Updates for November 22

[bfhealy]

Pull requests

• Fix inference bugs #170
• Include ra/dec/period columns in inference result #171
• Add probability threshold for classification uploads #174
• Convert ZTF ID to int before uploading #175
• Post all ZTF IDs for close sources #176
• Fix annotation bug for existing sources with no ZTF ID #177
• Reorganize and update defaults for API try/except loop #178

Issues

• Negative Gaia parallaxes #172
• Use annotations to post ML algorithm scores #103
• New taxonomy for scanner comments #173

Going forward

The SCoPe DNN Active Learning 1 group on Fritz (number 1503) now contains 11,536 sources and high-confidence (p ≥ 0.9) classifications from the latest trained DNN models. The sources in this group are ready for expert scanning.

The following taxonomies are used for these sources:

• SCoPe Phenomenological Taxonomy contains the phenomenological classifications.
• Sitewide Taxonomy contains the ontological classifications.
  • Note that the SCoPe binary star label currently maps to Eclipsing in this taxonomy (see 'Binary star' vs 'eclipsing' labels #133)
• SCoPe Scanner Taxonomy contains two classifications: Labeled, useful for marking which sources have been scanned previously, and Ignore, for cases where the classifications are better ignored than revised for the next round of training. We are currently not using the 'SCoPe Taxonomy' that may appear as another dropdown option.

The following workflow is recommended for scanning sources:

• Choose a specific label or set of related labels to scan

  • Filter by these classifications (at the top right of the group sources page)
  • Additionally filter by non-classification to exclude sources that already have the Labeled tag

• Click the dropdown arrow at the top left of the group sources page to reveal the scanning interface for all sources

• Drop down the ZTF_DR5 and UPLOADEDPERIOD annotations to be aware of multiple blended sources and period estimates

• Labeling:

  • Begin scanning photometry, using the dropdown/slider system to post new classifications (be sure to toggle Scale parent/child probabilities on to maintain consistency within the label hierarchy)
  • If necessary, remove incorrect labels by clicking on the Source ID (recommend opening in new tab), scrolling to Classifications and clicking the trash bin icon next to each undesired label
  • Fully label the source, including labels that you did not filter for (or mark as Ignore if appropriate)
  • When finished adding/removing classifications, mark the source as Labeled in the Scope Scanner Taxonomy

Thank you for your help with the active learning process!