AP Pipeline to include real/bogus classification

I’m about to merge DM-42217, which (among many other changes) adds lsst.meas.transiNet.RBTransiNetTask to the baseline Alert Production pipeline definition, and adds real/bogus classifications to the transformed DiaSource catalog and APDB DiaSource table as the reliability column. This is following up on a lot of development and testing by @Nima and @harshit over the past year.

The new AP pipeline will first be available in Science Pipelines d_2024_02_07 and w_2024_06.

RBTransiNetTask requires an extra pipeline input in the form of a machine learning model. A suitable model is included in /repo/main (AP runs should “just work”), and we’re planning on adding support to /repo/embargo and Prompt Processing soon. The shared repos will be updated with better classifiers as they are developed.

1 Like

I am interested in strongly lensed transients, which may have the PSF twisted/distorted. Can I hope to catch these by selecting diaSources with low reliability? Is there a document explaining how reliability is computed? Thank you!