J/A+A/677/A16 Light curves neural network approximation (Demianenko+, 2023)
Understanding of the properties of neural network approaches for
transient light curve approximation.
Demianenko M., Malanchev K., Samorodova E., Sysak M., Shiriaev A.,
Derkach D., Hushchyn M.
<Astron. Astrophys. 677, A16 (2023)>
=2023A&A...677A..16D 2023A&A...677A..16D (SIMBAD/NED BibCode)
ADC_Keywords: Supernovae ; Models ; Photometry ; Optical
Keywords: methods: data analysis - supernovae: general - methods: statistical
Abstract:
Modern time-domain photometric surveys collect a lot of observations
of various astronomical objects, and the coming era of large-scale
surveys will provide even more information. Most of the objects have
never received a spectroscopic follow-up, which is especially crucial
for transients, e.g., supernovae. Flux time series are actively used
as an affordable alternative for photometric classification and
characterization, e.g., peak identification and luminosity decline
estimation. However, the collected time series are multidimensional,
irregularly sampled, may contain outliers, and do not have
well-defined systematic uncertainties. This paper presents a search
for the best-performing methods to approximate over time and
wavelength the observed light curves for generating time series with
regular time steps in each passband. We examine several light curve
approximation methods based on neural networks such as Multilayer
Perceptrons, Bayesian Neural Networks, and Normalizing Flows to
approximate observations of a single light curve. Tests based on both
the simulated PLAsTiCC and real Zwicky Transient Facility Bright
Transient Survey data samples demonstrate that even a few observations
are enough to fit networks and improve the quality of approximation
compared to state-of-the-art models. The methods described in this
work have lower computational complexity and work significantly faster
than Gaussian Processes. Additionally, we analyze the performance of
the approximation techniques in perspective of further peak
identification and transients classification. The study results are
released in an open and user-friendly Fulu Python library available on
GitHub for the scientific community.
Description:
This table presents estimated peaks for the Zwicky Transient Facility
(ZTF) Bright Transient Survey (BTS) objects, which were downloaded at
23:29 09/22/2021 (only objects with at least ten observations per each
{g} and {r} passbands). The total number of selected light curves is
1870. The peak position in approximated light curve was estimated as
the timestamp with maximum total flux value (sum of fluxes in {r} and
{g} passbands). The timestamps of the peak in the light curve
observations are measured in Modified Julian Date (MJD). The
magnitudes are calculated in ZTF BTS units. The fluxes are in mJy. In
column explanations, we use following abbreviations for the names of
approximation models: BNN for Bayesian Neural Networks, MLP (sklearn)
for Multilayer Perceptron implemented using scikit-learn, MLP
(pytorch) for Multilayer Perceptron implemented using pytorch, NF for
Normalizing Flows, GP for Gaussian Processes.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table9.dat 866 1870 Estimated peaks for the Zwicky Transient Facility
(ZTF) Bright Transient Survey (BTS) objects
--------------------------------------------------------------------------------
See also:
https://github.com/HSE-LAMBDA/fulu : HSE-LAMBDA/fulu programs
Byte-by-byte Description of file: table9.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 4 I4 --- Num [0/1869] Sequnetial number of object (num)
6- 17 A12 --- Name ZTF 9-letter identifier (ZTFAAaaaaaaa)
(obj_names)
19- 36 F18.15 mag g+rmagBNN Magnitude of peak in {g} + {r} bands,
Bayesian Neural Networks (BBM)
(magnitudessumbnn)
38- 56 F19.16 mJy Fg+rBNN Flux of peak in {g} + {r} bands,
Bayesian Neural Networks (BBM)
(fluxessumbnn)
58- 75 F18.12 d Tg+rBNN Time of peak in {g} + {r} in MJD,
Bayesian Neural Networks (BBM)
(timessumbnn)
77- 94 F18.15 mag gmagBNN Magnitude of peak in {g} band,
Bayesian Neural Networks (BBM)
(magnitudesgbnn)
96-114 F19.16 mJy FgBNN Flux of peak in {g} band,
Bayesian Neural Networks (BBM)
(fluxesgbnn)
