The Automatic Learning for the Rapid Classification of Events (ALeRCE) Alert Broker

Abstract

We introduce the Automatic Learning for the Rapid Classification of Events (ALeRCE) broker, an astronomical alert broker designed to provide a rapid and self--consistent classification of large etendue telescope alert streams, such as that provided by the Zwicky Transient Facility (ZTF) and, in the future, the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST). ALeRCE is a Chilean--led broker run by an interdisciplinary team of astronomers and engineers, working to become intermediaries between survey and follow--up facilities. ALeRCE uses a pipeline which includes the real--time ingestion, aggregation, cross--matching, machine learning (ML) classification, and visualization of the ZTF alert stream. We use two classifiers: a stamp--based classifier, designed for rapid classification, and a light--curve--based classifier, which uses the multi--band flux evolution to achieve a more refined classification. We describe in detail our pipeline, data products, tools and services, which are made public for the community (see \url{https://alerce.science}). Since we began operating our real--time ML classification of the ZTF alert stream in early 2019, we have grown a large community of active users around the globe. We describe our results to date, including the real--time processing of 9.7×1079.7\times10^7 alerts, the stamp classification of 1.9×1071.9\times10^7 objects, the light curve classification of 8.5×1058.5\times10^5 objects, the report of 3088 supernova candidates, and different experiments using LSST-like alert streams. Finally, we discuss the challenges ahead to go from a single-stream of alerts such as ZTF to a multi--stream ecosystem dominated by LSST.

Full text

,, ,
Draft version August 11, 2020
Typeset using L
A
T
E
Xtwocolumn style in AASTeX63
The Automatic Learning for the Rapid Classification of Events (ALeRCE) Alert Broker
(The ALeRCE collaboration)
F. F¨
orster,1, 2, 3 G. Cabrera-Vives,4, 2 E. Castillo-Navarrete,1,2 P. A. Est´
evez,5, 2, 1 P. S´
anchez-S´
aez,2, 6, 7
J. Arredondo,2F. E. Bauer,6, 8, 2, 9 R. Carrasco-Davis,2, 5 M. Catelan,6,8, 2 F. Elorrieta,2, 10 S. Eyheramendy,2,7
P. Huijse,11, 2 G. Pignata,12, 2 E. Reyes,2, 5 I. Reyes,2, 1, 5 D. Rodr
´
ıguez-Mancini,2, 4 D. Ruz-Mieres,1, 2, 13
C. Valenzuela,2, 1 I. ´
Alvarez-Maldonado,2,1 N. Astorga,2, 5 J. Borissova,14,2 A. Clocchiatti,6, 2 D. De Cicco,2, 6
C. Donoso-Oliva,4, 2 M. J. Graham,15 R. Kurtev,14, 2 A. Mahabal,15, 16 J.C. Maureira,1R. Molina-Ferreiro,5
A. Moya,2, 1 W. Palma,2M. Prez-Carrasco,4,2 P. Protopapas,17 M. Romero,5L. Sabatini-Gacitua,2,1
A. Snchez,4, 2 J. San Mart
´
ın,1C. Sep´
ulveda-Cobo,2, 1 E. Vera,1J. R. Vergara,18, 2
1Center for Mathematical Modeling, University of Chile, AFB170001, Chile
2Millennium Institute of Astrophysics, Nuncio Monse˜nor S´otero Sanz 100, Providencia, Santiago, Chile
3Departamento de Astronom´ıa, Universidad de Chile, Casilla 36D, Santiago, Chile
4Department of Computer Science, University of Concepcin, Chile
5Department of Electrical Engineering, Universidad de Chile, Av. Tupper 2007, Santiago 8320000, Chile
6Instituto de Astrof´ısica, Facultad de F´ısica, Pontificia Universidad Cat´olica de Chile, Av. Vicu˜na Mackenna 4860, 7820436 Macul,
Santiago, Chile
7Department of Engineering and Science, Universidad Adolfo Ibaez, Av. Diagonal Las Torres 2700, Santiago, Chile
8Centro de Astroingenier´ıa, Pontificia Universidad Cat´olica de Chile, Av. Vicu˜na Mackenna 4860, 7820436 Macul, Santiago, Chile
9Space Science Institute, 4750 Walnut Street, Suite 205, Boulder, Colorado 80301, USA
10Department of Mathematics and Computer Science, Universidad de Santiago de Chile, Av. Libertador Bernardo OHiggins 3663,
Estaci´on Central, Santiago, Chile
11Instituto de Inform´atica, Universidad Austral de Chile, General Lagos 2086, Valdivia, Chile
12Departamento de Ciencias F´ısicas, Universidad Andres Bello, Avda. Republica 252, Santiago, Chile
13Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
14Instituto de F´ısica y Astronom´ıa, Universidad de Valpara´ıso, Av. Gran Breta˜na 1111, Playa Ancha, Casilla 5030, Chile
15Division of Physics, Mathematics, and Astronomy, California Institute of Technology, Pasadena, CA 91125, USA
16Center for Data Driven Discovery, California Institute of Technology, Pasadena, CA 91125, USA
17Institute for Applied Computational Science, Harvard University, Cambridge, MA, USA
18Departmento de Inform´atica y Computaci´on, Universidad Tecnol´ogica Metropolitana, Santiago, Chile
ABSTRACT
We introduce the Automatic Learning for the Rapid Classification of Events (ALeRCE) broker, an
astronomical alert broker designed to provide a rapid and self–consistent classification of large etendue
telescope alert streams, such as that provided by the Zwicky Transient Facility (ZTF) and, in the future,
the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST). ALeRCE is a Chilean–led
broker run by an interdisciplinary team of astronomers and engineers, working to become intermediaries
between survey and follow–up facilities. ALeRCE uses a pipeline which includes the real–time ingestion,
aggregation, cross–matching, machine learning (ML) classification, and visualization of the ZTF alert
stream. We use two classifiers: a stamp–based classifier, designed for rapid classification, and a light–
curve–based classifier, which uses the multi–band flux evolution to achieve a more refined classification.
We describe in detail our pipeline, data products, tools and services, which are made public for the
community (see https://alerce.science). Since we began operating our real–time ML classification of
the ZTF alert stream in early 2019, we have grown a large community of active users around the
globe. We describe our results to date, including the real–time processing of 9.7×107alerts, the stamp
Corresponding author: F. F¨orster
francisco.forster@gmail.com
arXiv:2008.03303v1 [astro-ph.IM] 7 Aug 2020

2F¨
orster et al.
classification of 1.9×107objects, the light curve classification of 8.5×105objects, the report of 3088
supernova candidates, and different experiments using LSST-like alert streams. Finally, we discuss
the challenges ahead to go from a single-stream of alerts such as ZTF to a multi–stream ecosystem
dominated by LSST.
Keywords: editorials, notices — miscellaneous — catalogs — surveys
1. INTRODUCTION
The exponential growth of the light collecting area of
telescopes and the number of pixels of digital detectors
has resulted in a new generation of survey telescopes
that are revolutionizing the way we study the time do-
main in astronomy (Tyson 2019). New surveys that
systematically scan the optical/near infrared sky with
deep, wide and fast cadence observations (e.g., Catalina
Real-Time Transient Survey, CRTS, Drake et al. 2009;
Palomar Transient Factory, PTF, Law et al. 2009; Op-
tical Gravitational Lensing Experiment, OGLE, Udal-
ski et al. 2015; Dark Energy Survey, DES, The Dark
Energy Survey Collaboration 2005; SkyMapper, Keller
et al. 2007;Kepler,Koch et al. 2010; Vista Vari-
ables in the Via Lactea Survey, VVV, Minniti et al.
2010; Korea Microlensing Telescope Network, KMTNet,
Kim et al. 2016; Hyper Suprime-Cam Subaru Strate-
gic Program, HSC-SSP, Aihara et al. 2017; Asteroid
Terrestrial–Impact Last Alert System, ATLAS, Tonry
et al. 2018; Zwicky Transient Facility, ZTF, Bellm et al.
2019; Deeper, Wider, Faster, DWF, Andreoni et al.
2020) are uncovering large populations of time–varying
astrophysical phenomena, including new populations of
dim, rare, and/or short-lived events (e.g., Kasliwal et al.
2012;Drout et al. 2014).
Meanwhile, the construction of the Vera C. Rubin Ob-
servatory and its Legacy Survey of Space and Time,
LSST (LSST Science Collaboration et al. 2009), is ad-
vancing, and a convergence is expected to happen with
surveys in other regions of the electromagnetic spec-
trum (e.g., Square Kilometer Array, SKA, Dewdney
et al. 2009; Wide-field Infrared Survey Explorer, WISE,
Wright et al. 2010; eROSITA, Merloni et al. 2012;
Fermi Gamma-ray Space Telescope, Atwood et al. 2009;
Cherenkov Telescope Array, CTA, Actis et al. 2011),
high energy particles (e.g., CTA; IceCube Neutrino Ob-
servatory, Aartsen et al. 2017), and gravitational waves
(Laser Interferometer Gravitational-Wave Observatory,
Abramovici et al. 1992; Advanced Virgo, Acernese et al.
2015), opening a new era of multi–messenger astronomy
(Abbott et al. 2017;IceCube Collaboration et al. 2018).
The fundamental quantity that defines a survey tele-
scope is the product of mirror area and field of view
(FOV), known as etendue, which is a simple proxy for
the volume in space that can be monitored by different
telescopes for the same exposure time and for a given
intrinsic lumininosity object. We show the FOV, col-
lecting area and number of pixels of a selection of large
etendue survey telescopes in Figure 1. These telescopes
vary from very large FOV or all–sky collections of small
aperture telescopes (“hedgehog” configurations) to large
aperture and large FOV detector mosaics (e.g., LSST).
The small aperture telescopes are able to explore very
fast cadences, but are restricted in practice to bright
objects or the nearby Universe. The large aperture tele-
scopes are able to explore dimmer objects and the more
distant Universe, but have more restricted cadences for
all–sky observations.
The detectors in these large etendue telescopes pro-
duce data at increasingly faster rates. Millions of events,
i.e., objects that are witnessed to change their brightness
or position in the sky, are being detected and reported in
the form of continuous astronomical alert streams (Pat-
terson et al. 2019). These streams create an opportunity
for a new generation of follow–up telescopes to charac-
terize large numbers of astronomical events in a coor-
dinated fashion, ultimately leading to a better under-
standing of the nature of variable phenomena and con-
sequently of the evolution of our local and more distant
Universe.
A new time–domain ecosystem is being built accord-
ingly, where telescopes specialize as either survey or
follow–up telescopes, but also where new digital in-
formation components are developed to connect them
seamlessly. The aggregation, annotation and classifi-
cation of alerts in a rapid and consistent fashion is
done by astronomical alert brokers, such as the Auto-
matic Learning for the Rapid Classification of Events,
ALeRCE, this work; Alert Management, Photometry
and Evaluation of Lightcurves, AMPEL (Nordin et al.
2019); Arizona-NOAO Temporal Analysis and Response
to Events System, ANTARES (Narayan et al. 2018);
Fink;1LASAIR;(Smith et al. 2019) and Make Alerts
Really Simple, MARS.2Different brokers typically spe-
cialize in different science cases. Their main role is to
provide a fast and consistent classification of the alert
1https://fink-broker.readthedocs.io/
2https://mars.lco.global/

The ALeRCE broker 3
Figure 1. FOV vs light collecting area for a selection of ground and space-based survey telescopes currently operational or
planned. The product of the two is called etendue and is indicated by the relative sizes. Note that if a survey contains several
identical telescopes we consider the sum of their etendues. The color of the circles indicates the number of pixels in the main
camera of the instrument, following the color coding on the right. Constant etendue loci are shown as gray dashed lines, with
the specific etendue value shown for each line. See Table B1 for telescope names and references.
stream using all the available data, but also to enable fil-
tering of the stream for different scientific communities.
The fast classification of events is critical for the study of
either short–lived phenomena or the early phases of evo-
lution of longer–lived processes, enabling follow–up ob-
servations to occur fast enough for some physical prop-
erties to be inferred (e.g., Gal-Yam et al. 2014). They
will also contribute to the detection of new astrophysical
phenomena in the form of outliers/anomalies (e.g., Nun
et al. 2016), and will help reveal new sub-populations
among known families of events (e.g., Baron & Poznan-
ski 2017).
An interoperable and agile ecosystem is needed, with
all the relevant parts able to interact automatically to
perform coordinated observations, but also capable of
adapting quickly to new science cases, instruments, or
digital technologies. In this new scenario, follow–up tele-
scopes will listen and react to Target and Observation
Managers (TOMs; e.g., Street et al. 2018). TOMs will
listen to alert broker classified streams, and brokers will
listen to survey telescope alert streams. When follow-
up observations are performed and their results become
available, TOMs will be able to modify their follow-up
strategy, brokers will be able to improve their classifica-
tion, and survey telescopes will be able to change their
surveying strategies, providing a feedback mechanism
for the entire time domain ecosystem to continuously
improve.
1.1. Alert Broker Challenges
Astronomical alert brokers are a new kind of tool in
the interface between astronomy and data science. They
face new challenges including infrastructure, machine
learning (ML), and community integration, but also or-
ganizational aspects which are important in order to ef-
fectively add value to the community. This makes them
important laboratories for testing new ideas on data sci-
ence going even beyond astronomy.

4F¨
orster et al.
In terms of infrastructure, the biggest challenge for as-
tronomical brokers is to ingest, annotate and classify, in
a scalable fashion, the large astronomical alert streams
coming from large etendue telescopes such as ZTF or
LSST. For example, we have received typically between
105–106alerts per night from the public ZTF stream, as-
sociated with 3.7×107objects as of Jun 2020. For com-
parison, LSST is expected to produce about 107alerts
per night and contain more than 109different objects,
which requires a distributed type of database and pro-
cessing. Additionally, there will be a diversity of surveys
streaming alerts which must be cross–matched and clas-
sified in real time (e.g., ZTF, ATLAS, LSST). Thus,
the challenge is to ingest data streams from a diversity
of telescopes in a scalable fashion and to classify them
using their combined information to enable a rapid reac-
tion by follow–up telescopes and a self–consistent anal-
ysis.
In terms of ML development, the challenges are di-
verse. What is an appropriate and relevant taxonomy
for the astronomical community? How should we bal-
ance classification purity and efficiency? How can we
develop ML classifiers and bring them into production
in a reasonable timescale? How should we include cross–
matched information in these classifiers? How can we
train models using data which may be highly unbalanced
and not fully representative of the unlabeled data? For
example, training a classifier with spectroscopically la-
beled data will tend to be biased towards the bright end
of the magnitude distribution. How can we train in a
semi-supervised fashion to take advantage of the unla-
beled data? How can we train using data from a dif-
ferent telescope with a different set of filters/cadences
(i.e., transfer learning and domain adaptation)? How
can we train models using synthetic or augmented data?
How can we detect outliers in a stream of data? All of
these are technically challenging problems which need to
be developed, validated with the community, and then
brought quickly into production.
Integration with the time–domain ecosystem and its
community of users is another important challenge.
First, brokers must be connected with other brokers,
follow–up infrastructure, and data exploration tools.
For this to happen, Application Programming Interfaces
(APIs) must be developed, using Virtual Observatory
(VO) or de facto standards. Second, in order to pro-
duce relevant data products and tools, a frequent inter-
action with the community is needed to provide feedback
and inject new ideas that can help improve the entire
ecosystem. This includes interaction with small to large
projects that interoperate with the community of survey
telescopes, brokers, TOMs, and follow–up telescopes.
A diversity of brokers must be encouraged, avoiding a
winner–take–all solution, and fostering an environment
where new, creative solutions rise faster into production.
1.2. The ALeRCE Broker
The ALeRCE broker is a Chilean-led project which
aims to become a community broker for LSST and other
large etendue survey telescopes. The project is run by an
interdisciplinary team composed by astronomers, com-
puter scientists and engineers, including faculty, post-
doctoral fellows, and students. The broker’s concept
was first announced in 2017 as the natural continuation
of the High cadence Transient Survey (HiTS), in which
we used the Dark Energy Camera on the 4 m Blanco
telescope to discover supernovae (SNe) in real–time by
combining tools from high performance computing and
ML (F¨orster et al. 2016). In 2018 a team of scientists
was consolidated, the key requirements were defined, the
first version of the front–end was developed, a memoran-
dum of understanding was signed with the ZTF project,
and the initial funding was secured. In early 2019, a
dedicated team of engineers was hired to start building
the tools needed to ingest the public ZTF alert stream
in preparation for LSST.
ALeRCE started to systematically classify the ZTF
stream using ML with astrophysically motivated tax-
onomies based on their light curves (S´anchez–S´aez 2020)
since March 2019, and on their image stamps (Carrasco–
Davis 2020) since July 2019. These classifiers are de-
signed to balance the needs for a fast and simple clas-
sification with a subsequent, but more complex classi-
fication. ALeRCE has reported 3088 SN candidates to
the Transient Name Server3, of which 361 have been
spectroscopically confirmed. It has classified 8.5×105
objects into a taxonomy that has expanded into 15
classes, including transient, periodic and stochastic vari-
able sources, and with continuously improving precision
and purity. All of ALeRCE’s data products can be ac-
cessed freely via several dashboards, APIs, or a direct
database connection.
ALeRCE has adopted Agile work methodologies4,
which have been adapted to academic environments by
several groups5. The main ideas behind these method-
ologies can be summarized as: 1) emphasizing individu-
als and interactions over processes and tools, 2) seeking
improvements over sustaining practices, 3) collaboration
over competition, and 4) adaptation to change over fol-
3https://wis-tns.weizmann.ac.il/
4https://agilemanifesto.org/
5https://www.agilealliance.org/resources/experience-reports/
reinventing-research-agile-in- the-academic-laboratory/

