful and precise match.

          Possible solutions to this are an automated system for calculating tolerances based off of statistics from either the source or reference catalog. This need not slow down calculating matches too much as the data used to produce these statistics can be reused in the matcher (e.g. no need to recompute distances between all pairs). It also may be helpful to decrease the maxMatchDistance criteria per WCS iteration by using a 3 sigma or similar cut rather than feeding back the previous max distance between matcher found. This could improve matching in dense fields where the matches are much noisier due to the high density of reference and source objects.

          These suggestions are currently being implemented and tested in DM-8113.

          Show
          cmorrison Chris Morrison [X] (Inactive) added a comment - - edited Current failures both within the stack code and my own implementation of matchOptimsiticB appear to largely come from inflexible starting tolerances and default config parameters. For dense stellar fields, a common is problem that using only the 50 brightest stars is not sufficient to find a pattern match between the reference source catalogs. Also a one size fits all coordinate distortion tolerance for dense and sparse fields appears to be insufficient to fit all field densities. This failure is also seen when the reference catalog is padded around the CCD to accommodate the maximum shift allowed. The padding causes the current stack matcher to fail on fields it once succeeded in finding a successful and precise match. Possible solutions to this are an automated system for calculating tolerances based off of statistics from either the source or reference catalog. This need not slow down calculating matches too much as the data used to produce these statistics can be reused in the matcher (e.g. no need to recompute distances between all pairs). It also may be helpful to decrease the maxMatchDistance criteria per WCS iteration by using a 3 sigma or similar cut rather than feeding back the previous max distance between matcher found. This could improve matching in dense fields where the matches are much noisier due to the high density of reference and source objects. These suggestions are currently being implemented and tested in DM-8113 .
          Hide
          krughoff Simon Krughoff added a comment -

          I word smithed a bit, so have a look to make sure I didn't distort your meaning before you hit Done. Otherwise, looks fine to me.

          Show
          krughoff Simon Krughoff added a comment - I word smithed a bit, so have a look to make sure I didn't distort your meaning before you hit Done . Otherwise, looks fine to me.
          Hide
          cmorrison Chris Morrison [X] (Inactive) added a comment -

          Summarized why and how failures occur in the current stack implementation of matchOptimisticB and my own implementation. Suggested and possible changes will be tested in DM-8113.

          Show
          cmorrison Chris Morrison [X] (Inactive) added a comment - Summarized why and how failures occur in the current stack implementation of matchOptimisticB and my own implementation. Suggested and possible changes will be tested in DM-8113 .
          Hide
          cmorrison Chris Morrison [X] (Inactive) added a comment -

          Further discussed this ticket with ctslater. Suggested adding specifics on the automated tolerance calculation and other steps to DM-8113.

          Show
          cmorrison Chris Morrison [X] (Inactive) added a comment - Further discussed this ticket with ctslater. Suggested adding specifics on the automated tolerance calculation and other steps to DM-8113 .

            People

            Assignee:
            cmorrison Chris Morrison [X] (Inactive)
            Reporter:
            cmorrison Chris Morrison [X] (Inactive)
            Reviewers:
            Colin Slater, Simon Krughoff
            Watchers:
            Chris Morrison [X] (Inactive), Colin Slater, Simon Krughoff
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

              Dates

              Created:
              Updated:
              Resolved:

                Jenkins Builds

                No builds found.