Automated transient source detection algorithms are needed to fully leverage the massive datasets generated by modern and future sky surveys like ASAS-SN and Large Synoptic Survey Telescope (LSST). Along with improving ASAS-SN source detection through the implementation of new machine-learning algorithms, I spent a summer working with researcher at the SLAC National Accelerator Laboratory to enhance the realism of supernovae planted in simulated LSST data. These simulations are being used to improve the LSST pipeline before the survey begins real-time operations. As part of this work, I wrote two new Python software packages, which are described below.

XDGMM is a new program compatible with scikit-learn machine learning methods that uses Gaussian mixtures to perform density estimation of noisy or incomplete data using XD algorithms. Users can select different XD fitting algorithms and, most crucially, they can condition a model based on the known values of a subset of model parameters, creating a tool that can predict unknown parameters based on known parameters. It is a generic tool that is being used for a wide variety of applications.

empiriciSN is an example application of XDGMM that can be used for fitting a Gaussian mixture model to observed supernova/host datasets, creating a tool to predict likely supernova parameters using a model conditioned on observed host properties. This eliminates the need for theoretically modeling physical host quantities from observations. We have trained a default model on a sample of 1432 Type Ia SNe and their hosts from the Supernova Legacy Survey and the Sloan Digital Sky Survey that is being implemented to sample realistic Type Ia SNe for host galaxies in simulated LSST data.

Both tools are open-source and available for download on my github page. For more information, see the video and paper below.