The ALeRCE broker 5
lowing a fixed plan. We use development sprints of two
weeks and short daily meetings where product owners
are the leading scientists of the different science cases,
and where scrum masters rotate among a few members
of the team. It has been important to define precise and
achievable objectives and associated deliverables in each
sprint, coupling the team’s skills and motivations around
them. Adopting this methodology has important impli-
cations for the broker, which becomes a continuously
evolving product with regular data and code releases.
All the major components become dynamic: the classifi-
cation taxonomy, as the available data sources grow and
the product owners identify new scientific questions; the
ML classification models, as new training sets and ideas
are brought from development into production; and the
tools and products, in order to adapt to the changing re-
quirements of the community of users. This means that
special attention needs to be given to version control of
the broker pipeline, tools and data products. This is
done via the use of GitHub repositories to track code
changes, and the use of the Semantic Versioning6nam-
ing convention for our future pipeline and associated
data releases, starting with version 1.0.0.
The outline of this document is the following. In Sec-
tion 2we introduce the science goals of the ALeRCE
broker, including a discussion of the broker taxonomy.
In Section 3we describe the ML classifiers used by our
broker. In Section 4we present the pipeline structure
and its associated infrastructure. In Section 5we discuss
our main data products, services and tools. In Section 6
we present some of the main results. Finally, in Section 7
we draw some conclusions and discuss future directions.
2. SCIENCE GOALS
Our primary science goals are the study of three broad
categories of objects: transients, variable stars and ac-
tive galactic nuclei (AGN); we also provide Solar System
object classifications as a secondary science goal.
2.1. Transients
Two important questions which can be answered via
the study of transients are: 1) what is the nature of
explosive phenomena, and 2) what can they teach us
about the dynamics of the Universe. Rapid classifica-
tion is key to answer these questions since it can fa-
cilitate dedicated follow-up observations, either rapid
or slow, spectroscopic or photometric. Rapid follow–
up is critical to understand short–lived transients and
the progenitors of stellar explosions in general, since it
probes the outermost, unprocessed layers of exploding
6https://semver.org/
stars and the possible interaction with the circumstellar
medium (e.g., Yaron et al. 2017;F¨orster et al. 2018).
Early spectroscopy can be used to measure the compo-
sition and velocity structure of their ejecta. Late-time
follow-up, either photometric or spectroscopic, probes
the nature of the progenitor and explosion mechanism
by constraining the composition and velocity structure
of the innermost layers of the star (e.g., Fang et al.
2019). Having large samples of classified transient events
cross–matched with multi–band/messenger or contex-
tual information will help characterize the parameter
space and provide clues of new, unrecognized popula-
tions of events. Furthermore, the ability to cross–match
different streams in real-time, e.g., the LIGO and LSST
streams, will offer possibilities which can lead to new,
unexpected discoveries. Finally, these larger and better
calibrated samples, with well–understood systematics,
can be used for cosmological distance and/or event rate
estimations.
2.2. Variable Stars
Some of the important questions which can be an-
swered via the study of variable stars are: 1) what
is the nature of these systems and the physical mech-
anisms of variability, and 2) what can they teach us
about the structure and formation of our own galaxy,
its satellites, and other galaxies in the Local Group
(e.g., Catelan & Smith 2015, and references therein).
There are various reasons to obtain a uniform and rapid
classification of variable stars. Rapid follow-up of stars
entering/leaving the instability strip or changing their
pulsation modes could provide new insights about the
physics of stellar pulsation (e.g., Clement & Goranskij
1999;Buchler & Koll´ath 2002;Soszy´nski et al. 2014).
Detection and follow-up of eclipses in pulsating stars
can help provide direct stellar mass measurements (e.g.,
Pietrzy´nski et al. 2010,2012). Rapid follow-up of grav-
itational microlensing events can allow the detection of
planets with masses and separations resembling those
in our Solar System (e.g., Bennett & Rhie 1996;Gould
et al. 2010), while microlensing events with timescales
of the order of years can provide clues about the na-
ture of black holes (BHs) and dark matter (e.g., Green
2016). Moreover, microlensing may allow spectroscopic
follow-up of sources that might otherwise have been too
faint for spectroscopy (e.g., Bensby et al. 2020). The de-
tection of eruptive events and the spectroscopic follow-
up immediately after the beginning of the eruption can
provide new insights about the physics of young stel-
lar objects (Contreras Pe˜na et al. 2017;Connelley &
Reipurth 2018). Finally, larger and more distant sam-
ples of consistently classified variable stars (e.g., Gaia

6F¨
orster et al.
Collaboration et al. 2019a) will be key to understand-
ing the tridimensional structure and formation history
of our galaxy, along with that of its neighbors, ranging
from the ultra–faint dwarfs to the Magellanic Clouds
(e.g., D´ek´any et al. 2019;Jacyszyn-Dobrzeniecka et al.
2020a,b;Vivas et al. 2020).
2.3. Active Galactic Nuclei
Some of the most exciting questions which can be an-
swered from the study of AGN are: 1) what drives the
growth of BHs (Alexander & Hickox 2012); 2) what
are the physical mechanisms behind AGN variability
(S´anchez-S´aez et al. 2018;Ross et al. 2018); 3) are there
intermediate-mass BHs (IMBHs; Mezcua 2017;Greene
et al. 2019), with masses between stellar and super-
massive BHs (SMBHs); 4) what is the structure and
size of AGNs (Lawrence 2016); and 5) what can tidal
disruption events (Arcavi et al. 2014) teach us about
BH properties. Rapid classification could help identify
and follow-up optical changing-look AGNs, a popula-
tion which may unlock numerous clues to BH accretion
physics (LaMassa et al. 2015;Graham et al. 2019). Se-
lecting large samples of targets based on their multi–
band variability for reverberation mapping studies can
enable better physical constraints on the BH surround-
ing medium and distance (Peterson et al. 2004). Fast
cadence data can help assemble large samples of IMBHs
candidates (Mart´ınez-Palomera et al. 2020), which are
known to vary on shorter timescales. The early detec-
tion of tidal disruption events can provide independent
constraints on the BH properties that drive these phe-
nomena (Komossa 2015). All of the above can be done
while simultaneously cross-matching the LSST stream
with future surveys that will provide critical additional
information, such as eROSITA (Merloni et al. 2012),
SKA precursors, IceCube (Abbasi et al. 2009), etc. Fi-
nally, exploring larger samples of AGNs that are dimmer
and redder can lead to the discovery of new populations
of events and a better understanding of the AGN phe-
nomena.
3. ML CLASSIFICATION
3.1. Classification Taxonomy
An important component of an automatic classifier
is the taxonomy used for classification, which defines
the classes into which the alert stream will be classified.
Choosing a good taxonomy is about achieving a balance
between a reasonably accurate classifier, which depends
on finding good training sets and the intrinsic separabil-
ity of the classes, and meeting the demands of different
communities of users. More complex taxonomies can
be useful for a larger set of communities, but the addi-
tion of subclasses can lead to potentially less accurate
classification models. The best compromise between the
accuracy of the classifier and the complexity of the tax-
onomy is difficult to define, therefore in order to guide
our choice of taxonomy we performed a survey of the
taxonomies used in other studies that carried out ML
classification of variable astronomical objects.
3.1.1. Light Curve Classifier Taxonomy
First, we consider those works that use only light
curves in their analysis. We divide them into those that
include both persistent variable and transient sources
(Table 1), those that include only persistent variable
objects (Tables B2 and B3), and those that include only
transient objects (Table B4). We examined four publi-
cations that include both transient and persistent vari-
able objects in their taxonomy, 22 publications which
include only persistent variable objects, and 8 publica-
tions which include only transient objects. There were
19 different sources of observational data, mostly for per-
sistent variable sources (Table B5), and five sources of
synthetic data (Table B6).
A large diversity of taxonomies was found, with fewer
classes in general being used in the last five years with
respect to older works. This may be due to the appear-
ance of more exploratory efforts in recent years, which
look for variations from more traditional classification
methods while using fewer classes for simplicity. We
found more classes of persistent variable objects of stel-
lar origin, probably because of the relative abundance of
curated light curve training sets for these classes. The
synthetic data sources were applied mostly for transient
data, probably because of the relative difficulty in find-
ing large numbers of observed transients. A brief de-
scription of the classes is included in the Appendix. The
pulsating star variable classes included in the previous
publications are shown in Tables B7 and B8, other stel-
lar variable sources in Table B9, SMBH-related sources
in Table B10, and transients in Table B11.
In general, there are certain families of objects which
seem to be included consistently among most classifiers,
but whose decomposition into subclasses varies greatly.
Taking this into account we have decided to develop
a hierarchical classifier which groups families of classes
and which will gradually be refined as the amount and
quality of the data grows (S´anchez–S´aez 2020). The
first level of the classifier considers transients, peri-
odic, and stochastic variable phenomena. In the sec-
ond level, the transient branch divides into (class names
between parenthesis) the Type Ia SNe (SNIa), Type
Ib and Ic SNe (SNIbc), Type II and IIn SNe (SNII),
and Super Luminous SNe (SLSN) classes. The peri-

The ALeRCE broker 7
Table 1. Light curve-based ML classifiers that include both transient and persistent variable objects. Note that
S´anchez–S´aez (2020) is an accompanying publication where we describe the ALeRCE Light Curve classifier in more
detail.
Reference Data source Data type #classes classes
S´anchez–S´aez (2020) ZTF Observed 15 SNIa, SNIbc, SNII, SLSN,
(See Section 3.3) AGN, QSO, Blazar, CV/Nova, YSO,
DSCT, RRL, Ceph, LPV, E,
Periodic–Other
Boone (2019) PLAsTiCC Simulated 14 AGN, RRL, E, Mira, Mdwarf, ML,
TDE, kN, SNIa, SNIa-91bg,
SNIax, SNIbc, SNII, SLSN-I
Mart´ınez-Palomera et al. (2018) HiTS Observed 8 NV, QSO, CV, SN, DSCT, E, ROT, RRL
Narayan et al. (2018) OGLE,OSC Observed 7 SN, BPer, RRL, LPV, Ceph, DSCT, DPV
D’Isanto et al. (2016) CRTS Observed 6 CV, SN, Blazar, AGN, Mdwarf, RRL
odic branch divides into the eclipsing binary (E), δScuti
(DSCT), RR Lyrae (RRL), Cepheid (Ceph), long period
variables (LPVs, including Miras, semi–regular and ir-
regular variables), and other (Periodic-Other) classes.
The Periodic-Other class corresponds to periodic objects
which are not members of the E, DSCT, RRL, Ceph or
LPV classes. The stochastic branch divides into host-
dominated AGN, core-dominated AGN or quasi–stellar
objects (QSO), blazars, cataclysmic variables and novae
(CV/Nova), and young stellar objects (YSO).
ALeRCE’s current classification taxonomy is shown
in Figure 2. This figure draws inspiration from the vari-
ability diagram of Eyer & Mowlavi (2008), most recently
updated in Gaia Collaboration et al. (2019b), but signif-
icantly simplified and with a more observationally based
hierarchy, more resolution in the transient classes, and
less resolution in the stellar variability classes. The rea-
son for having more resolution in the transient classes
is that in many cases the reaction time for the photo-
metric or spectroscopic follow–up of these classes needs
to be fast, e.g., to get spectroscopic confirmation or to
characterize a short–lived phase of evolution, while for
the persistent variability classes it is not as common to
require fast follow–up. Thus, our main goal is to provide
a first filter for the expert communities to explore fur-
ther and classify into more complex taxonomies in more
branches of the classification tree.
3.1.2. Stamp Classifier Taxonomy
In addition to the classifiers which work solely on light
curves, there are classifiers which use the pixel informa-
tion contained on the variable object detection images.
Alerts are generated from a difference image which re-
sults from aligning, scaling, convolving and subtracting
the reference image from the science image. We have
listed the ML classification studies which use the object
“image stamps” in Table 2for the classification of images
into either real or bogus, but also as members of more
astrophysically-motivated classes. The latter efforts are
relevant for the taxonomy of our stamp–based classifier,
a classification model which uses as input the first set of
science, template and difference images associated with
a new object in the alert stream7, and which is used
as the first classification step in ALeRCE. Although the
complexity of the taxonomy associated with this classi-
fier is less refined, this early classification is critical to
enable the triggering of fast photometric and spectro-
scopic follow–up and characterization of extragalactic
transient sources. In the case of our stamp–based clas-
sifier (Carrasco–Davis 2020), we have used the classes
SN, AGN, variable star (VS), asteroid and bogus, trying
to mimic how astronomers have historically looked for
transients and variables. SNe tend to be near extended
sources, AGNs are either relatively isolated point–like
sources or at the center of extended sources depend-
ing on luminosity, variable stars are point–like sources
which are frequently near other point–like sources and
are present in both the science and reference images, as-
teroids are present only in the science image and not in
the reference image, and bogus sources are not shaped
like the point spread function of the image.
Finally, we found one publication that uses time series
of image stamps (Carrasco-Davis et al. 2019), following
7Note that the same object can have many associated alerts.

8F¨
orster et al.
Real Bogus
Transient
Stochastic
SN Ia
SN Ib/c SN II Super
luminous
SN
Long
Period
Variable
Young Stellar
Object
Nova /
Cataclysmic
variable
Illustrations: @wandering_astro
Blazar QSO AGN
Pulsating
Periodic
RR Lyrae δ Scuti
Cepheid
Alert
?
Periodic-Other
Eclipsing Binary
Figure 2. The hierarchical taxonomy used by the ALeRCE broker for classifying light curves (v1.0.0). This classifier uses
four models: one which separates transients, stochastic and periodic ob jects; another which separates transients into SNe Ia,
SNe Ib/c, SNe II and Superluminous SNe; another which separates stochastic objects into blazars, QSOs, AGNs and YSOs; and
another which classifies periodic stars into LPVs, Ceph, RRL, DSCT, Es or Periodic–Other.
an approach that combines time series and image stamps
using a convolutional recurrent neural network classifier.
They use seven classes: non–variable, galaxy, asteroid,
SN, RRL, Ceph and E. This type of work could become
more important in the future because it combines spatial
and temporal information as well as simulated and real
data.
3.2. Training Sets
In order to compile training sets, we use only sources
observed by ZTF whose labels have been cross–matched
from different catalogs available in the literature, or
compiled by our collaboration. For each catalog we de-
fine a function which maps the catalog’s taxonomy into
our own taxonomy, allowing us to aggregate labels from
different catalogs into a unified taxonomy. Then, we as-
sign a priority order that defines which labels to use in
case of disagreement between catalogs. These priorities
are based on discussions with community experts, a crit-
ical analysis of the methods that were used to classify
objects (e.g., manual vs. automatic), and an analysis
of which catalogs tend to disagree more with other cat-
alogs, from a visual exploration of catalog label matri-
ces (similar to confusion matrices, but with rows and
columns as the classes in each catalog, potentially with
different taxonomies).
The catalogs we use to extract labels from are, in order
of priority:
1. Cataclysmic variables catalog: compiled by Abril
et al. (2020), including Ritter & Kolb 2003.
2. ROMABZCAT: Multi–frequency catalog of
blazars from Massaro et al. (2015).
3. Catalog of Type I AGNs from Oh et al. (2015).
4. The Million Quasars (MILLIQUAS) Catalogue
from Flesch (2019).
5. Spectroscopically classified SNe in the Transient
Name Server, TNS.8
6. Objects classified as YSOs in Simbad (Wenger
et al. 2000).
7. Catalina Real Time Transient Survey (CRTS) cat-
alog of northern periodic sources (Drake et al.
2014).
8https://wis-tns.weizmann.ac.il/

The ALeRCE broker 9
Table 2. Single image stamp ML classifiers. Empirical data are used in all cases. Note that Carrasco–
Davis (2020) is an accompanying work where we describe the ALeRCE Stamp Classifier in more detail.
Reference Data source #classes classes
Carrasco–Davis 2020 (Section 3.4) ZTF 5 SN, AGN, VS, SN, asteroid, bogus
Duev et al. (2019) ZTF 2 real, bogus
Wright et al. (2017) PanSTARRS1 3 real, asteroid, bogus
Cabrera-Vives et al. (2017) HiTS 2 real, bogus
Kimura et al. (2017) HSC-SSP 2 SNIa, other
du Buisson et al. (2015) SDSS 2 real, bogus
Carrasco et al. (2015) RCS-2 2 stars, QSOs
Bloom et al. (2012) PTF 2 real, bogus
Bailey et al. (2007) PTF 2 real, bogus
8. CRTS catalog of southern periodic sources (Drake
et al. 2017).
9. The LINEAR catalog of periodic variables
(Palaversa et al. 2013).
10. Gaia Data Release 2 (DR2) catalog of variable
stars (Mowlavi et al. 2018).
11. The ASAS-SN catalog of variable stars (Jayas-
inghe et al. 2019).
3.3. The Light Curve Classifier
This classifier computes classification probabilities for
objects with ≥6 detections in gor ≥6 detections in
r. We represent individual light curves as a vector of
features compiled from the literature and new features
developed by the ALeRCE collaboration as described in
S´anchez–S´aez (2020). One of the most relevant new fea-
tures comes from an irregularly sampled autoregressive
model (IAR) introduced in Eyheramendy et al. (2018),
which is able to estimate autocorrelation in irregularly
sampled time series in a statistically robust way. The
classification is done in a hierarchical fashion using a
balanced random forest classifier9, which in our tests
achieved better accuracies than recurrent neural net-
works (e.g., Muthukrishna et al. 2019). As described
before, a given object will be first classified as either pe-
riodic, stochastic or transient and subsequently refined
into 15 different classes as described in Section 3.1. The
latest confusion matrix associated with this classifier can
be seen in Figure 3, described in S´anchez–S´aez (2020).
9Using the imblearn library
3.4. The Stamp Classifier
Inspection of ZTF image stamps suggests that it
should be possible to classify alerts based on the first
detection set of stamps (see Section 3.1.2). Therefore,
we designed and trained a stamp classifier based on a
convolutional neural network with the main motivation
of finding SN candidates using as input the information
contained in the first alert, including the science, ref-
erence and difference stamp set, as well as other meta
data, such as spatial location and data quality metrics.
The stamp classifier (Carrasco–Davis 2020) is able to
discriminate among five classes: SNe, AGN, variable
stars, asteroids, and bogus alerts, achieving 90% accu-
racy on a balanced test set, and a recall of 81% among
spectroscopically confirmed SNe from TNS. To improve
the model interpretability, we added a regularization
term that maximizes the entropy of the predicted prob-
ability for each class, enhancing the different certainties
for each prediction. This model is currently running on
ZTF alerts and its results are publicly available in the
ALeRCE SN Hunter at https://snhunter.alerce.online
(see Section 5.2.1). The confusion matrix associated
with this classifier can be seen in Figure 4, reproduced
from Carrasco–Davis (2020).
3.5. Metrics and Selection of Classification Model
In order to evaluate the classifiers that will go from
initial model training into production, we use a combi-
nation of metrics and tests that take into account the
labeled and unlabeled data. We have found this to be
relevant when using a labeled training set known to be
non–representative of the unlabeled data. First, we com-
pute the test set classification balanced (averaged per
class) accuracy (ratio between correct and total labels),
and F1–score (the harmonic mean between precision and

