DSC (Discrete Source Classifier)
The Discrete Source Classifier (DSC) is one of the two top level classification modules in CU8, the other being the Object Clustering Analysis (OCA) package. DSC is responsible for classifying sources using supervised machine learning techniques. The output is probabilistic and divides the sources into broad astrophysical classes (single stars, white dwarfs, binaries, galaxies and quasars). This classification allows objects to be streamed to the correct parameterizing modules.
DSC is currently designed along modular lines, with several different classification algorithms working in parallel on different types of data. Probabilistic outputs from individual subclassifiers will be combined into a final output. The subclassifiers are;
- Photometric classifier, working on BPRP spectra. Based on a support vector machine (SVM)
- Position G magnitude classifier. Provides a prior based on source magnitude and sky position. Based on a Kernel density estimate of the sky distributions.
- Astrometric Classifier. Classifies based on proper motion and parallax. Based on a Gaussian mixture model.
- Variability classifier. Classify sources based on a photometric time series. Will be based on a structure function analysis and probably a mixture model. Not yet implemented.