Photometric Classification of Astronomical Transients and Variables With Biased Spectroscopic Samples
avocado is a general photometric classification code that is designed to
produce classifications of arbitrary astronomical transients and variable
objects. This code is designed from the ground up to address the problem of
biased spectroscopic samples. It does so by generating many lightcurves from
each object in the original spectroscopic sample at a variety of redshifts and
with many different observing conditions. The "augmented" samples of
lightcurves that are generated are much more representative of the full
datasets than the original spectroscopic samples.
The original codebase of avocado was developed for and won the 2018 Kaggle PLAsTiCC
challenge. A paper describing the algorithms
implemented in this package can be found
here, and there is
also a discussion available on the kaggle
forum. The PLAsTiCC results
can be replicated using the latest version of avocado following the steps in the
documentation, and all of the code used to generate the figures in the paper can be
found in the avocado_paper_figures.ipynb notebook.
Carrick et al. 2021 (submitted to MNRAS) used avocado to study how to optimize
spectroscopic training samples for photometric classification of supernovae. An example
of the code used in that analysis can be found in the spcc_augment_methods.ipynb
notebook.
Instructions on how to install and use avocado can be found on the avocado
readthedocs page.