10 F¨
orster et al.
recall) to take into account the accuracy, precision and
recall of the classifier while considering the class imbal-
ance, which is very important when using observational
data as training sets. Second, we look at the confu-
sion matrix to search for signs of over–representation
of certain classes which may not be evident in the bal-
anced accuracy. Third, for the light curve classifier we
look for classification biases with certain relevant vari-
ables; e.g., looking for a relatively constant recall vs.
apparent magnitude relation for individual classes when
no significant bias exists. Fourth, we compare the ex-
pected and inferred spatial and class distributions of
the unlabeled data to discard models using astrophysi-
cal knowledge. For example, if the classification model
were correct one would expect the spatial distribution
of the different classes to follow known patterns, such
as that most Galactic classes should be concentrated
around the Galactic plane, extragalactic classes should
be homogeneously distributed outside the Galactic plane
due to extinction and source confusion, and asteroids
should be distributed around the ecliptic. Additionally,
we would expect the distribution of class labels in the
unlabeled set to follow known population ratios, for ex-
ample we expect SNe Ia to be more abundant than SNe
Ibc. Therefore, the final choice of a classification model
is made considering all these metrics and tests before
the model is brought into production, i.e., applying the
model using the available infrastructure with our latest
pipeline for nightly operations.
3.6. Stamp and Light Curve Classifier Comparison
As a consistency check between the two aformentioned
classifiers, we compare the distribution of classes of the
Stamp Classifier among those objects classified by the
Light Curve classifier. In Figure 5we show a ma-
trix of Stamp Classifier classes and Light Curve Classi-
fier classes, normalized along the Light Curve Classifier
classes. We can see that there is overall agreement be-
tween the two classifiers, which highlights the comple-
mentarity between our two classifiers, and emphasizes
the value of using the image stamps for early classifica-
tions as shown in Carrasco-Davis et al. (2019).
3.7. Outlier/Novelty Detection
Outlier/novelty detection refers to the automatic iden-
tification of abnormal or unexpected phenomena embed-
ded in data (Faria et al. 2016). We are developing outlier
detection methods experimentally to focus on two prob-
lems: the discrimination of outlier clusters of time series
or image stamps, i.e., cohesive and representative sets
of examples associated with interesting phenomena that
are not characterized in the current training database;
and the detection of unexpected events occurring within
a particular time series. To solve the first problem we
are developing online one–class/semi–supervised outlier
detection methods (Sch¨olkopf et al. 2001;Chapelle et al.
2009;Reyes & Est´evez 2020) to find similarities be-
tween objects and automatically detect outlier phenom-
ena. We are addressing this problem from three different
perspectives: using autoencoders, generative adversarial
networks, and one–class neural networks. To find unex-
pected events within time series, we are using robust on-
line nonlinear filters (Liu et al. 2011;Huentelemu et al.
2016). Traditional methods such as Kalman filters and
kernel filters are being extended to incorporate measure-
ment uncertainties, the heteroscedasticity of the noise,
and the use of state space formulations where states are
unevenly separated in time.
For both problems, Active Learning techniques (Zhu
et al. 2003) are being explored to select sets of the most
uncertain objects and/or events to be shown to human
experts. We are aiming to use information theoretic
feature selection (Est´evez et al. 2009) and feature ex-
traction methods to reduce dimensionality and generate
visualizations that can be presented to the experts.
4. ALERCE PIPELINE AND INFRASTRUCTURE
ALeRCE is currently processing the alert stream
provided by the ZTF survey, but we expect to in-
gest other alert streams in the future, such as those
provided by ATLAS, HATPi10 and LSST (see Fig-
ure 1). The ZTF pipeline and alert distribution sys-
tem are described in Masci et al. (2019) and Patterson
et al. (2019). Alert packets contain image difference
stamps and other metadata, whose detailed description
can be found in https://zwickytransientfacility.github.
io/ztf-avro-alert/schema.html. The ALeRCE system in-
gests these alerts and processes them through a pipeline
which is divided into a combination of sequential and
parallel steps, shown schematically in Figure 6and de-
scribed below.
4.1. Ingestion and Kafka Topics
ZTF alerts are sent as Avro packets11 which con-
tain associated image stamps, metadata and infor-
mation related to previous detections as described in
https://zwickytransientfacility.github.io/ztf-avro-alert/
schema.html. We use Apache Kafka12 to receive the
ZTF alert stream and to communicate information be-
tween the different steps of our pipeline as independent
10 https://hatpi.org/science/
11 https://avro.apache.org
12 https://kafka.apache.org

The ALeRCE broker 11
Figure 3. Confusion matrix obtained with the balanced hierarchical random forest light curve classifier model in S´anchez–S´aez
(2020).
Kafka topics. We use an Apache Zookeeper cluster
with a replication factor of three, following recom-
mended practices, and three independent machines of
Kafka consumers, which are responsible for reading data
from the alert queue. We have set up a Kafka cluster
in Amazon Web Services (AWS) to manage different
topics associated with different steps in the pipeline.
Assigning different topics for each step in the pipeline
has the advantage of allowing for alerts to be grouped
in different batch sizes optimized for performance. For
example, querying the database for several objects si-
multaneously can be faster than doing it sequentially
for a list of objects depending on the type of query, or
in the case of cross–matching, it may be more efficient
to group alerts by their spatial location if the external
catalog is stored hierarchically, e.g., a tessellation of
the sky. Another advantage is that we can configure
each topic independently for performance, e.g., using
different numbers of Kafka partitions per topic.
We have tested different configurations of Kafka pro-
ducers to mimic an LSST–like stream of data, and we
have found that a cluster of three Kafka consumers with
12 partitions each is capable of ingesting all the differ-
ent topics at a rate of 119.7 MB/s, which is about three
times faster than the average alert production rate ex-
pected for LSST.
4.2. Database and Avro Repository
As alerts arrive, we store the original Avro files in
AWS Simple Storage Service (S3) buckets for future
analysis and extract a selection (in order to limit the size
of the database) of the fields contained in these packets
to be added directly to a database using a PostgreSQL
database engine. As the data are processed and object
alerts aggregated, we add different statistics to different
tables. The main tables in our database are:

12 F¨
orster et al.
Figure 4. Confusion matrix obtained with the stamp clas-
sifier model in Carrasco–Davis (2020).
•objects table, which contains basic filter and
time–aggregated statistics such as location, num-
ber of observations, and the times of first and last
detection.
•magstats table, which contains time–aggregated
statistics separated by filter, such as the average
magnitude, or the initial magnitude change rate.
•detections table, which contains the object light
curves including their difference and corrected
magnitudes and associated errors separated by fil-
ter (see Section 4.4).
•non_detections table, which contains the limit-
ing magnitudes of previous non–detections sepa-
rated by filter.
•features table, which contains the object light
curve statistics and other features used for ML
classification and which are stored as json files in
our database.
•xmatch table, which contains the object cross–
matches and associated cross–match catalogs.
•classification tables, which contain the object
classification probabilities, including those from
the stamp and light curve classifiers, and from dif-
ferent versions of these classifiers.
•taxonomy table, that contains details about the
different taxonomies used in our stamp and light
curve classifiers, which can evolve with time.
A webpage containing an updated description of the
different tables can be found in https://alerce.science.
As the volume of alerts grows for different projects, we
expect to migrate some of the previous tables to NoSQL
database engines such as Cassandra or MongoDB. Af-
ter ingestion, the alerts undergo the processing steps
described next.
4.3. Stamp Classification
When an alert from a previously unreported object ar-
rives, its first available image stamps are used to classify
it as either SN, AGN, variable star, asteroid or bogus,
as explained in Section 3.4. Note that if the first de-
tection from an object did not pass the ZTF real/bogus
test, but a subsequent detection did, the first available
image stamp will not be from the former. This stamp
classification is done within one second of the alert being
received and is automatically available in our database
and in the SN Hunter tool (see Section 5.2.1), if the can-
didate is consistent with being a SN. The details of the
stamp classifier are described in a parallel publication
(Carrasco–Davis 2020).
4.4. Light curve Correction
As explained before, ZTF alerts are produced when a
science image contains a significant change with respect
to a reference image, after aligning, scaling, convolving
and subtracting the reference image from the science im-
age. Flux differences with respect to the reference image
are reported as difference magnitudes and an associated
flag (isdiffpos) is included to indicate whether the dif-
ference is positive or negative. In the case of ZTF, a ref-
erence image is defined by a unique reference field iden-
tifier (rfid). If the source was present in the reference
image it is possible to recover its actual apparent magni-
tude from the difference and reference magnitudes. We
do this correction when the nearest catalogued object is
closer than 1.4” (distnr<1.4), providing a flag to indi-
cate whether we think the object is extended based on
PanSTARRS and ZTF shape parameters. The actual
apparent magnitude and associated errors in the case of
an point–like source which was present in the reference
are the following:
mcorr =−2.5 log1010−0.4mref + sgn 10−0.4mdiff (1)
δmcorr =10−0.8mdiff δm2
diff −10−0.8mref δm2
ref 0.5
10−0.4mref +sgn 10−0.4mdiff (2)
where mref is the magnitude of the object in the ref-
erence image, mdiff is the magnitude associated with
the absolute flux difference between the science and
reference images, sgn is the sign of the difference
(isdiffpos), δmref is the error associated with the ref-
erence magnitude, and δmdiff is the error associated with

The ALeRCE broker 13
Figure 5. Fraction of objects predicted to belong to a given Stamp Classifier class (rows), normalized among the objects
predicted to belong to a given Light Curve Classifier class (columns). We considered a sample of 186,794 unlabeled objects
which were classified with the Stamp Classifier Carrasco–Davis (2020) and the Light Curve Classifier S´anchez–S´aez (2020).
ZTF
stream
S3 upload
LC correction LC features
LC classifier Outliers
ALeRCE
stream
Stamp classifier
Cross match
≥ 6 detections
all alerts
Figure 6. ALeRCE pipeline structure from ZTF alert inges-
tion to the ALeRCE streaming of the processed alert. Alerts
ingested from the public ZTF stream are first sent to four
parallel Kafka topics: an Avro backup service in AWS S3, the
stamp classifier for early SN detections, a cross–match step
to gather information from public catalogs, and a light curve
(LC) correction step. The LC correction step is followed by
a LC features computation step, and a LC classifier and out-
lier detection steps, which are only applied to objects with
6 or more detections. Note that the ML classification steps
can also be fed with information from the crossmatch step.
The tables of our database are modified inside the pipeline
steps for subsequent access via APIs.
the difference magnitude. Note that we provide both the
original and corrected photometry. For the corrected
photometry, we include errors values with and without
the term inside square brackets in Equation 2, which
originates from the correlation between the reference
and difference fluxes (see derivation in Appendix A).
It is important to note that if the difference flux is
equal to the reference flux and the sign of the difference
is negative, both the corrected magnitude and associ-
ated errors will diverge, which is a limitation of using a
logarithmic scale for difference fluxes. This should nor-
mally not occur, since an alert is triggered only when
there is a significant difference with respect to the ref-
erence. However, if the reference image contains a tran-
sient source, the difference flux can eventually become
exactly minus the reference flux, and the corrected flux
zero, which will lead to divergences depending on the
noise. We treat these cases by assigning values of 100 to
the corrected magnitudes and their associated errors.
We discuss in detail the derivation of these formulae,
how to include the effect of a change in reference im-
age, and how we treat extended sources in the reference
image in the Appendix A.
4.5. Xmatch
A cross-match step runs in parallel with the stamp
classifier and light curve correction, querying external
catalogs in order to extract additional information about
the objects of interest. The ZTF alert packets already
contain the nearest Solar System, PanSTARRS and
Gaia catalogued sources. In addition to this informa-
tion, we query WISE and SDSS in order to obtain in-
frared and spectroscopic information if available, which
can be critical to better constrain some of the classes
included in our taxonomy. Additional catalogs will be
included as they prove relevant. These queries are done
using the CDS cross–match API13.
4.6. Feature Computation
With the corrected light curves we can compute light
curve characteristics or features based on both the de-
tections and non–detections of a given object, but also
on available crossmatches. Advanced light curve fea-
13 http://cdsxmatch.u-strasbg.fr/xmatch/doc/

14 F¨
orster et al.
tures are only triggered for objects with ≥6 detections
in gor ≥6 detections in r. The features computed
are a significantly extended version of the FATS library
(Nun et al. 2017), called Turbo FATS, which is opti-
mized for computation speed and adds several new fea-
tures. A description of these features, which are con-
tained in the features table of our database, can be
found in S´anchez–S´aez (2020).
4.7. Light Curve Classification
Objects having computed features are then processed
by the light curve classifier described in Section 3.3.
The results of this classifier are obtained within a few
seconds from ingestion for 95% of the objects. For a
larger stream this could be maintained by scaling the
infrastructure given the embarrassingly parallel nature
(i.e., no need of communication between parallel tasks)
of the light curve correction, feature computation and
light curve classification tasks between different alerts.
The current model used for the light curve classifier is
a hierarchical balanced random forest, as described in
S´anchez–S´aez (2020).
After the light curve classification step we perform an
outlier detection step, which as of Jun 2020 is being
actively developed experimentally (see Section 3.7).
4.8. Database Integrity Tests
After the nightly ingestion and processing of the
alerts, we perform a series of database integrity tests
during the day. This consists in reanalyzing the Kafka
topic associated with the last night of observations to
check that no alerts were lost during the processing due
to unexpected errors. If any alerts were missed during
the night, we add them to a specially created Kafka topic
which is then processed by our pipeline until no missing
alerts exist.
5. DATA PRODUCTS AND SERVICES
The ALeRCE broker provides several data products
and services which are constantly growing as we identify
new requirements from our community of users. New
requirements are defined by user stories, informal de-
scriptions of desired features from the perspective of an
end user, which are translated into different data prod-
ucts and services by astronomers in our team follow-
ing an Agile methodology. In this section we list the
most important data products and services provided by
ALeRCE as of Jun 2020, which are summarized in Ta-
ble 3.
5.1. Data Products
The ALeRCE data products can be divided into sev-
eral categories: the tables of a database, a repository
of Avro files, a repository of jupyter notebooks, an out-
put stream of annotated and classified alerts, a GitHub
repository with our open source code, a Grafana dash-
board to monitor the status of the pipeline, our main
webpage, documentation webpages, and tutorial videos
for new users. We provide a brief description of each of
them in what follows.
5.1.1. Database
The tables in our database integrate the informa-
tion about individual objects. A description of the
database can be found in Section 4.2. The tables from
our database are open for direct exploration in read–
only mode as shown in some of our use case jupyter
notebooks (https://github.com/alercebroker/usecases),
although we recommend accessing them using our dif-
ferent APIs for simple queries (see Section 5.2.2). A
detailed description of the tables and schema used in
our database can be found in http://shorturl.at/cJS34.
5.1.2. Avro Repository
Apart from the previous tables, a copy of the original
Avro files contained in the ZTF stream are stored in
AWS S3. These Avro files can be accessed using our
Avro/stamp API.
5.1.3. GitHub Repositories
All of our open source code can be found in the
GitHub repository https://github.com/alercebroker. In
the course of developing this project and as of Jun 2020
we have created 113 repositories, 27 of which have been
made public for our community of users. These repos-
itories can be forked or modified for external use. The
pipeline steps are contained in these repositories and
new version numbers are defined when dockerized ver-
sions of the steps are created.
5.1.4. Use Case Jupyter Notebooks
We have compiled a list of example jupyter notebooks
which show how to use our API or directly access our
database, focused around different science cases, such
as SN, variable stars, AGN, or even asteroid studies.
They can be found at https://github.com/alercebroker/
usecases.
Apart from these notebooks, we have created a
special notebook and associated GitHub repository
for the inspection and submission of SN candi-
dates to TNS (https://github.com/alercebroker/TNS
upload). In this notebook users can interact with Hi-
erarchical Progressive Surveys (HiPS, Fernique et al.
2015) PanSTARRS images to easily select the candi-
date host galaxies using ipyaladin,NED,Simbad, and

The ALeRCE broker 15
Table 3. Summary of ALeRCE data products & services as of Jun 2020.
Type Name Address
Database ALeRCE DB PostgreSQL repository db.alerce.online
GitHub repositories ALeRCE open source repositories http://github.com/alercebroker
Jupyter notebooks Science use cases notebooks http://github.com/alercebroker/usecases
Jupyter notebooks TNS upload notebooks http://github.com/alercebroker/TNS upload
Output stream ALeRCE output Kafka stream Please contact us.
Website ALeRCE main webpage http://alerce.science/
Dashboard ALeRCE Grafana pipeline dashboardahttp://grafana.alerce.online/
Documentation ALeRCE API documentation http://alerceapi.readthedocs.io/en/latest/
Documentation ALeRCE client documentation http://alerce.readthedocs.io/en/latest/
Documentation ALeRCE tutorial videos https://bit.ly/2NHDagc
Web interface ALeRCE explorer http://alerce.online
Web interface SN Hunter http://snhunter.alerce.online
Web interface Crossmatch interface http://xmatch.alerce.online
Web interface ALeRCE reporter http://reporter.alerce.online/
Web interface TOM Toolkit plugin http://tom.alerce.online/
API ZTF DB access http://ztf.alerce.online
API Avro/stamp service http://avro.alerce.online
API ZTF crossmatch service http://xmatch-api.alerce.online
API catsHTM crossmatch service http://catshtm.alerce.online
API TNS crossmatch service http://tns.alerce.online
API Finding chart generator http://findingchart.alerce.online
aRequest access
SDSS DR15. This repository includes a tutorial explain-
ing all the steps required to upload candidates to TNS,
including tutorial videos to guide users in the process.
5.1.5. Output Stream
A real–time output stream is provided to report
database changes as new alerts arrive and are processed
by our pipeline, including an update on the classification
probabilities and basic statistics. Users can connect to
this stream using Apache Kafka upon request.
5.1.6. Grafana Dashboard
A Grafana dashboard is available to monitor the
ALeRCE pipeline and associated database and infras-
tructure (http://grafana.alerce.online). This dashboard
shows the status of the Apache Kafka servers and rele-
vant metrics about the number of alerts being processed,
the PostgreSQL database and associated servers, and
the front–end servers. Access to this dashboard can be
given upon request.
5.1.7. Main Website, Documentation and Tutorial Videos
ALeRCE’s main website, which summarizes all our
data products and services, can be accessed at http://
alerce.science. Documentation for our API services and
client (see Section 5.2.1), and a series of tutorial videos
for our community of users can be found at https://bit.
ly/2NHDagc.
5.2. Services
Apart from the previous data products, several ser-
vices are provided to facilitate the exploration of the
ZTF stream and associated objects. They are divided
into web interfaces, which are web pages that allow the
simple exploration of the alert stream; and APIs, which
power the previous web interfaces and allow for the flexi-
ble integration of ALeRCE into the time domain ecosys-
tem.
5.2.1. Web Interfaces
ALeRCE Explorer (http:// alerce.online)—The ALeRCE
explorer is the main tool to explore the astronomical
objects recovered from the ZTF alert stream. Its land-
ing page consists of two main sections: the Search and
Results sections (see Figure 7). The Search section is
where users can filter objects by selecting their unique
identifier, or by selecting different combinations of clas-
sifier, class, class probability, number of detections, and
sky coordinates. The Results section is where the results
of the filtered objects are shown, sorted by classification
probability or other variables. Clicking on an individual

