To identify scientifically valuable transients or moving objects on the sky, astronomical imaging surveys have typically taken a manual approach, employing humans to vet candidate events extracted from difference images by eye. Usually, the vast majority of these are artifacts of processing or instrumentation, making the search process tedious, error-prone, time-consuming, and subjective.
Difference image objects scored from 0 (garbage) to 1 (real) by
autoscan is a python package developed for the
Dark Energy Survey Supernova Survey that uses the
supervised machine learning technique known as Random Forest
to quickly, accurately, automatically, and consistently
identify difference image objects caused by genuine
astrophysical variability, implementing the framework
Goldstein et al. (2015).
autoscan is provided
on GitHub as
a reference implementation only, with no guarantee of
general usability. It was developed for use in the
difference imaging pipeline for
the Dark Energy
Survey, and it assumes an Oracle database structure and
input file structure that are specific to the survey.
You can download the training data (features and postage stamps)
autoscan below. A description of the training data
appears in section 3
of Goldstein et
al. (2015). Note that many of the data files exceed 1GB in
size. We recommend that you use a command line tool such
curl to execute the
The following download contains the class labels and values of the
38 features presented in Table 2 of
Goldstein et al. (2015) computed over each of
the 898,963 detections in the
autoscan training data
set. The structure of the table is described in the first few lines of
The following downloads contain the postage stamp images (search,
template, and subtraction) for the training instances
If you make use of
autoscan source code or
training data in your work, please cite:
Copyright 2015 the author. All rights reserved.
autoscan is free software licensed under the GNU General Public License Version 3. See the file
LICENSE for the full terms of this license.