SRC workflow example: Source Classification (not CNN)

Input data: catalog data, preferably cross-matched with other wavelengths

Output data: catalog of class labels with confidence 

Software involved: Python machine learning libraries

Workflow steps: Train a model on some labeled data, apply the model to unlabeled data.

State of existing workflows: SDC1 solution has these workflows done already, but only applies to simulated SDC1 radio data, so the results are not very representative of actual science. Alex’s SDSS & WISE star/galaxy/quasar classification paper is a complete multi-wavelength example, but excludes radio. Lots of options here to distribute data, build models separately and then combine them later in one location. Distributed data workflow would be a lot of work to implement.

Architecture and hardware: Multithreaded CPU. Flexible RAM. Data in a single location, or can distribute data for training/classification and combine models afterwards.