16 F¨
orster et al.
A1
A2 A3
A4
A5
A6
A7
A8 A9
A10
A11 A12
Search Results
...
B2
B1
B3 B4 B5 B6 B7
Figure 7. The ALeRCE explorer web interface (http:
//alerce.online) initial Search and Results view. The
Search panel allows users to directly filter by object iden-
tifier (A1); by inferred type using either the stamp or light
curve classification models (A2), a given class (A3), and a
minimum classification probability (A4); by the minimum
and maximum number of detections (A5); by minimum (A6)
and/or maximum (A7) discovery date in modified Julian
dates or calendar dates; or by location in the sky using a
cone search defined by a right ascension (A8), declination
(A9), and search radius (A10). The Search button (A11)
submits queries and the Clear button (A12) clears the search
options. The Results panel shows the results of the previous
query. First, it shows the total number of results (B1), which
are displayed in a paginated format. Users can select which
columns to display (B2). The columns shown in this figure
are the object identifier (B3), the number of detections (B4),
the time of first (B5) and last (B6) detection, and the coordi-
nates (B7). Other columns displayed by default (not shown
in this image) are whether the object has cross–matches, and
the stamp and light curve classifier classes and probabilities.
Clicking on an object links to the Object view (Figure 8).
object will take the user to the object view page (see
Figure 8).
The object view page is divided into two tabs: the
General Information and the Cross Matches tabs, with
different panels each (see Figure 8). In the General In-
formation tab users can see some basic statistics about
the object, generate a finding chart, query different cat-
alogs at the position of the object (NED, Simbad, TNS,
PanSTARRS, or SDSS), or quickly see basic TNS infor-
mation about the object. The user can see the object’s
light curve, including detections and non–detections,
with the capability of plotting the raw difference light
curve, a corrected apparent magnitude (which includes
the contribution of the reference image), or a folded ver-
sion of the corrected apparent magnitude using the best–
fitting period. The light curve information can be down-
loaded as comma separated values (CSV), and every
point in the light curve can be hovered over to see more
information, or clicked on to show its associated image
stamp. HiPS images and catalogs around the position of
the object are shown using Aladin, with superimposed
NED and Simbad clickable objects. The science, refer-
ence and difference image stamps associated with any
point in the light curve can be shown in the Stamps
section, where the stamps can be explored by selecting
different dates or hovering over them, seen in full screen,
or downloaded as fits files. The full Avro packet infor-
mation can also be explored. The classification proba-
bilities are shown in the Stamp and Light Curve Clas-
sifier tabs, where a radar plot is used to show the class
probabilities assigned by the light curve or stamp based
classifiers, if available. Finally, in the Cross Matches tab
users can see all the cross–matches contained in the cat-
sHTM set of catalogs for a given separation, which can
be selected manually with a sliding bar (see Figure 9).
The ALeRCE explorer is where most of our web de-
velopment has been focused, including new tools as re-
quested by our community of users, but also new sources
of data which in the future will allow for the multi–
stream exploration of astrophysical objects. We are de-
veloping a modular data exploration library which will
be gradually expanded to include new sources of stream-
ing data14.
SN Hunter (https:// snhunter.alerce.online)—The SN
Hunter platform allows users to visualize and explore
the best and most recent SN candidates (see Figure 10).
These candidates are obtained using the convolutional
neural network which powers the ALeRCE stamp classi-
fier and can be seen in the SN Hunter just seconds after
being received from ZTF. Users can see the spatial dis-
tribution of the candidates in celestial coordinates and
in comparison to the Milky Way plane or the ecliptic, as
well as a table which shows them sorted by classification
probability, discovery date, or number of observations.
Selecting a candidate displays an Aladin HiPS image at
the location of the object, as well as the science, refer-
ence and difference images contained in the Avro file.
The candidates’s unique identifier, coordinates, first ob-
servation properties, and the properties of the closest
PanSTARRS object are also shown, as well as links to
the ALeRCE explorer for the same object, or for NED,
TNS and Simbad sources around the position of the ob-
ject. Users can also see the full alert information con-
14 https://vue-components.alerce.online/

The ALeRCE broker 17
A1
A4
A2
A3
A5
Object information
A9
A7
A8
A10 A11
A12 A13
A14
Light curve explorer
B1
B3
B2
B4 B5
B7 B8 B9 B6
Aladin explorer
C2
C3
C4
C1
C5
E2
E3
Classification probabilities
E1
E4
Stamp explorer
F1 F2 F3
F4
F5
F6
F7
Light curve statistics
D1 D2 D3
iii iii iv
F10
F8
F9
A6
Figure 8. The ALeRCE explorer web interface (http://alerce.online) object view. At the top left, users can switch between
the General information (i, this figure) or the Cross–matches (ii) ob ject views, and at the top right, between different objects
(iii, or the arrow keys) if directed from the results table of a previous query, or to go back to the Search and Results view (iv, or
the escape key). The General information view contains six different panels which we demarcate with colored text. At the
top left, the Object information panel shows the object’s unique identifier (A1); most likely class (A2); coordinates (A3) in
different formats (A7); number of detections and non–detections (A4); and the first and last detection times (A5) in calendar
or modified Julian dates (A6). It also contains links to the finding chart generator tool (A8); the NASA Extragalactic Database
(A9, NED); Simbad (A10); TNS (A11); PanSTARRS (A12); and the SDSS DR15 navigation tool (A13). The latest type, name
and redshift associated with the object in TNS are also shown (A14). At the top middle, the Light curve explorer panel
displays the latest light curve of the object, including both detections and non–detection upper limits in both bands (B1), which
can be turned on/off individually (B2). The light curve can be zoomed in and out (B3) and users can hover over individual
points (B4) to see the exact date, magnitude and alert identifier (B5), or click on to display its associated stamps and full alert
information in the Stamp explorer panel. The light curve and associated data can be downloaded (B6) and users can select
whether to show: the difference magnitude (B7); the apparent magnitude (B8), i.e., corrected by the flux in the reference image;
or the period–folded apparent magnitude (B9), assuming that the light curve is periodic and using a periodogram to compute the
most likely period. At the top right, the Aladin explorer panel shows an interactive Aladin window (C1) with a PanSTARRS
image at the location of the candidate (C2), in this case a confirmed SN near its likely host galaxy (C3). An overlaid catalog of
objects can be clicked on to view more information (C4), such as the host redshift (C5). At the bottom left, the Light curve
statistics panel shows different statistics (D1) computed over the g(D2) and r(D3) bands of the apparent magnitude light
curve. At the middle bottom, the Classification probabilities panel shows the classification probabilities according to our
light curve (E1) or stamp (E2) classifiers, when available. A radar plot of the class probabilities for the taxonomy used in the
classification model (E3) is shown. Hovering over the radar plot displays the numerical values of the probabilities (E4). At the
bottom right, the Stamp explorer panel shows the science (F1), reference (F2) and difference (F3) image stamps associated
with any point in the light curve, which can be downloaded for further analysis (F4), or displayed in full screen mode (F5).
Users can switch between the previous or next stamps in time (F6), or select any particular date (F7) of the light curve which
is contained in the public ZTF stream. Users can select between displaying cross hairs (F8) or simultaneously hovering and
zooming (F9) over the stamps. They can also see the full alert information in the associated alert packet (F10). Note that those
points in the light curve that do not pass the ZTF’s real/bogus test will not have stamps available for display since they do not
trigger an alert in the public stream.

18 F¨
orster et al.
A1
A6A5
A4
A2 A3
iii iii iv
A7
distance
Max
Figure 9. The Object Cross matches view of the ALeRCE
explorer. Labels i, ii, iii and iv as in Figure 8. This panel
allows users to find the closest cross–matching sources in
the catsHTM dataset, given a maximum cross–matching dis-
tance (A1) defined via a sliding bar (A2) or directly via its
numeric value (A3). The closest cross–matches among differ-
ent catalogs (A4) are shown with their associated distances
(A5), allowing for an expanded view of the columns avail-
able in each catalog (A6). For more information, see the
catsHTM (A7) reference (Soumagnac & Ofek 2018).
tained in the original Avro file of the alert by clicking in
Full Alert Information button.
A key feature of the SN Hunter is the ability to receive
feedback from users who have logged in. If a candidate
appears to be bogus, users can label the candidate as
such to further enhance the training set. Moreover, if the
candidate appears to be a SN or extragalactic transient,
the user can label it as a possible SN to be sent to the
ALeRCE reporter tool (see below). The list of possible
SNe can then be explored by the team with our reporter
tool, which can then be used to submit targets to the
TOMs for follow–up.
Reporter (https: // reporter.alerce.online)—The ALeRCE
reporter tool is a platform which serves to manage user
feedback in general (see Figure 11). As of Jun 2020 it
serves three purposes: to manage the feedback provided
by the SN Hunter interface, to connect with the TOM
Toolkit interface, and to manage internal data classifi-
cation challenges. The user feedback provided via the
SN Hunter consists of bogus alert labels, for those alerts
which appear to be bogus; and possible SN alert labels,
for those alerts or groups of alerts which appear to be
originated by extragalactic transients. The connection
of SN candidates with the TOM Toolkit interface is also
done from the reporter tool, sending users to the TOM
Toolkit Interface after clicking on a reported candidate.
Finally, the reporter tool can be used to create data chal-
lenges, manage associated user entries, produce metrics
and confusion matrices, and show leader boards as in
Kaggle. The data challenges are key for the collabora-
tion’s periodic hackathons, where we set different classi-
fication challenges and which motivate the ML team to
develop new ideas and tools.
TOM Toolkit Plugin (https:// tom.alerce.online)—This
platform is used to manage and submit candidates to the
TOM Toolkit (https://lco.global/tomtoolkit/). Users
that have access rights to the ALeRCE reporter can con-
nect with the TOM Toolkit via this interface, allowing
them to submit observational requests with detailed in-
strumental specifications to the queue of different obser-
vatories.
Xmatch Service (http:// xmatch.alerce.online )—ALeRCE
provides a cross–match service which allows users to sub-
mit an arbitrary CSV file with objects and coordinates
of their favorite targets (see Figure 12). After a file is
uploaded, the user is asked to select the names of the
identifier, right ascension and declination columns. Af-
ter this is done the closest objects in ZTF are returned,
adding several columns from the ALeRCE object table
to the submitted objects. A paginated table is shown
for exploration, and the output can be downloaded as a
CSV file.
5.2.2. APIs
All the interactions between the Web Interfaces and
the database or the Avro/stamp repository are done
via APIs. These APIs serve most of ALeRCE’s data
exploration tools following the principle of maximiz-
ing the modularization of our different services. They
are also the key elements which will allow ALeRCE
to integrate seamlessly with the astronomical time–
domain ecosystem. These APIs are documented in
the ALeRCE API Documentation webpage: https://
alerceapi.readthedocs.io/en/latest/. Here we describe
the services available as of Jun 2020:
ZTF Database Access Service (http:// ztf.alerce.online)—
This service allows users to query the ALeRCE database
tables without needing any authentication. This API
includes services to query objects filtered by unique ob-
ject identifier, number of detections, class, class proba-
bilities, coordinates, or detection times. Users can also
get the associated SQL command for a given query, all
the detections for a given object, all the non–detections
for a given object, the classification probabilities for a
given object, or the features used as input for the ML
classifiers for a given object. The documentation can

The ALeRCE broker 19
C1
C4
C2 C3
C5
Aladin explorer
D2
E2 E3
Stamps & user feedback
Top candidates
Alert information
C8
A1
Celestial map
C6
C7
C9 C10 C11
C12
E4
E1
E5
B3
B4
B2
B1
D3
D1
A2 A3
A4 A5 A6 A7
D4
i
Figure 10. The SN Hunter web interface (http://snhunter.alerce.online), which allows users to find the highest stamp–
classification probability and most recent SN candidates in the ZTF alert stream in real–time. This tool is divided into five
panels and is used by our collaboration to select candidates for submission to TNS. Starting at the bottom right, the Top
candidates panel shows a list of the top 10 – 1000 (default 100; A1) SN candidates in terms of their stamp classifier SN
probabilities within the last 1-7 days (default 24 hours; A2). This list can be refreshed at any moment (A3). The results are
shown in a paginated table sorted by either object identifier (A4), discovery date (A5), score or stamp classifier SN probability
(A6), or number of detections (A7). Each candidate can be clicked on for exploration, opening up the top panels. At the bottom
left, the Celestial map panel shows the spatial distribution of all the candidates in the Top candidates panel, with a circle
size proportional to their score (B1) and centered around the currently selected candidate (B2). Also shown are the position of
the ecliptic (B3) and Milky Way plane, where the white contour levels denote crudely the density distribution of Galactic stars
(B4). At the top left, the Alert information panel shows the information about the currently selected candidate, including its
object identifier (C1); coordinates (C2); band (C3), magnitude and time (C4) at first detection; information about the closest
PanSTARRS source, including its identifier (C5), distance (C6), and star galaxy score (C7, varying between 0 and 1 between
galaxies and stars). Links to the ALeRCE explorer Object view (C8), NED (C9), TNS (C10), and Simbad (C11) are provided.
All additional information contained in the alert is also available for exploration (C12). At the middle top, the Aladin explorer
panel provides an interactive Aladin window (D1) centered around the selected candidate (D2), where a host galaxy may be
seen in PanSTARRS DR1 gri color images (D3). Note that although there is a clear host galaxy associated with this candidate,
its closest source is a star (D4), which explains the star galaxy score displayed in C7. Finally, at the top right, the Stamps
& user feedback panel is where the science (E1), reference (E2) and difference (E3) ZTF image stamps are displayed for the
currently selected candidate. If users are logged in using a Google account (i), they can label candidates as possible SNe (E4)
or report them as bogus (E5) in order to improve the stamp classifier training set.
be found in https://alerceapi.readthedocs.io/en/latest/
ztf db.html. This service is used in the ALeRCE ex-
plorer and the SN Hunter (see Section 5.2.1).
Avro/Stamps Service (http:// avro.alerce.online )—This
service allows users to access the alert Avro files and
their associated stamps. The input is the unique ob-
ject identifier and the unique stamp identifier. Users
can get the Avro file, a specific field from an Avro file,
or the science, reference and difference image stamps
contained in an Avro file. The documentation can
be found in https://alerceapi.readthedocs.io/en/latest/
avro.html. This service is used in the ALeRCE explorer
and the SN Hunter (see Section 5.2.1).
ZTF Xmatch Service (http:// xmatch-api.alerce.online)—
This service allows users to submit an arbitrary catalog
and get the nearest ZTF sources, their separation, and
their properties. It is used in the Xmatch interface (see
Section 5.2.1).

20 F¨
orster et al.
Challenger
Reporter
B2
B
A
B3 B4 B5 B6 B7 B8
B1
B9 B10
B11B12
B13
Figure 11. The ALeRCE reporter web interface (http://reporter.alerce.online) is used to manage user input in the ALeRCE
ecosystem. Here we show two types of inputs: the Reporter tool, that manages input labels from the SN hunter, either bogus
(A) or possible SN (B), which in the latter case become candidates to be sent to the TOM; and the Challenger tool, which we use
to manage data classification challenges or hackathons. In the TOM list of possible SNe, users can select a given period of time
of recently reported candidates (B1), which returns a given number of candidates (B2). Users can explore the ob ject identifiers
(B3), number of independent reports for the given candidate (B4), source of the label (B5), date of first (B6) and last (B7)
reports, and possible actions (B8). Among the possible actions, users can explore who has reported a candidate (B9), create
a target for observations in the TOM Toolkit (B10), edit the observational properties of an already created TOM candidate
(B11), or remove the target from the TOM Toolkit (B12). The full list can be downloaded as a CSV file (B13).
catsHTM Crossmatch Service (http:// catshtm.alerce.
online)—This service allows users to do cone searches
to a given location using the catsHTM catalogs (Sou-
magnac & Ofek 2018). This includes cone searches re-
turning all the objects closer than a given distance from
all the catalogs, from a specific catalog, or only the clos-
est object from all or a given catalog. This service is
used in the ALeRCE explorer Cross Matches view (see
Section 5.2.1). The documentation, indicating also a
list of all the available catalogs, can be found in https:
//alerceapi.readthedocs.io/en/latest/catshtm.html.
TNS Crossmatch Service (http: // tns.alerce.online)—This
service allows users to query TNS information about an
object centered around a given position in the sky. It
queries the TNS API and returns the TNS name, type
and redshift, and it is used by the ALeRCE explorer
General Information Tab (see Section 5.2.1).
Finding Chart Service (http:// findingchart.alerce.online )—
This service provides a finding chart associated with a
given object’s unique identifier. It returns a pdf file with
a PanSTARRS reference image indicating the location
of the candidate, as well as the science, reference and
difference image stamps. An example finding chart can
be seen in Figure 13. This service is used in the ALeRCE
explorer (see Section 5.2.1).
Python API Client —We provide a Python client for eas-
ier access to the previous API services. It can be in-
stalled via pip and is documented in https://alerce.
readthedocs.io/en/latest/. You can find examples of
how to use the client in the use case notebooks.
6. RESULTS
The ALeRCE broker has processed 9.7×107alerts
from the public ZTF stream, at a rate of about 5×107
per year, which corresponds to about 1.4×105per night,
or about 5 alerts per second on average. This is ∼80×
less than the expected alert rate of LSST of about 107
per night. However, the ZTF public stream alert pro-
duction rate is not constant, with some nights producing
a few million alerts, which we have been able to ingest
without significant wait time increases. In Figure 14 we
show the distribution of processing times (CPU + wait-
ing times) at the different steps of our pipeline for a typi-
cal ZTF night, including the distribution of ZTF stream-
ing times (time between observation and ingestion) for
comparison. With our current infrastructure we can pro-
cess ZTF alerts in real–time, with classification delays
being dominated by the ZTF streaming times. The lat-
est version of the ALeRCE pipeline has been tested at
rates of about 150 alerts per second, which is approxi-
mately 45% of the expected rates of LSST.

The ALeRCE broker 21
ABC
B1 B2 B3 B4
B5
B6
B7
B8
Figure 12. The cross–match service interface (https://xmatch.alerce.online). Users can input arbitrary catalogs as csv files to
be cross–matched to the ZTF database. The procedure consists in selecting an input catalog CSV file (A), and then indicating
the columns in the file which will be used as identifier (B1), right ascension (B2) and declination (B3), as well as the maximum
radius used to search for the closest cross–matching source (B4). The information provided allows for the partial exploration
of the input file (B5) by a given number of rows (B6) in paginated form (B7). After submitting the catalog (B8), users can
visually explore and download the cross–matched catalog (C).
As of Jun 2020, we have 3.7×107objects, 9.7×107de-
tections, and 1.1×109non–detections in our database.
There are 8.5×105objects classified by the light curve
classifier and 1.9×107objects classified by the stamp
classifier, which started being applied to new alerts
in Aug 2019. For a distribution of the ML inferred
classes in these samples, see our accompanying papers
(Carrasco–Davis 2020;S´anchez–S´aez 2020). The asso-
ciated confusion matrices can be seen in Figures 3and
4and a comparison between the two classifiers can be
seen in Figure 5. Note that our classifiers are contin-
uously improving and that the choice of model is not
based solely on a balanced accuracy score, but also on a
study of the relative frequency and spatial distribution
of classes in the unlabeled set, which we have found to
be an important verification when the training set is not
representative of the unlabeled set.
An important tool to connect ALeRCE with the SN
community of users is the SN Hunter. We have used it
to report 3088 previously unreported astrophysical tran-
sient candidates to TNS, 408 of which have been clas-
sified spectroscopically (with 1% contamination among
those classified spectroscopically, see Figure 15). Among
these, we have found 64 SN candidates rising faster than
0.4 mag/day, and ten faster than 1.0 mag/day, at dis-
covery (see Figure 16). In the process, we have visu-
ally inspected about 20,000 candidates, saving in our
database more than 6500 bogus candidates since Oct
2019 and 1100 transient candidates since Jan 2020, when
we added the Bogus and Possible SN buttons to the SN
Hunter, respectively. The bogus examples have been
used to increase the size and diversity of our training
set and have resulted in significant improvements to the
stamp classifier.
We are slowly building an international community
of users. In order to facilitate the adoption of our
tools by the community, we do not require users to
create accounts to access our system, which makes it
difficult to precisely estimate the number of ALeRCE
users. However, we can use Google Analytics15 to quan-
tify our online community of users. Since Jul 2019,
when Google Analytics was added to the ALeRCE Ex-
plorer and SN Hunter tools, we have had 2.1/1.3 k users
(unique combinations of device and browser, as per the
Google definition) and 7.7/2.2 k sessions in the AleRCE
Explorer/SN Hunter. This does not include the use
of APIs or direct connections to our database. Our
users are currently distributed in 52 countries (see Fig-
ure 17), with the top ones being Chile (27.2%), U.S.
(25.8%), Spain (8.9%), Japan (7.3%), China (6.5%),
and U.K. (5.1%). We are continuously listening to our
users to include new features and we have created new
use case jupyter notebooks for different science cases.
We encourage users to create additional use case note-
books and contribute to our open source repository
(https://github.com/alercebroker/usecases).
15 https://analytics.google.com

22 F¨
orster et al.
A2
A5
A1 A6
A3
A4
A7
A8
A9
Figure 13. A section of the finding chart generated auto-
matically for object ZTF20aaelulu, or SN 2020oi, a Type Ic
SN that occurred in the nearby galaxy M100. The finding
chart shows a PanSTARRS DR1 image (A1) centered around
this object (A2, A3), indicating the direction of the north
and east axes (A4), the coordinates (A5), and the pixel scale
and field size (A6). It also shows the ZTF science (A7), ref-
erence (A8) and difference image stamps (A9). Additional
information, such as the coordinates in a different format,
magnitude statistics, or the time of first and last detection,
are also included. Note that this SN was reported to TNS
by ALeRCE after being classified as a possible SN with just
a single detection using the SN Hunter tool (see Figure 10).
7. DISCUSSION AND CONCLUSIONS
The ALeRCE broker is a new–generation astronomi-
cal alert broker, processing alerts in real–time from ZTF
and preparing to become a community broker for LSST.
We are an interdisciplinary, inter–institutional and in-
ternational team led from Chile, using Agile method-
ologies to develop new digital components for the as-
tronomical time–domain ecosystem in the era of large
etendue telescopes.
In this document we have reported the motiva-
tion, challenges, methodologies and first results of the
ALeRCE broker. The main motivation for ALeRCE
is to provide a rapid classification of events to enable
fast follow–up and characterization, but also to provide
a systematic classification of all variable objects for a
self–consistent analysis of large volumes of events in the
observable Universe. Our primary scientific drivers are
the study of transients, variable stars, and AGN, but
we also provide Solar System object classifications for
further analysis.
Figure 14. Cumulative distribution function (CDF) of ZTF
streaming times compared to the CDF of ALeRCE pipeline
processing times. The ZTF streaming times corresponds to
the difference between the reported observation time and the
alert ingestion time, obtained empirically in a typical night of
operations. The ALeRCE pipeline step elapsed times stands
for the time needed for an alert to move from ingestion to the
completion of a given step, including CPU and wait times. In
this figure we consider an incoming alert rate of about 25 s−1
(c.f., we expect about 5 and 350 s−1for ZTF and LSST on
average, respectively). The embarrassingly parallel nature of
the processing steps suggests that our infrastructure should
scale linearly with the number of incoming alerts to manage
the LSST alert stream.
We describe the infrastructure, processing steps, data
products, tools & services that work in real–time. We
ingest, aggregate, and cross–match the alert stream, and
apply two ML based classifiers to the data (see Sec-
tion 3). First, a stamp classifier is applied to all alerts
associated with previously unreported objects using the
first image stamps as input and a simple taxonomy. Sec-
ond, a light curve classifier with a more complex taxon-
omy is applied to all objects with ≥6 detections in g
or ≥6 detections in r. We are also experimentally ap-
plying outlier detection methods to the data, which we
hope to make public in real–time after significant testing
is done. To our knowledge, ALeRCE was the first pub-
lic broker to provide real–time classification of the ZTF
alert stream into an astrophysically motivated taxonomy
based on the alert image stamps or their light curves.
Regarding the processing of the data, our processing
times per alert are of the order of seconds, significantly
smaller than the current ZTF streaming times (see Sec-
tion 6). Moreover, we have run experiments at ingestion
rates similar to those expected for LSST.
Our database contains object, detection and non–
detection based families of tables, with increasing num-
bers of rows, which are indexed for fast query speeds.

The ALeRCE broker 23
Figure 15. The sample of spectroscopically classified tran-
sients first reported by ALeRCE to TNS, from 3088 SN can-
didates submitted based on their first alert. Out of 408
candidates observed spectroscopically, 401 are confirmed as
SNe, two are unclear, one is a likely SN misclassified as a
galaxy, and four are not SNe. Of the 401 confirmed SNe,
268 are SNe Ia, 86 are SNe II, 29 are SNe Ib/c, 16 are
other peculiar types, and two are classified just as SNe.
The two unclear cases, both of which had SN–like light
curves, are AT 2019yzs (ZTF19adcbnty), which could be a
SN, TDE, or AGN; and AT 2020bdh (ZTF20aaivtof), which
has a very noisy spectrum. The likely galaxy misclassifica-
tion is AT 2019tkd (ZTF19aciiuta), which also has a SN–
like light curve. The four cases confirmed as not SNe are
AT 2019qiz (ZTF19abzrhgq), which is a TDE; AT 2020fx
(ZTF20aadymod), which is a high proper motion star in the
line of sight of a galaxy; AT 2019uzg (ZTF19acssnul), which
is a badly subtracted galaxy, likely a bad zero point cali-
bration; and AT2020csk (ZTF20aaodhzr), which is an AGN.
All relevant tables are public with read–only access, al-
though we recommend accessing them via our differ-
ent APIs which power all our web–based services and
Python client. We provide extensive documentation for
our different data products and services, which can be
found in our main website, http://alerce.science. All
our data products, documentation, tools and services
are summarized in Table 3.
Apart from providing a classified stream of data upon
request, our two most important web services are the
ALeRCE Explorer (https://alerce.online) and the SN
Hunter (https://snhunter.alerce.online), which are pub-
licly available and described in detail in Sections 5.2.1.
The ALeRCE Explorer is the main tool to explore the
objects contained in the ZTF public stream, allowing for
simple queries and providing a user friendly visualiza-
tion of their light curves, cross–matches, image stamps
and classification probabilities. The SN Hunter tool is
Figure 16. Detection magnitude vs. magnitude rise rate at
time of detection for the SN candidates reported to TNS by
ALeRCE based on their first alert image stamps. The color
indicates the peak magnitude of the candidate. We only show
candidates detected rising faster than 0.4 mag/day, a sam-
ple which includes 64 SN candidates. We individually label
ten candidates which rose faster than one mag/day at detec-
tion. Of these candidates, ZTF19abueupg,ZTF20aapjiwl,
ZTF20aapycrh,ZTF20aatzhhl and ZTF20abccixp are SNe
II; ZTF20aaelulu is a SN Ic (shown in the inset plot);
ZTF19abvdgqo is a SN Ib; ZTF19abkrbjt is a SNe Ia;
ZTF20aafdhqm is a transient which coincided with a previ-
ous SN candidate (PS1-13dgc); and ZTF19aadnhaw is prob-
ably a nova based on the shape of its light curve and the
presence of a blue stellar source at its position.
Figure 17. The geographic distribution of users of the
ALeRCE Explorer according to Google Analytics. The num-
ber of users is estimated counting the unique combinations
of device & browser accessing our website. In total, we have
more than 2788 estimated users coming from 52 different
countries accessing the ALeRCE Explorer.
targeted for the transient community to enable a rapid
reaction, allowing users to quickly explore and provide
feedback on the latest SN candidates contained in the
stream. We use this tool to submit new SN candidates
to the TNS at an average rate of about 9 per night, with
3088 reported candidates since Aug 2019. We also use

24 F¨
orster et al.
this tool to select candidates for follow–up via the TOM
Toolkit.
An important goal of ALeRCE is to provide a good
user experience, which should allow for a smooth transi-
tion into a time–domain ecosystem dominated by large
alert streams and automated components where as-
tronomers and data scientists are not replaced, but in-
stead are aided by ML tools to achieve new discoveries.
Thus, we are developing different modular components
for the visualization of the alert stream data, optimized
for usability after testing with our community of users
in regular tutorials and hackathons. The use of Agile
methodologies with a fully dedicated interdisciplinary
team of engineers and astronomers has been critical to
develop ALeRCE at the speed required by the commu-
nity. Collaboration remains essential among brokers to
bring a more diverse set of ideas into our community
and add resilience to the time–domain ecosystem in the
era of large etendue telescopes.
One of the biggest challenges ahead for ALeRCE is
the ability to scale to significantly larger streams, from
∼1.4×105alerts per night to >107alerts per night;
and with significantly more objects generating alerts,
from a few 107objects to >109objects. For this, we
will need to migrate some of our tables from a SQL,
centralized database engine, to a NoSQL, distributed
database engine (e.g., Cassandra, MongoDB). We are
running different tests to determine the efficiency and
cost of the different available solutions in collaboration
with other brokers (Fink). Another important challenge
is to determine what fraction of our storage and comput-
ing services should be located in the cloud (e.g., AWS,
where we currently operate some of our services) vs on–
premise infrastructure. It seems likely that the answer
will be a hybrid solution, with cloud and on–premise in-
frastructure optimized for a better user experience while
minimizing the operational costs.
Achieving more complex taxonomies in an era of
multi–stream, multi–messenger astronomy is another
important challenge ahead. In fact, the large number
of events expected, combined with the addition of het-
erogeneous streams spanning different depths, cadences,
wavelengths, and messengers will likely unveil new pop-
ulations which would not have been possible to identify
otherwise. Encompassing the full diversity of variable
classes in the Universe with a fixed taxonomy is un-
feasible, and thus our taxonomy will continue to grow
and evolve with time. Eventually, a combination be-
tween domain knowledge via supervised training, with
unsupervised, more data–driven taxonomies, will be-
come necessary. Training and classifying with missing
data, as most streams of data will be sparse in compar-
ison to that of LSST, will also become important.
Regarding the challenges of ML classification, we are
trying different strategies. We are introducing new fea-
tures, e.g. a complex number extension to the IAR
model that allows for positive as well as negative au-
tocorrelation (CIAR, Elorrieta et al. 2019), further ex-
panded to bivariate or higher dimensional time series
and to include different covariance structures. From
these models we expect to extract useful features for
classification, as well as be able to do prediction, in-
terpolation and forecasting on time series. We are also
testing ways to combine real, augmented and simulated
data; new ways to combine and expand our Stamp and
Light Curve classifiers; or different recurrent neural net-
works applied to the light curve (e.g., Muthukrishna
et al. 2019) and images stamp series (e.g., Carrasco-
Davis et al. 2019); or different outlier detection methods.
Finally, we note that, given the continuously evolving
nature of ALeRCE, this document provides a snapshot
of the current status of ALeRCE as of Jun 2020. We
are constantly listening to our community of users in an
effort to introduce new data products, tools and services.
Our preferred way of communication is through issues
in our GitHub repositories (https://www.github.com/
alercebroker), but users can also contact us directly via
https://alerce.science.

The ALeRCE broker 25
ACKNOWLEDGMENTS
The authors acknowledge support from the Chilean
Ministry of Economy, Development, and Tourism’s
Millennium Science Initiative through grant IC12009,
awarded to the Millennium Institute of Astrophysics
(FF, GCV, ECN, PAE, PSS, JA, FEB, RCD, MC, FE,
SE, PH, GP, ER, IR, DR, DRM, CV, IAM, NA, JB, AC,
DDC, CDO, RK, AM, WP, MPC, LSG, AS, CSC, JRV)
and from the National Agency for Research and Devel-
opment (ANID) grants: BASAL Center of Mathemati-
cal Modelling AFB-170001 (FF, ECN, PAE, IR, DRM,
CV, IAM, JCM, AM, LSG, JSM, CSC, EV) and Centro
de Astrofsica y Tecnologas Afines AFB-170002 (FEB,
MC, DDC, PSS); FONDECYT Regular #1200710 (FF),
#1190818 (FEB), #1200495 (FEB), #1171273 (MC),
#1201793 (GP); FONDECYT Postdoctorado #3200250
(PSS) and #3200222 (DDC); ANID infrastructure funds
QUIMAL140003 and QUIMAL190012; Magister Na-
cional 2019 #22190947 (ER). We acknowledge support
from REUNA Chile who hosts and maintains some
of our infrastructure. This work has been possible
thanks to the use of AWS-U.Chile-NLHPC credits. Pow-
ered@NLHPC: This research was partially supported
by the supercomputing infrastructure of the NLHPC
(ECM-02).
Software: Aladin (Bonnarel et al. 2000), Apache
ECharts16, Apache Kafka17, Apache Spark (Zaharia
et al. 2016), ASTROIDE (Brahem et al. 2020), Astropy
(Astropy Collaboration et al. 2013), catsHTM (Sou-
magnac & Ofek 2018), Dask (Rocklin 2015), FATS (Nun
et al. 2017), Grafana18, Imbalanced-learn (Lemaˆıtre
et al. 2017), ipyladin (Boch & Desroziers 2020), Jupyter
(Kluyver et al. 2016), Keras (Chollet et al. 2015),
Matplotlib (Hunter 2007), NED (Steer et al. 2017),
P4J (Huijse et al. 2012), Pandas (McKinney et al.
2010), Prometheus19, Python (Van Rossum & Drake Jr
1995), scikit-learn (Pedregosa et al. 2011), Simbad-CDS
(Wenger et al. 2000), Tensorflow (Abadi et al. 2016),
Vue20 , Vuetify21, PostgreSQL22, XGBoost23.
APPENDIX
A. LIGHT CURVE CORRECTION DERIVATION
A.1. Light Curve Fluxes
An alert is originated when a significant flux is detected at some location of a difference image between a science and
reference images. In the ZTF alert stream, the difference and reference fluxes are reported for every alert. The science
flux is not reported, but it can be recovered from the difference and reference images. The difference flux is reported
by its absolute magnitude, mdiff , and sign, sgn; and the reference flux is reported by the PSF photometry magnitude,
mref ,of the closest source in the reference, with associated errors, distance and shape parameters. This leads to three
types of cases: 1) the closest source in the reference coincides with the location of the alert, and it is unresolved; 2)
the closest source in the reference coincides with the location of the difference image alert, but it is resolved; and 3)
the closest source does not coincide with the position of the difference alert. In 1) the science flux can be recovered
exactly, in 2) it can be recovered plus a constant which depends on how much contamination from an extended source
occurs in the reference, and in 3) one needs to assume that the science flux is equal to the difference flux. These cases
are typically represented by variable stars (1), AGNs (2), or transients (3). Since it is not possible to know a priori
16 https://echarts.apache.org
17 https://kafka.apache.org/
18 https://grafana.com/
19 https://prometheus.io/
20 https://vuejs.org/
21 https://vuetifyjs.com/
22 https://www.postgresql.org/
23 https://xgboost.readthedocs.io/

26 F¨
orster et al.
which correction should be applied to each object, e.g., it is difficult to distinguish an AGN from a nuclear transient
until the flux evolution can be observed, we report both the corrected photometry, which is useful for variable stars
and AGNs, and the uncorrected photometry, which is useful for transients.
If the reference source is resolved, its reported flux contains two components: a variable/compact component, which
is normally the object of study, and a static/extended component, which is difficult to separate using only the ZTF
photometry. Because of the convolution done during the image difference process, the extended component should not
contribute to the difference flux. Then, we note the following relations:
fref =fext
ref +fvar
ref ,(A1)
fsci =fext
sci +fvar
sci ,(A2)
sgnfdiff =fvar
sci −fvar
ref ,(A3)
where fref is the reference flux, fsci is the science flux, sgn is the sign and fdiff is the absolute value of the difference
flux, fext
ref is the contribution from the extended component in the reference image, fvar
ref is the contribution of the
variable component in the reference image, fext
sci is the contribution from the extended component in the science image,
and fvar
sci is the contribution of the variable component in the science image. Note that the contribution of the extended
component can vary between the reference and science images due to seeing effects, which can create an artificial source
of variability. The scientifically relevant component for variability studies is the flux of the compact component, but
it is difficult to separate it from the extended component. The second best alternative is to recover the flux of the
compact component plus a constant contribution from the extended component. For this we can define an effective
science flux, ˆ
fsci:
ˆ
fsci ≡fext
ref +fvar
sci (A4)
=fref + sgn fdiff ,(A5)
which considers the same contribution of the extended component at all times. If the reference image changes, we can
introduce a new effective science flux, ˆ
fref,0, that considers the contribution from the extended component from the
first reference image used to generate alerts:
ˆ
fsci,0=fext
ref,0+fvar
sci (A6)
=ˆ
fsci +fext
ref,0−fext
ref ),(A7)
where fext
ref,0is the (unknown) contribution from the extended component from the first reference image. Note that the
expected value from the second term is zero.
A.2. Light Curve Variances
The computation of errors of the science flux must take into account the relation between the difference and ref-
erence fluxes, which are correlated. We can estimate the variance of the effective science flux, V[ˆ
fsci], starting from
Equation A5 and using Equations A1 and A3:
V[ˆ
fsci] = V[fref + sign fdiff ] (A8)
=V[fref ] + V[fdiff ] + 2 Cov[fref ,sign fdiff ] (A9)
=V[fref ] + V[fdiff ] + 2 Cov[fext
ref +fvar
ref , f var
sci −fvar
ref ] (A10)
=V[fref ] + V[fdiff ]−2V[fvar
ref ].(A11)
Note that the variance due to sky emission is contained in the first two terms of Equation A16. One can also include
additional terms in Equation A10 to reflect the contribution of the sky, but because these terms are not correlated
they have no additional contribution in the covariance. We can expand Equation A11 to get the following:
V[ˆ
fsci] = V[fref ] + V[fdiff ]−2V[fvar
ref ] (A12)
=V[fext
ref +fvar
ref ] + V[fdiff ]−2V[fvar
ref ] (A13)
=V[fext
ref ] + V[fvar
ref ] + 2 Cov[fext
ref , f var
ref ] + V[fdiff ]−2V[fvar
ref ] (A14)
=V[fext
ref ] + V[fvar
ref ] + V[fdiff ]−2V[fvar
ref ] (A15)
=V[fdiff ]−V[fvar
ref ] + V[fext
ref ].(A16)

The ALeRCE broker 27
and in the case of a change in the reference image, using Equation A7,A16 and A4:
V[ˆ
fsci,0] = V[ˆ
fsci + (fext
ref,0−fext
ref )] (A17)
=V[ˆ
fsci] + V[fext
ref,0] + V[fext
ref ]−2 Cov[ ˆ
fsci, f ext
ref ] (A18)
=V[fdiff ]−V[fvar
ref ] + V[fext
ref ] + V[fext
ref,0] + V[fext
ref ]−2 Cov[fext
ref +fvar
sci , f ext
ref ] (A19)
=V[fdiff ]−V[fvar
ref ] + V[fext
ref,0].(A20)
To summarize, we show Equations A5,A7,A16,A20:
ˆ
fsci =fref + sgnfdiff
ˆ
fsci,0=ˆ
fsci + (fext
ref,0−fext
ref )
V[ˆ
fsci] = V[fdiff ]−V[fvar
ref ] + V[fext
ref ]
V[ˆ
fsci,0] = V[fdiff ]−V[fvar
ref ] + V[fext
ref,0].
A problem with these formulae is that neither the variable nor extended components are known. However, they led
us to consider the following cases:
1. The contribution from the extended component is negligible in all the reference images:
fext
ref = 0
⇒
ˆ
fsci,0=ˆ
fsci =fref + sgn fdiff (A21)
V[ˆ
fsci,0] = V[ˆ
fsci] = V[fdiff ]−V[fref ].(A22)
2. The contribution from the extended component is similar in all the reference images, and its contribution is
similar to that from the variable component:
fext
ref,0=fext
ref &fvar
ref =fext
ref
⇒
ˆ
fsci,0=ˆ
fsci =fref + sgn fdiff (A23)
V[ˆ
fsci,0] = V[ˆ
fsci] = V[fdiff ].(A24)
3. The contribution from the extended component is similar in all the reference images, and its contribution is
dominant over the variable component:
fext
ref,0=fext
ref &fvar
ref = 0
⇒
ˆ
fsci,0=ˆ
fsci =fref + sgn fdiff (A25)
V[ˆ
fsci,0] = V[ˆ
fsci] = V[fdiff ] + V[fref ].(A26)
A visual inspection of variable star light curves confirms that Equation A22 is a better approximation in the case
where there is no contribution from an extended component. In the case of AGNs, we have found that Equation A24
appears to be a better reflection of the measurement errors, which is consistent with having a similar contribution
from the extended and variable components. In the case of transients, the extended component dominates the flux in
the reference, but for these cases the scientifically relevant flux is the difference flux and its error. For this reason, we
report the difference flux with its error, as well as the effective science flux with the errors (after a conversion of the
fluxes to magnitudes) from Equations A22 and A24 for every object where it is possible to correct the photometry,
letting the users decide which flux and error to use for their particular science.

28 F¨
orster et al.
A.3. Light curve magnitudes
The corrected photometry magnitude results from adding/subtracting the fluxes from the reference and difference
in the same unit system and then converting to magnitudes. We can compute ˆ
fsci by transforming the reference and
difference magnitudes using the zero points of the science image:
ˆ
fsci =fref + sgn fdiff = 10
ZPsci−mref
2.5+ sgn 10
ZPsci−mdiff
2.5,
where ZPsci is the zero point of the science image. This implies that the effective science magnitude, ˆmsci, will be:
ˆmsci =−2.5 log fsci + ZPsci
=−2.5 log10
ZPsci−mref
2.5+ sgn 10
ZPsci−mdiff
2.5+ ZPsci
=−2.5 log10−
mref
2.5+ sgn 10−
mdiff
2.5.(A27)
Finally, we show the reported errors for Equations A22 and A24:
δˆmsci =10−0.8mdiff δm2
diff −10−0.8mref δm2
ref 0.5
10−0.4mref + sgn 10−0.4mdiff ,(A28)
to be used when there is no significant contribution from an extended component; or
δˆmsci =10−0.4mdiff δmdiff
10−0.4mref + sgn 10−0.4mdiff ,(A29)
to be used when there is a contribution from an extended component, assumed to be similar to the variable component.
B. TABLES
Table B1 provides the list of telescopes that was used in preparing Figure 1, along with their names and a relevant
accompanying reference.
Tables B2,B3, and B4 refer to a number of studies in which light curves were used to perform ML-based classification
of variable and transient sources. Tables B2 and B3 both refer to studies in which only persistent variable star classes
were used; the former refers to papers published between 2017-2019, whereas the latter includes studies that appeared
in print before 2017. Table B4, in turn, refers to those studies in which only transient sources were considered. These
three tables have the same structure, with the reference given in the first column, an acronym for the source of the data
given in the second column (with keys provided in Tables B5 and B6 for empirical and synthetic data, respectively),
the number of classes considered shown in the third column, and the fourth column displaying acronyms representing
the actual classes that were considered in each case. These acronyms, along with the classes that they are intended to
represent, are laid out in Tables B7 through B11.
In the case of Tables B7 and B8, the pulsating variable star classes are shown. Table B7 includes pulsating stars in
the upper and lower main sequence, Cepheids, RR Lyrae, blue subdwarfs, and compact (WD) pulsators. Table B8, in
turn, includes red giant and supergiant pulsators.
Table B9 presents a number of additional stellar variability classes, including eclipsing, eruptive, cataclysmic, and
rotational variables. Additional classes that are shown in this table include microlensing events, R CrB stars, Be stars,
and X-ray binaries, among others.
Primarily extragalactic variable sources are shown in Tables B10 and B11. In the case of B10, the variability is
typically related to the presence of SMBHs, as in the case of AGNs and QSOs. Table B11, in turn, includes primarily a
variety of SN classes, although a few transient events of non-SN origin, such as TDEs and kilonovae, are also included.
We emphasize that the classes and associated taxonomies that are implied by Tables B2 through B11 do not reflect
our own choices, but are rather simply a summary of what has been used in the ML literature to date. In particular, the
reader should be aware that the list of classes, as given, suffers from several shortcomings, such as being incomplete,
containing redundant entries, and including classes that may not be sufficiently well defined. Still, our best effort to
interpret what the different authors have intended to express in each case is reflected in these tables, with definitions
given following, among others, the General Catalog of Variable Stars (GCVS; Kholopov et al. 1998), Variable Star
Index (VSX; Watson et al. 2006), and the broad overview of stellar variability classes presented in Catelan & Smith

The ALeRCE broker 29
Table B1. Selection of telescopes shown in Figure 1.
Short name Long name Reference
ASAS-SN All–Sky Automated Survey for Supernova Kochanek et al. (2017)
ATLAS Asteroid Terrestrial-impact Last Alert System Tonry et al. (2018)
BlackGEM BlackGEM https://astro.ru.nl/blackgem/
Blanco-DECam V´ıctor Blanco telescope – Dark Energy Camera Flaugher et al. (2015)
Clay-MegaCam Clay Telescope – Megacam McLeod et al. (2015)
CFHT-MegaCam Canada France Hawaii Telescope – Megacam Boulade et al. (2003)
CRTS Catalina Real–Time Transient Survey (CSS, MLS, SSS) Drake et al. (2009)
Euclid Euclid Mission Laureijs et al. (2011)
Evryscope Evryscope – South Law et al. (2015)
Gaia Gaia Mission Gaia Collaboration et al. (2018)
HATPI HATPI https://hatpi.org/science/
Kepler Kepler Mission Borucki et al. (2010)
KMTNet Korea Microlensing Transient Network Kim et al. (2016)
KISO Kiso Observatory Morokuma et al. (2014)
LS-QUEST La Silla 4000 ESO Schmidt Telescope – QUEST camera Vivas et al. (2004)
LSST Vera C. Rubin Observatory Legacy Survey of Space and Time LSST Science Collaboration et al. (2009)
PanSTARRS Panoramic Survey Telescope and Rapid Response Response System Kaiser et al. (2002)
PTF Palomar Transient Factory Law et al. (2009)
SDSS Sloan Digital Sky Survey York et al. (2000)
Subaru-HSC Subaru telescope – Hyper Suprime-Cam Aihara et al. (2018)
SkyMapper SkyMapper Southern Sky Survey Keller et al. (2007)
TESS Transiting Exoplanet Survey Satellite Ricker et al. (2015)
VISTA Visible and Infrared Survey Telescope for Astronomy Dalton et al. (2006)
VST-OmegaCam VLT Survey Telescope – OmegaCam Cappellarao (2005)
WFIRST Wide Field Infrared Survey Telescope Spergel et al. (2015)
(aka Nancy Grace Roman Space Telescope)
ZTF Zwicky Transient Facility Bellm et al. (2019)
(2015). In the future, as the ALeRCE project matures, we will work towards producing and refining our own taxonomy,
which we will perfect along the way as we enter the LSST era.
REFERENCES
Aartsen, M. G., Ackermann, M., Adams, J., et al. 2017,
Journal of Instrumentation, 12, P03012,
doi: 10.1088/1748-0221/12/03/P03012
Abadi, M., Barham, P., Chen, J., et al. 2016, in 12th
{USENIX}Symposium on Operating Systems Design
and Implementation ({OSDI}16), 265–283
Abbasi, R., Ackermann, M., Adams, J., et al. 2009, Nuclear
Instruments and Methods in Physics Research Section A:
Accelerators, Spectrometers, Detectors and Associated
Equipment, 601, 294 ,
doi: https://doi.org/10.1016/j.nima.2009.01.001
Abbott, B. P., Abbott, R., Abbott, T. D., et al. 2017,
ApJL, 848, L12, doi: 10.3847/2041-8213/aa91c9
Abramovici, A., Althouse, W. E., Drever, R. W. P., et al.
1992, Science, 256, 325, doi: 10.1126/science.256.5055.325
Abril, J., Schmidtobreick, L., Ederoclite, A. r., &
L´opez-Sanjuan, C. 2020, MNRAS, 492, L40,
doi: 10.1093/mnrasl/slz181
Acernese, F., Agathos, M., Agatsuma, K., et al. 2015,
Classical and Quantum Gravity, 32, 024001,
doi: 10.1088/0264-9381/32/2/024001

30 F¨
orster et al.
Table B2. Light curve based ML classifiers that include only persistent variable classes (more than 2 classes) between 2017
and 2019. Class abbreviations are defined in Tables B7 to B11
Reference Data source #classes classes
Rimoldini et al. (2019)Gaia DR2 18 E, CV, RSCvn, BLAP,
Mira+SR, DSCT+SXPh, RRL(ab, c, d, Ad),
CephCl, ACEP, CephII,
Low amp.:DSCT+GDOR, ELL, OSARG, FL+ROT, Other
Tsang & Schultz (2019) ASAS-SN 8 DSCT, RRL(ab, cd), Ceph, E, ROT,
Mira, SR
Jayasinghe et al. (2019) ASAS-SN 10 Ceph, DSCT, E(EW,EA—EB,EB), RRL(ab,c),
M, SR, Irregular
Hosenie et al. (2019) CSDR2 12 RRL(ab, c, d), Blazhko, E(C+SD,D),
ROT, LPV, DSCT, Ceph(II,A)
Johnston et al. (2019) UCR 3 RRL, Ceph, E
LINEAR 5 RRL(ab, c), DSCT, E(C,SD)
Aguirre et al. (2019) OGLE+VVV 9 Ceph(F, 01), RRL(ab, c),
+CoRoT E(C, SD+D), Mira, SR, OSARG
Castro et al. (2018) MACHO 8 NV, QSO, BeS, Ceph, RRL, E, ML, LPV
OGLE 6 Ceph, CephII, RRL, E, DSCT, LPV
Naul et al. (2018) ASAS 5 RRLab, Ceph, SR, BPer, WUMa
LINEAR 5 DSCT, RRL(ab, c), BPer, WUMa
MACHO 8 Ceph(F, O1), LPVW, RRL(ab, c, e, GB)
Valenzuela & Pichara (2018) OGLE 8 Ceph(CL, II, A), RRL, LPV, DPV, DSCT, E
MACHO 11 RRL(ab, c, e, GB), Ceph(F, O1),
LPVW(A, B, C, D), E
Mahabal et al. (2017) CSDR2 7 E(C, SD), RRL(ab, c, d), RSCVn, LPV
Benavente et al. (2017) EROS, 5 Ceph, E, QSO, RRL, LPV
MACHO, HiTS
Zinn et al. (2017) OGLE 8 Mira, QSO, SR, OSARG, Ceph(F, O1),
RRL(ab+d, c+e)
Actis, M., Agnetta, G., Aharonian, F., et al. 2011,
Experimental Astronomy, 32, 193,
doi: 10.1007/s10686-011-9247-0
Aguirre, C., Pichara, K., & Becker, I. 2019, MNRAS, 482,
5078, doi: 10.1093/mnras/sty2836
Aihara, H., Arimoto, N., Armstrong, R., et al. 2017,
Publications of the Astronomical Society of Japan, 70,
doi: 10.1093/pasj/psx066
Aihara, H., Arimoto, N., Armstrong, R., et al. 2018, PASJ,
70, S4, doi: 10.1093/pasj/psx066
Alcock, C., Allsman, R. A., Alves, D. R., et al. 2000, ApJ,
542, 281, doi: 10.1086/309512
Alexander, D. M., & Hickox, R. C. 2012, NewAR, 56, 93,
doi: 10.1016/j.newar.2011.11.003
Andreoni, I., Cooke, J., Webb, S., et al. 2020, MNRAS,
491, 5852, doi: 10.1093/mnras/stz3381
Arcavi, I., Gal-Yam, A., Sullivan, M., et al. 2014, ApJ, 793,
38, doi: 10.1088/0004-637X/793/1/38
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J.,
et al. 2013, A&A, 558, A33,
doi: 10.1051/0004-6361/201322068
Atwood, W. B., Abdo, A. A., Ackermann, M., et al. 2009,
ApJ, 697, 1071, doi: 10.1088/0004-637X/697/2/1071
Baglin, A., Auvergne, M., Boisnard, L., et al. 2006, in 36th
COSPAR Scientific Assembly, Vol. 36, 3749
Bailey, S., Aragon, C., Romano, R., et al. 2007, ApJ, 665,
1246, doi: 10.1086/519832
Baron, D., & Poznanski, D. 2017, MNRAS, 465, 4530,
doi: 10.1093/mnras/stw3021
Bass, G., & Borne, K. 2016, MNRAS, 459, 3721,
doi: 10.1093/mnras/stw810

The ALeRCE broker 31
Table B3. Light curve based ML classifiers that include only persistent variable objects (more than 2 classes) before
2017. Class abbreviations are defined in Tables B7 to B11
Reference Data source #classes classes
Kim & Bailer-Jones (2016) MACHO, 19 DSCT, RRL(ab, c, d, e),
LINEAR, ASAS Ceph(F, O1, other, II), E(C, SD, D),
LPV(MAGBC, MAGBO, OSARGAGB,
OSARGRGB, SRAGBC, SRAGBO), NV
Mackenzie et al. (2016) OGLE 6 Ceph(CL, II), RRL, E, DSCT, LPV
MACHO 8 NV, QSO, BeS, Ceph, RRL, E, ML, LPV
Pichara et al. (2016) MACHO 8 BeS, Ceph, E, LPV, ML, NV, QSO, RRL
EROS 11 E, RRL, Ceph(F, O1, DM, II),
LPV(OSARGRGBO, SRAGBO,
SRAGBC, MAGBC, MAGBO)
Nun et al. (2016) MACHO 8 NV, QSO, BeS, Ceph, RRL, E, ML, LPV
Bass & Borne (2016)Kepler 14 ACT, BCep, Ceph, DSCT, E, ELL, GDor, ROT,
RRL(ab, c), RVTau, SPB, SR, MISC/NV
Faraway et al. (2016)
K¨ugler et al. (2015) OGLE 3 Ceph, E, RRL
ASAS 7 Mira, RRLab, E(C, D, SD), DSCT, CephF
Kim et al. (2014) EROS-2 26 DSCT, RRL(ab, c, d, e), Ceph(F, O1, Other), CephII
E(C, SD, D, SD+D, Other), BeS, QSO, NV
LPV(MAGB(C, O), OSARGAGB(C, O),
OSARGRGB(C, O), SRAGB(C, O))
Pichara & Protopapas (2013) SAGE, 2MASS, 7 NV, QSO, BeS, Ceph, RRL, E, LPV
UBVI, MACHO
Richards et al. (2012) ASAS 28 DSCT, SXPh, RRL(ab, c, d), Ceph(CL, MM, II),
Mira, SR, LPVW(A, B), RVTau, BCep, RSG,
BPer, BLyr, WUMa, ChemPec, ELL, RSCvn,
HAeBe, CTTau, WLTTau, RCB, LBV, BeS
Debosscher et al. (2009) CoRoT 29 sdBV, DSCT, LBoo, SXPh, roAp, GDor,
RR(ab, c, d), Ceph(CL, DM, II), RVTau,
Mira, SR, PVSG, BCep, SPB, E,
ChemPec, ELL, FUOri, HAeBe, TTau,
LBV, WR, XB, BeS, LAPV
Debosscher et al. (2007) OGLE 35 DAV, DBV, sdBV, GWVir,
DSCT, LBoo, SXPh, roAp, GDor,
RRL(ab, c, d), Ceph(Cl, DM, II),
PVSG, Mira, SR, RVTau, BCep, SPB,
E(C, SD, D), ChemPec, ELL,
FUOri, HAeBe, TTau, LBV,
SLR, WR, XB, CV, BeS

32 F¨
orster et al.
Table B4. Light curve based ML classifiers that include only transient objects. Class abbreviations are defined in Table B11
Reference Data source #classes classes
Villar et al. (2019) PS1-MDS 5 SNIa, SNIbc, SNII, SNIIn, SLSN
Muthukrishna et al. (2019) PLAsTiCC 12 TDE, CART, ILOT, PISN, kN, .Ia,
SNIa, SNIax, SNIa-91bg, SNIbc, SNII
M¨oller & de Boissi`ere (2019) SNANA 2 SNIa, other
Brunel et al. (2019) SNANA, SPCC 2 SNIa, other
Revsbech et al. (2018) SPCC 3 SNIa, SNII, SNIbc
Charnock & Moss (2017) SPCC 3 SNIa, SNII, SNIbc
Lochner et al. (2016) SPCC 3 SNIa, SNII, SNIbc
Karpenka et al. (2013) SPCC 2 SNIa, other
Table B5. Observational data sources used for ML classification.
Abbreviation Long name Reference
ZTF Zwicky Transient Facility Bellm et al. (2019)
HSC–SSP Hyper Suprime-Cam Subaru Strategic Program Aihara et al. (2018)
UCR University of California Riverside Dau et al. (2018)
Time Series Classification Archive
OSC Open Supernova Catalog Guillochon et al. (2017)
ASAS-SN All-Sky Automated Survey for Supernovae Kochanek et al. (2017)
CSDR2 The Catalina Surveys Data Release 2 Drake et al. (2017)
HiTS High cadence Transient Survey F¨orster et al. (2016)
PS1-MDS Pan-STARRS-1 Medium Deep Survey Huber et al. (2011)
LINEAR Lincoln Near-Earth Asteroid Research Survey Sesar et al. (2011)
UBVI UBVI photometry of six open cluster candidates Piatti et al. (2011)
VVV Vista Variables in the Via Lactea Minniti et al. (2010)
OGLE The Optical Gravitational Lensing Experiment Udalski et al. (2008)
2MASS The Two Micron All Sky Survey Skrutskie et al. (2006)
SAGE Spitzer Survey of the Large Magellanic Cloud: Meixner et al. (2006)
Surveying the Agents of a Galaxy’s Evolution
CoRoT Convection, Rotation, and planetary Transits Baglin et al. (2006)
SDSS The Sloan Digital Sky Survey York et al. (2000)
MACHO Massive Compact Halo Objects survey Alcock et al. (2000)
EROS Exp´erience pour la Recherche d’Objets Sombres Palanque-Delabrouille et al. (1998)
ASAS All Sky Automated Survey Pojmanski (1997)
Table B6. Synthetic data sources used for ML classification.
Abbreviation Long name / Description Reference
PLAsTiCC Photometric LSST Astronomical Kessler et al. (2019)
Time-Series Classification Challenge
SNANA SuperNova ANAlysis software Kessler et al. (2009)
SPCC Supernova Photometric Classification Challenge Kessler et al. (2010)
Type II SNe confined wind acceleration model Moriya et al. (2019)
Type Ia SNe spectral templates Hsiao et al. (2007)

The ALeRCE broker 33
Table B7. Pulsating variable star classes (excluding red giants and supergiants) found in the ML literature (see text for further
details).
Type Class abbrev. Brief description
Lower MS
DSCT δScutis. Low-order p-mode pulsators. Both radial and non-radial modes can be present.
Periods typically shorter than 0.42 d. Pop. I.
LBoo λB¨ootis. A–type MS dwarf with low metallicities. Part of the DSCT class.
SXPh SX Phoenicis. Pop. II counterparts of the DSCT. Typically found in globular clusters
and dSph galaxies. Includes pulsating blue straggler stars.
roAp Rapidly oscilating Ap stars. High-order, non-radial p-mode pulsators. Amplitudes typi-
cally do not exceed 0.012 mag in V.
GDor γDoradus. High-order, non-radial g-mode pulsators. Periods between 0.3 and 3 d,
amplitudes less than 0.1 mag in V.
Upper MS
BCep βCepheids. Non-radial p-mode pulsators. Periods between 0.1–0.6 d, amplitudes in V
between 0.01–0.32 mag.
SPB Slowly pulsating blue stars, aka 53 Per stars. Non-radial g-mode pulsators. Periods
between 0.4–6 d, amplitudes in Vless than 0.03 mag.
RR Lyrae
RRL(ab,c,d,Ad,e,GB) RR Lyrae. Pulsating horizontal-branch stars, with periods of order 0.5 d. Subtypes: ab
(fundamental-mode), c (first overtone), d (double-mode), Ad (anomalous double mode),
e (second overtone). Also classified by location (Galactic bulge, GB).
Blazhko RRL with long-period modulations (Blazhko effect).
Cepheids
Ceph(CL,F,O1,
DM,MM,other)
δCepheids, aka classical (CL) Cepheids or type I Cepheids. Pulsating G-K giant and
supergiant stars. Often found pulsating in the fundamental (F), first (OI), or second
overtone; double (DM) or multi-mode (MM) pulsation also common.
ACEP Anomalous Cepheids, aka BL Boo stars. Evolved counterparts of the SX Phe stars.
Commonly found in dSph galaxies.
CephII Type II Cepheids. Low-mass Pop. II stars, often subdivided into BL Her, W Vir, and
RV Tau subclasses with increasing periods.
RVTau Type II Cepheids with periods in excess of 30 d. Light curves are well-behaved and show
double minima at the short-period end, but become increasingly irregular with increasing
period.
Subdwarf
sdBV Pulsating subdwarf B stars, aka V361 Hya, EC 14026, sdBVp, or sdBVrstars. p-mode
pulsators in which both radial and non-radial modes can be present. Periods between 60
and 570 s, amplitudes in Vless than 65 mmag.
Compact
GW Vir Pulsating pre-WD stars, aka pulsating PG 1159 stars. Includes both pulsating O-type
WD stars (DOVs) and so-called planetary nebulae nucleus variables (PNNVs).
DAV Pulsating A-type WD star, aka ZZ Ceti variables. Non-radial g-mode pulsators with
H-dominated atmospheres.
DBV Pulsating B-type WD stars, aka V777 Her stars. Non-radial g-mode pulsators with He-
dominated atmospheres.
Bellm, E. C., Kulkarni, S. R., Graham, M. J., et al. 2019,
PASP, 131, 018002, doi: 10.1088/1538-3873/aaecbe
Benavente, P., Protopapas, P., & Pichara, K. 2017, ApJ,
845, 147, doi: 10.3847/1538-4357/aa7f2d
Bennett, D. P., & Rhie, S. H. 1996, ApJ, 472, 660,
doi: 10.1086/178096
Bensby, T., Feltzing, S., Yee, J. C., et al. 2020, A&A, 634,
A130, doi: 10.1051/0004-6361/201937401
Bloom, J. S., Richards, J. W., Nugent, P. E., et al. 2012,
PASP, 124, 1175, doi: 10.1086/668468
Boch, T., & Desroziers, J. 2020, in Astronomical Society of
the Pacific Conference Series, Vol. 522, Astronomical
Data Analysis Software and Systems XXVII, ed.
P. Ballester, J. Ibsen, M. Solar, & K. Shortridge, 117
Bonnarel, F., Fernique, P., Bienaym´e, O., et al. 2000,
A&AS, 143, 33, doi: 10.1051/aas:2000331
Boone, K. 2019, AJ, 158, 257,
doi: 10.3847/1538-3881/ab5182
Borucki, W. J., Koch, D., Basri, G., et al. 2010, Science,
327, 977, doi: 10.1126/science.1185402

34 F¨
orster et al.
Table B8. As in Table B7, but for pulsating red giants and supergiants.
Type Class abbrev. Brief description
Red Giants
LPV Long Period Variable. Pulsating cool giant or supergiant stars. Often subdivided into
Miras, SRs, Irregulars, and OSARGs.
Mira Mira variables. LPV red giants with very red colors and large amplitudes (by definition,
exceeding 2.5 mag in V). Can be C- or O-rich, depending on evolutionary history.
SR Semi-regular variables. Similar to the Miras, but with smaller amplitudes (by defini-
tion, not exceeding 2.5 mag in V). Often subdivided into SRa (persistent periodicity),
SRb (poorly defined periodicity), SRc (red supergiant SRs), and SRd (orange/yellow
supergiant SRs).
OSARG OGLE Small Amplitude Red Giant. Less evolved/luminous counterpart of the Miras and
SRs, with smaller amplitudes and frequently multiple pulsation modes present.
LPVW(A,B,C,D) LPVs classified according to the sequence that they follow in a so-called Wood diagram
(Wood et al. 1999).
LPV(MAGB[C,O]) C- or O–rich Mira-type LPVs on the asymptotic giant branch (AGB)
LPV(OSARGAGB) OSARG-type LPVs on the AGB
LPV(OSARGRGB[O]) Normal or O–rich OSARG-type LPVs on the red giant branch
LPV(SRAGB[C,O]) C- or O–rich SR-type LPVs on the AGB
Supergiants
RSG Red supergiant stars with irregular or semi-regular light curves (Lc and SRc, respectively,
as per the GCVS). According to (Chatys et al. 2019), periodicities may include two
groups, related to pulsations (P∼300 −1000 d) and LSPs (P∼1000 −8000 d).
LSP LPV red giants with long secondary periods.
PVSG Periodic variable supergiant star.
Boulade, O., Charlot, X., Abbon, P., et al. 2003, in Society
of Photo-Optical Instrumentation Engineers (SPIE)
Conference Series, Vol. 4841, Proc. SPIE, ed. M. Iye &
A. F. M. Moorwood, 72–81, doi: 10.1117/12.459890
Brahem, M., Zeitouni, K., & L., Y. 2020, ApJ
Brunel, A., Pasquet, J., Pasquet, J., et al. 2019, arXiv
e-prints, arXiv:1901.00461.
https://arxiv.org/abs/1901.00461
Buchler, J. R., & Koll´ath, Z. 2002, ApJ, 573, 324,
doi: 10.1086/340550
Cabrera-Vives, G., Reyes, I., F¨orster, F., Est´evez, P. A., &
Maureira, . J. C. 2017, ApJ, 836, 97,
doi: 10.3847/1538-4357/836/1/97
Cappellarao, E. 2005, The Messenger, 120, 13
Carrasco, D., Barrientos, L. F., Pichara, K., et al. 2015, ˚ap,
584, , doi: 10.1051/0004-6361/201525752
Carrasco–Davis, R. 2020, ApJ
Carrasco-Davis, R., Cabrera-Vives, G., F¨orster, F., et al.
2019, PASP, 131, 108006, doi: 10.1088/1538-3873/aaef12
Castro, N., Protopapas, P., & Pichara, K. 2018, AJ, 155,
16, doi: 10.3847/1538-3881/aa9ab8
Catelan, M., & Smith, H. A. 2015, Pulsating Stars
(Wiley-VCH, Weinheim) (Wiley & Sons)
Chapelle, O., Scholkopf, B., & Zien, A. 2009, IEEE
Transactions on Neural Networks, 20, 542
Charnock, T., & Moss, A. 2017, ApJL, 837, ,
doi: 10.3847/2041-8213/aa603d
Chatys, F. W., Bedding, T. R., Murphy, S. J., et al. 2019,
MNRAS, 487, 4832, doi: 10.1093/mnras/stz1584
Chollet, F., et al. 2015, Keras, GitHub.
https://github.com/fchollet/keras
Clement, C. M., & Goranskij, V. P. 1999, ApJ, 513, 767,
doi: 10.1086/306885
Connelley, M. S., & Reipurth, B. 2018, ApJ, 861, 145,
doi: 10.3847/1538-4357/aaba7b
Contreras Pe˜na, C., Lucas, P. W., Minniti, D., et al. 2017,
MNRAS, 465, 3011, doi: 10.1093/mnras/stw2801
Dalton, G. B., Caldwell, M., Ward, A. K., et al. 2006, in
Society of Photo-Optical Instrumentation Engineers
(SPIE) Conference Series, Vol. 6269, Proc. SPIE,
62690X, doi: 10.1117/12.670018
Dau, H. A., Keogh, E., Kamgar, K., et al. 2018, The UCR
Time Series Classification Archive
Debosscher, J., Sarro, L. M., Aerts, C., et al. 2007, A&A,
475, 1159, doi: 10.1051/0004-6361:20077638

The ALeRCE broker 35
Table B9. Stellar variability classes, other than the pulsating ones, in the ML literature (see text for further details).
Var. Type Class Brief description
Non–
variable
NV Non–variable star
Eclipsing
E(C, SD, D) Eclipsing binary, classified according to its physical status as contact (C), semi-detached
(SD), or detached (D)
BPer, BLyr, WUMa Eclipsing binary, phenomenologically classified, according to its light curve shape, into
βPer (Algol, EA), βLyr (EB), and W UMa (EW), respectively.
Rotational
ROT Rotational variable. Rotating stars with non–uniform surface (starspots).
ChemPec Chemically peculiar rotational variable star.
ELL Close binary systems with ellipsoidal components (not eclipsing).
RSCVn RS Canum Venaticorum variable. Binary systems in which the primary star is typically
a giant, characterized by semi–periodic light curves due to active chromospheres and the
presence of starspots.
Chromosph. ACT Stars presenting surface activity due to active coronae and chromospheres.
Mdwarf M–dwarf flaring star; flares are caused by magnetic field reconnection events.
[C,WL]TTau Classic (C) or weak-lined (WL) T Tauri stars. Low-mass YSOs undergoing accretion from
their surrounding disks. Depending on the Hαemission strength, they are subdivied into
C (strong emission) and WL (weak emission). Possible evolutionary link with EX Lupi
(EXor) and FU Ori (FUor) stars, according to the mass accretion rate.
YSO
HAeBe Herbig Ae/Be star. Higher-mass counterparts of the T Tauri stars. When large, irregular
dust obscuration events are present, they may also be classified as UX Ori (UXor) stars.
FUOri FU Orionis stars. Pre–MS stars undergoing abrupt mass accretion episodes.
Outburst
LBV Luminous blue variable (aka S Doradus) star. Hot, luminous stars near or above the Ed-
dington limit undergoing vigorous mass loss and outbursts, followed by quiescent states.
CV/Nova Cataclysmic variable star (including classical novae). Mass transferring binary system
in which a MS star transfers mass onto a WD via Roche Lobe overflow. In the case of
classical novae, thermonuclear explosions take place at the surface of the mass-accreting
WD, followed by a quiescent state.
Lensing ML Microlensing event. Star whose brightness is magnified due to a gravitational lensing
event.
Other
RCB R Coronae Borealis stars. F- or G-type self-eclipsing supergiant stars that undergo dra-
matic dimming events, brought about by mass loss episodes followed by dust condensa-
tion.
DPV Double periodic variable. Binary system with variability due to eclipses or ellipsoidal
modulations on timescales of order a few days, accompanied by a long cycle lasting
about 33 times the orbital period.
BeS Be stars. Non–supergiant B star rotating close to break-up speed and presenting decretion
disks, accompanied by variable Balmer emission.
LAPV Low amplitude periodic variable. Defined in Debosscher et al. (2009), include low-
amplitude Cepheids and also rotational variable stars with regular light curves.
WR Wolf–Rayet star. Evolved, massive stars that have lost their H envelopes and show
signatures of strong stellar winds.
XB X–ray binary. CV-like systems in which the accreting star is typically not a WD, but
rather a neutron star or black hole, and which thus emit their energy mostly in the form
of X rays.

36 F¨
orster et al.
Debosscher, J., Sarro, L. M., L´opez, M., et al. 2009, A&A,
506, 519, doi: 10.1051/0004-6361/200911618
D´ek´any, I., Hajdu, G., Grebel, E. K., & Catelan, M. 2019,
ApJ, 883, 58, doi: 10.3847/1538-4357/ab3b60
Dewdney, P. E., Hall, P. J., Schilizzi, R. T., & Lazio,
T. J. L. W. 2009, IEEE Proceedings, 97, 1482,
doi: 10.1109/JPROC.2009.2021005
D’Isanto, A., Cavuoti, S., Brescia, M., et al. 2016, MNRAS,
457, 3119, doi: 10.1093/mnras/stw157
Drake, A. J., Djorgovski, S. G., Mahabal, A., et al. 2009,
ApJ, 696, 870, doi: 10.1088/0004-637X/696/1/870
Drake, A. J., Graham, M. J., Djorgovski, S. G., et al. 2014,
ApJS, 213, 9, doi: 10.1088/0067-0049/213/1/9
Drake, A. J., Djorgovski, S. G., Catelan, M., et al. 2017,
MNRAS, 469, 3688, doi: 10.1093/mnras/stx1085
Drout, M. R., Chornock, R., Soderberg, A. M., et al. 2014,
ApJ, 794, 23, doi: 10.1088/0004-637X/794/1/23
du Buisson, L., Sivanandam, N., Bassett, B. A., & Smith,
M. 2015, MNRAS, 454, 2026, doi: 10.1093/mnras/stv2041
Duev, D. A., Mahabal, A., Masci, F. J., et al. 2019,
MNRAS, 489, 3582, doi: 10.1093/mnras/stz2357
Elorrieta, F., Eyheramendy, S., & Palma, W. 2019, A&A,
627, A120, doi: 10.1051/0004-6361/201935560
Est´evez, P. A., Tesmer, M., Perez, C. A., & Zurada, J. M.
2009, IEEE Transactions on Neural Networks, 20, 189
Eyer, L., & Mowlavi, N. 2008, in Journal of Physics
Conference Series, Vol. 118, Journal of Physics
Conference Series, 012010,
doi: 10.1088/1742-6596/118/1/012010
Eyheramendy, S., Elorrieta, F., & Palma, W. 2018,
MNRAS, 481, 4311, doi: 10.1093/mnras/sty2487
Fang, Q., Maeda, K., Kuncarayakti, H., Sun, F., &
Gal-Yam, A. 2019, Nature Astronomy, 3, 434,
doi: 10.1038/s41550-019-0710-6
Faraway, J., Mahabal, A., Sun, J., et al. 2016, Statistical
Analysis and Data Mining: The ASA Data Science
Journal, 9, 1, doi: 10.1002/sam.11305
Faria, E. R., Gon¸calves, I. J., de Carvalho, A. C., & Gama,
J. 2016, Artificial Intelligence Review, 45, 235
Fernique, P., Allen, M. G., Boch, T., et al. 2015, A&A, 578,
A114, doi: 10.1051/0004-6361/201526075
Flaugher, B., Diehl, H. T., Honscheid, K., et al. 2015, AJ,
150, 150, doi: 10.1088/0004-6256/150/5/150
Flesch, E. W. 2019, arXiv e-prints, arXiv:1912.05614.
https://arxiv.org/abs/1912.05614
F¨orster, F., Maureira, J. C., San Mart´ın, J., et al. 2016,
The Astrophysical Journal, Volume 832, Issue 2, article
id. 155, 22 pp. (2016)., 832, 155,
doi: 10.3847/0004-637X/832/2/155
F¨orster, F., Moriya, T. J., Maureira, J. C., et al. 2018,
Nature Astronomy, 2, 808,
doi: 10.1038/s41550-018-0563-4
Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al.
2018, A&A, 616, A1, doi: 10.1051/0004-6361/201833051
Gaia Collaboration, Eyer, L., Rimoldini, L., et al. 2019a,
A&A, 623, A110, doi: 10.1051/0004-6361/201833304
—. 2019b, A&A, 623, A110,
doi: 10.1051/0004-6361/201833304
Gal-Yam, A., Arcavi, I., Ofek, E. O., et al. 2014, Nature,
509, 471, doi: 10.1038/nature13304
Gould, A., Dong, S., Gaudi, B. S., et al. 2010, ApJ, 720,
1073, doi: 10.1088/0004-637X/720/2/1073
Graham, M. J., Ross, N. P., Stern, D., et al. 2019, arXiv
e-prints, arXiv:1905.02262.
https://arxiv.org/abs/1905.02262
Green, A. M. 2016, PhRvD, 94, 063530,
doi: 10.1103/PhysRevD.94.063530
Greene, J. E., Strader, J., & Ho, L. C. 2019, arXiv e-prints,
arXiv:1911.09678. https://arxiv.org/abs/1911.09678
Guillochon, J., Parrent, J., Kelley, L. Z., & Margutti, R.
2017, ApJ, 835, 64, doi: 10.3847/1538-4357/835/1/64
Hosenie, Z., Lyon, R. J., Stappers, B. W., & Mootoovaloo,
A. 2019, MNRAS, 488, 4858, doi: 10.1093/mnras/stz1999
Hsiao, E. Y., Conley, A., Howell, D. A., et al. 2007, ApJ,
663, 1187, doi: 10.1086/518232
Huber, M., Rest, A., Narayan, G., et al. 2011, in American
Astronomical Society Meeting Abstracts, Vol. 218,
American Astronomical Society Meeting Abstracts #218,
328.12
Huentelemu, P., Est´evez, P. A., & F¨orster, F. 2016, in
Neural Networks (IJCNN), 2016 International Joint
Conference on, IEEE, 3322–3329
Huijse, P., Estevez, P. A., Protopapas, P., Zegers, P., &
Principe, J. C. 2012, IEEE Transactions on Signal
Processing, 60, 5135, doi: 10.1109/TSP.2012.2204260
Hunter, J. D. 2007, Computing in Science Engineering, 9,
90
IceCube Collaboration, Aartsen, M. G., Ackermann, M.,
et al. 2018, Science, 361, 147,
doi: 10.1126/science.aat2890
Jacyszyn-Dobrzeniecka, A. M., Soszy´nski, I., Udalski, A.,
et al. 2020a, ApJ, 889, 25, doi: 10.3847/1538-4357/ab61f1
Jacyszyn-Dobrzeniecka, A. M., Mr´oz, P., Kruszy´nska, K.,
et al. 2020b, ApJ, 889, 26, doi: 10.3847/1538-4357/ab61f2
Jayasinghe, T., Stanek, K. Z., Kochanek, C. S., et al. 2019,
MNRAS, 486, 1907, doi: 10.1093/mnras/stz844
Johnston, K. B., Caballero-Nieves, S. M., Petit, V., Peter,
A. M., & Haber, R. 2019, MNRAS, 2752,
doi: 10.1093/mnras/stz3165

The ALeRCE broker 37
Table B10. Extragalactic BH-related variability classes, as found in the ML literature (see text for further details).
Abbreviation Description
AGN Active Galactic Nuclei. Central accreting SMBH (>105M) where the host galaxy
dominates the total light. Variability likely due to accretion-disk instabilities.
QSO Quasi Stellar Object. Central accreting SMBH which dominates over the host galaxy in
the total light. Variability likely due to accretion-disk instabilities.
Blazar Central accreting SMBH with a relativistic jet directed towards the observer. Variability
due to sychrotron and inverse-compton relativistic beaming. This category does not
distinguish between Blazars, BL Lacs, and optical violent variables (OVVs), which peak
in different wavebands.
Table B11. Transient classes, as found in the ML literature (see text for further details).
Abbreviation Description
SNIa Type Ia supernova (SN). Thermonuclear explosion of a CO white dwarf.
SNIa-91bg Underluminous SNe Ia. SN1991bg–like.
SNIax Type Iax SNe. Deflagration dominated SN Ia.
.Ia “.Ia” SNe. He shell detonation explosion.
SNIbc Type Ib or Ic SNe. Core collapse (CC) of envelope–stripped massive star.
SNII Type II SNe. CC of red supergiant star.
SNIIn Type IIn SNe. SN explosion in dense circumstellar medium.
TDE Tidal Disruption Event. Stellar disruption due to BH proximity.
CART Calcium Rich Transient.
ILOT Intermediate Luminosity Optical Transient.
PISN Pair instability SNe. CC and thermonuclear explosion due to e−/e+pair production.
SLSN Super Luminous SNe. Class of explosions about 10 times brighter than standard SNe.
kN Kilonova. Neutron star merger optical counterpart.

38 F¨
orster et al.
Kaiser, N., Aussel, H., Burke, B. E., et al. 2002, in Society
of Photo-Optical Instrumentation Engineers (SPIE)
Conference Series, Vol. 4836, Proc. SPIE, ed. J. A. Tyson
& S. Wolff, 154–164, doi: 10.1117/12.457365
Karpenka, N. V., Feroz, F., & Hobson, M. P. 2013,
MNRAS, 429, 1278, doi: 10.1093/mnras/sts412
Kasliwal, M. M., Kulkarni, S. R., Gal-Yam, A., et al. 2012,
ApJ, 755, 161, doi: 10.1088/0004-637X/755/2/161
Keller, S. C., Schmidt, B. P., Bessell, M. S., et al. 2007,
PASA, 24, 1, doi: 10.1071/AS07001
Kessler, R., Conley, A., Jha, S., & Kuhlmann, S. 2010,
arXiv e-prints, arXiv:1001.5210.
https://arxiv.org/abs/1001.5210
Kessler, R., Bernstein, J. P., Cinabro, D., et al. 2009,
PASP, 121, 1028, doi: 10.1086/605984
Kessler, R., Narayan, G., Avelino, A., et al. 2019, PASP,
131, 094501, doi: 10.1088/1538-3873/ab26f1
Kholopov, P. N., Samus, N. N., Frolov, M. S., et al. 1998,
Combined General Catalogue of Variable Stars, 0
Kim, D. W., & Bailer-Jones, C. A. L. 2016, ˚ap, 587, ,
doi: 10.1051/0004-6361/201527188
Kim, D.-W., Protopapas, P., Bailer-Jones, C. A., et al.
2014, Astronomy & Astrophysics, 566, A43
Kim, S.-L., Lee, C.-U., Park, B.-G., et al. 2016, Journal of
Korean Astronomical Society, 49, 37,
doi: 10.5303/JKAS.2016.49.1.037
Kimura, A., Takahashi, I., Tanaka, M., et al. 2017, arXiv
e-prints, arXiv:1711.11526.
https://arxiv.org/abs/1711.11526
Kluyver, T., Ragan-Kelley, B., P´erez, F., et al. 2016, in
Positioning and Power in Academic Publishing: Players,
Agents and Agendas, ed. F. Loizides & B. Schmidt, IOS
Press, 87 – 90
Koch, D. G., Borucki, W. J., Basri, G., et al. 2010, ApJL,
713, L79, doi: 10.1088/2041-8205/713/2/L79
Kochanek, C. S., Shappee, B. J., Stanek, K. Z., et al. 2017,
PASP, 129, 104502, doi: 10.1088/1538-3873/aa80d9
Komossa, S. 2015, Journal of High Energy Astrophysics, 7,
148, doi: 10.1016/j.jheap.2015.04.006
K¨ugler, S. D., Gianniotis, N., & Polsterer, K. L. 2015,
MNRAS, 451, 3385, doi: 10.1093/mnras/stv1181
LaMassa, S. M., Cales, S., Moran, E. C., et al. 2015, ApJ,
800, 144, doi: 10.1088/0004-637X/800/2/144
Laureijs, R., Amiaux, J., Arduini, S., et al. 2011, arXiv
e-prints, arXiv:1110.3193.
https://arxiv.org/abs/1110.3193
Law, N. M., Kulkarni, S. R., Dekany, R. G., et al. 2009,
PASP, 121, 1395, doi: 10.1086/648598
Law, N. M., Fors, O., Ratzloff, J., et al. 2015, PASP, 127,
234, doi: 10.1086/680521
Lawrence, A. 2016, Astronomical Society of the Pacific
Conference Series, Vol. 505, Clues to the Structure of
AGN Through Massive Variability Surveys
(Astronomical Society of the Pacific), 107
Lemaˆıtre, G., Nogueira, F., & Aridas, C. K. 2017, Journal
of Machine Learning Research, 18, 1.
http://jmlr.org/papers/v18/16-365.html
Liu, W., Principe, J. C., & Haykin, S. 2011, Kernel
adaptive filtering: a comprehensive introduction, Vol. 57
(John Wiley & Sons)
Lochner, M., McEwen, J. D., Peiris, H. V., Lahav, O., &
Winter, M. K. 2016, ApJS, 225, 31,
doi: 10.3847/0067-0049/225/2/31
LSST Science Collaboration, Abell, P. A., Allison, J., et al.
2009, arXiv e-prints, arXiv:0912.0201.
https://arxiv.org/abs/0912.0201
Mackenzie, C., Pichara, K., & Protopapas, P. 2016, ApJ,
820, 138, doi: 10.3847/0004-637X/820/2/138
Mahabal, A., Sheth, K., Gieseke, F., et al. 2017, ArXiv
e-prints. https://arxiv.org/abs/1709.06257
Mart´ınez-Palomera, J., Lira, P., Bhalla-Ladd, I., F¨orster,
F., & Plotkin, R. M. 2020, ApJ, 889, 113,
doi: 10.3847/1538-4357/ab5f5b
Mart´ınez-Palomera, J., F¨orster, F., Protopapas, P., et al.
2018, AJ, 156, 186, doi: 10.3847/1538-3881/aadfd8
Masci, F. J., Laher, R. R., Rusholme, B., et al. 2019,
PASP, 131, 018003, doi: 10.1088/1538-3873/aae8ac
Massaro, E., Maselli, A., Leto, C., et al. 2015, Ap&SS, 357,
75, doi: 10.1007/s10509-015-2254-2
McKinney, W., et al. 2010, in Proceedings of the 9th
Python in Science Conference, Vol. 445, Austin, TX,
51–56
McLeod, B., Geary, J., Conroy, M., et al. 2015, PASP, 127,
366, doi: 10.1086/680687
Meixner, M., Gordon, K. D., Indebetouw, R., et al. 2006,
AJ, 132, 2268, doi: 10.1086/508185
Merloni, A., Predehl, P., Becker, W., et al. 2012, arXiv
e-prints, arXiv:1209.3114.
https://arxiv.org/abs/1209.3114
Mezcua, M. 2017, International Journal of Modern Physics
D, 26, 1730021, doi: 10.1142/S021827181730021X
Minniti, D., Lucas, P. W., Emerson, J. P., et al. 2010,
NewA, 15, 433, doi: 10.1016/j.newast.2009.12.002
M¨oller, A., & de Boissi`ere, T. 2019, MNRAS, 2989,
doi: 10.1093/mnras/stz3312
Moriya, T. J., Liu, D., Wang, B., & Liu, Z.-W. 2019,
MNRAS, 488, 3949, doi: 10.1093/mnras/stz1908
Morokuma, T., Tominaga, N., Tanaka, M., et al. 2014,
PASJ, 66, 114, doi: 10.1093/pasj/psu105

The ALeRCE broker 39
Mowlavi, N., Lecoeur-Ta¨ıbi, I., Lebzelter, T., et al. 2018,
A&A, 618, A58, doi: 10.1051/0004-6361/201833366
Muthukrishna, D., Narayan, G., Mandel, K. S., Biswas, R.,
& Hloˇzek, R. 2019, PASP, 131, 118002,
doi: 10.1088/1538-3873/ab1609
Narayan, G., Zaidi, T., Soraisam, M. D., et al. 2018, ApJS,
236, 9, doi: 10.3847/1538-4365/aab781
Naul, B., Bloom, J. S., P´erez, F., & van der Walt, S. 2018,
Nature Astronomy, 2, 151,
doi: 10.1038/s41550-017-0321-z
Nordin, J., Brinnel, V., van Santen, J., et al. 2019, A&A,
631, A147, doi: 10.1051/0004-6361/201935634
Nun, I., Protopapas, P., Sim, B., & Chen, W. 2016, AJ,
152, 71, doi: 10.3847/0004-6256/152/3/71
Nun, I., Protopapas, P., Sim, B., et al. 2017, FATS: Feature
Analysis for Time Series. http://ascl.net/1711.017
Oh, K., Yi, S. K., Schawinski, K., et al. 2015, ApJS, 219, 1,
doi: 10.1088/0067-0049/219/1/1
Palanque-Delabrouille, N., Afonso, C., Albert, J. N., et al.
1998, ˚ap, 332, 1
Palaversa, L., Ivezi´c, ˇ
Z., Eyer, L., et al. 2013, AJ, 146, 101,
doi: 10.1088/0004-6256/146/4/101
Patterson, M. T., Bellm, E. C., Rusholme, B., et al. 2019,
PASP, 131, 018001, doi: 10.1088/1538-3873/aae904
Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011,
Journal of machine learning research, 12, 2825
Peterson, B. M., Ferrarese, L., Gilbert, K. M., et al. 2004,
ApJ, 613, 682, doi: 10.1086/423269
Piatti, A. E., Clari´a, J. J., & Ahumada, A. V. 2011, NewA,
16, 161, doi: 10.1016/j.newast.2010.08.009
Pichara, K., & Protopapas, P. 2013, ApJ, 777, 83,
doi: 10.1088/0004-637X/777/2/83
Pichara, K., Protopapas, P., & Le´on, D. 2016, ApJ, 819, 18,
doi: 10.3847/0004-637X/819/1/18
Pietrzy´nski, G., Thompson, I. B., Gieren, W., et al. 2010,
Nature, 468, 542, doi: 10.1038/nature09598
—. 2012, Nature, 484, 75, doi: 10.1038/nature10966
Pojmanski, G. 1997, AcA, 47, 467.
https://arxiv.org/abs/astro-ph/9712146
Revsbech, E. A., Trotta, R., & van Dyk, D. A. 2018,
MNRAS, 473, 3969, doi: 10.1093/mnras/stx2570
Reyes, E., & Est´evez, P. A. 2020, arXiv e-prints,
arXiv:2005.07779. https://arxiv.org/abs/2005.07779
Richards, J. W., Starr, D. L., Miller, A. A., et al. 2012,
ApJS, 203, 32, doi: 10.1088/0067-0049/203/2/32
Ricker, G. R., Winn, J. N., Vanderspek, R., et al. 2015,
Journal of Astronomical Telescopes, Instruments, and
Systems, 1, 014003, doi: 10.1117/1.JATIS.1.1.014003
Rimoldini, L., Holl, B., Audard, M., et al. 2019, A&A, 625,
A97, doi: 10.1051/0004-6361/201834616
Ritter, H., & Kolb, U. 2003, A&A, 404, 301,
doi: 10.1051/0004-6361:20030330
Rocklin, M. 2015, in Proceedings of the 14th Python in
Science Conference, ed. K. Huff & J. Bergstra, 130 – 136
Ross, N. P., Ford, K. E. S., Graham, M., et al. 2018,
MNRAS, 480, 4468, doi: 10.1093/mnras/sty2002
S´anchez–S´aez, P. 2020, ApJ
S´anchez-S´aez, P., Lira, P., Mej´ıa-Restrepo, J., et al. 2018,
ApJ, 864, 87, doi: 10.3847/1538-4357/aad7f9
Sch¨olkopf, B., Platt, J., Shawe–Taylor, J., Smola, A., &
Williamson, R. 2001, Neural Computation, 13, 1443
Sesar, B., Stuart, J. S., Ivezi´c, ˇ
Z., et al. 2011, AJ, 142, 190,
doi: 10.1088/0004-6256/142/6/190
Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ,
131, 1163, doi: 10.1086/498708
Smith, K. W., Williams, R. D., Young, D. R., et al. 2019,
Research Notes of the American Astronomical Society, 3,
26, doi: 10.3847/2515-5172/ab020f
Soszy´nski, I., Dziembowski, W. A., Udalski, A., et al. 2014,
AcA, 64, 1. https://arxiv.org/abs/1403.6476
Soumagnac, M. T., & Ofek, E. O. 2018, PASP, 130, 075002,
doi: 10.1088/1538-3873/aac410
Spergel, D., Gehrels, N., Baltay, C., et al. 2015, arXiv
e-prints, arXiv:1503.03757.
https://arxiv.org/abs/1503.03757
Steer, I., Madore, B. F., Mazzarella, J. M., et al. 2017, AJ,
153, 37, doi: 10.3847/1538-3881/153/1/37
Street, R. A., Bowman, M., Saunders, E. S., & Boroson, T.
2018, in Society of Photo-Optical Instrumentation
Engineers (SPIE) Conference Series, Vol. 10707,
Proc. SPIE, 1070711, doi: 10.1117/12.2312293
The Dark Energy Survey Collaboration. 2005, arXiv
e-prints, astro. https://arxiv.org/abs/astro-ph/0510346
Tonry, J. L., Denneau, L., Heinze, A. N., et al. 2018, PASP,
130, 064505, doi: 10.1088/1538-3873/aabadf
Tsang, B. T. H., & Schultz, W. C. 2019, ApJL, 877, L14,
doi: 10.3847/2041-8213/ab212c
Tyson, J. A. 2019, Journal of Physics: Conference Series,
1290, 012001, doi: 10.1088/1742-6596/1290/1/012001
Udalski, A., Szymanski, M., Soszynski, I., & Poleski, R.
2008, Acta Astronomica, 58, 69
Udalski, A., Szyma´nski, M. K., & Szyma´nski, G. 2015,
AcA, 65, 1. https://arxiv.org/abs/1504.05966
Valenzuela, L., & Pichara, K. 2018, MNRAS, 474, 3259,
doi: 10.1093/mnras/stx2913
Van Rossum, G., & Drake Jr, F. L. 1995, Python reference
manual (Centrum voor Wiskunde en Informatica
Amsterdam)
Villar, V. A., Berger, E., Miller, G., et al. 2019, ApJ, 884,
83, doi: 10.3847/1538-4357/ab418c

40 F¨
orster et al.
Vivas, A. K., Mart´ınez-V´azquez, C., & Walker, A. R. 2020,
ApJS, 247, 35, doi: 10.3847/1538-4365/ab67c0
Vivas, A. K., Zinn, R., Abad, C., et al. 2004, AJ, 127, 1158,
doi: 10.1086/380929
Watson, C. L., Henden, A. A., & Price, A. 2006, Society for
Astronomical Sciences Annual Symposium, 25, 47
Wenger, M., Ochsenbein, F., Egret, D., et al. 2000, A&AS,
143, 9, doi: 10.1051/aas:2000332
Wood, P. R., Alcock, C., Allsman, R. A., et al. 1999, in
IAU Symposium, Vol. 191, Asymptotic Giant Branch
Stars, ed. T. Le Bertre, A. Lebre, & C. Waelkens, 151
Wright, D. E., Lintott, C. J., Smartt, S. J., et al. 2017,
MNRAS, 472, 1315, doi: 10.1093/mnras/stx1812
Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al.
2010, AJ, 140, 1868, doi: 10.1088/0004-6256/140/6/1868
Yaron, O., Perley, D. A., Gal-Yam, A., et al. 2017, Nature
Physics, 13, 510, doi: 10.1038/nphys4025
York, D. G., Adelman, J., Anderson, John E., J., et al.
2000, AJ, 120, 1579, doi: 10.1086/301513
Zaharia, M., Xin, R. S., Wendell, P., et al. 2016,
Communications of the ACM, 59, 56
Zhu, X., Lafferty, J., & Ghahramani, Z. 2003, in ICML 2003
workshop on the continuum from labeled to unlabeled
data in machine learning and data mining, Vol. 3
Zinn, J. C., Kochanek, C. S., Kozlowski, S., et al. 2017,
MNRAS, 468, 2189, doi: 10.1093/mnras/stx586