116-133 F18.12 d TgBNN Time of peak in {g} band in MJD,
Bayesian Neural Networks (BBM)
(timesgbnn)
135-152 F18.15 mag rmagBNN Magnitude of peak in {r} band,
Bayesian Neural Networks (BBM)
(magnitudesrbnn)
154-172 F19.16 mJy FrBNN Flux of peak in {r} band,
Bayesian Neural Networks (BBM)
(fluxesrbnn)
174-191 F18.12 d TrBNN Time of peak in {r} band in MJD,
Bayesian Neural Networks (BBM)
(timesrbnn)
193-210 F18.15 mag g+rmagNNdk Magnitude of peak in {g} + {r} bands,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(magnitudessumnn_sk)
212-230 F19.16 mJy Fg+rNNdk Flux of peak in {g} + {r} bands,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(fluxessumnn_sk)
232-249 F18.12 d Tg+rNNdk Time of peak in {g} + {r} bands in MJD,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(timessumnn_sk)
251-268 F18.15 mag gmagNNdk Magnitude of peak in {g} band,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(magnitudesgnn_sk)
270-288 F19.16 mJy FgNNdk Flux of peak in {g} band, Multilayer
Perceptron implemented (MLP)
implemented using scikit-learn
(fluxesgnn_sk)
290-307 F18.12 d TgNNdk Time of peak in {g} band in MJD,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(timesgnn_sk)
309-326 F18.15 mag rmagNNdk Magnitude of peak in {r} band,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(magnitudesrnn_sk)
328-346 F19.16 mJy FrNNdk Flux of peak in {r} band,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(fluxesrnn_sk)
348-365 F18.12 d TrNNdk Time of peak in {r} band in MJD,
Multilayer Perceptron implemented (MLP)
implemented using scikit-learn
(timesrnn_sk)
367-384 F18.15 mag g+rmagNNpt Magnitude of peak in {g} + {r} bands,
Multilayer Perceptron implemented (MLP)
(magnitudessumnn_pt)
386-404 F19.16 mJy Fg+rNNpt Flux of peak in {g} + {r} bands,
Multilayer Perceptron implemented (MLP)
implemented using pytorch
(fluxessumnn_pt)
406-423 F18.12 d Tg+rNNpt Time of peak in {g} + {r} bands in MJD,
Multilayer Perceptron implemented (MLP)
implemented using pytorch (timessumnn_pt)
425-442 F18.15 mag gmagNNpt Magnitude of peak in {g} band,
Multilayer Perceptron implemented (MLP)
implemented using pytorch
(magnitudesgnn_pt)
444-462 F19.16 mJy FgNNpt Flux of peak in {g} band,
Multilayer Perceptron implemented (MLP)
implemented using pytorch (fluxesgnn_pt)
464-481 F18.12 d TgNNpt Time of peak in {g} band in MJD,
Multilayer Perceptron implemented (MLP)
implemented using pytorch, (timesgnn_pt)
483-500 F18.15 mag rmagNNpt Magnitude of peak in {r} band,
Multilayer Perceptron implemented (MLP)
implemented using pytorch
(magnitudesrnn_pt)
502-520 F19.16 mJy FrNNpt Flux of peak in {r} band,
Multilayer Perceptron implemented (MLP)
implemented using pytorch (fluxesrnn_pt)
522-539 F18.12 d TrNNpt Time of peak in {r} band in MJD,
Multilayer Perceptron implemented (MLP)
implemented using pytorch (timesrnn_pt)
541-558 F18.15 mag g+rmagNF Magnitude of peak in {g} + {r} bands,
Normalizing Flows (NF)
(magnitudessumnf)
560-571 F12.9 mJy Fg+rNF Flux of peak in {g} + {r} bands,
Normalizing Flows (NF)
(fluxessumnf)
573-590 F18.12 d Tg+rNF Time of peak in {g} + {r} bands in MJD,
Normalizing Flows (NF)
(timessumnf)
592-609 F18.15 mag gmagNF Magnitude of peak in {g} band,
Normalizing Flows (NF)
(magnitudesgnf)
611-622 F12.9 mJy FgNF Flux of peak in {g} band,
Normalizing Flows (NF) (fluxesgnf)
624-641 F18.12 d TgNF Time of peak in {g} band in MJD,
Normalizing Flows (NF)
(timesgnf)
643-660 F18.15 mag rmagNF Magnitude of peak in {r} band,
Normalizing Flows (NF)
(magnitudesrnf)
662-673 F12.9 mJy FrNF Flux of peak in {r} band,
Normalizing Flows (NF) (fluxesrnf)
675-692 F18.12 d TrNF Time of peak in {r} band in MJD,
Normalizing Flows (NF)
(timesrnf)
694-711 F18.15 mag g+rmagGP Magnitude of peak in {g} + {r} bands,
Gaussian Processes (GP)
(magnitudessumgp)
713-731 F19.16 mJy Fg+rGP Flux of peak in {g} + {r} bands,
Gaussian Processes (GP)
(fluxessumgp)
733-750 F18.12 d Tg+rGP Time of peak in {g} + {r} bands in MJD,
Gaussian Processes (GP)
(timessumgp)
752-769 F18.15 mag gmagGP Magnitude of peak in {g} band,
Gaussian Processes (GP)
(magnitudesggp)
771-789 F19.16 mJy FgGP Flux of peak in {g} band,
Gaussian Processes (GP) (fluxesggp)
791-808 F18.12 d TgGP Time of peak in {g} band in MJD,
Gaussian Processes (GP)
(timesggp)
810-827 F18.15 mag rmagGP Magnitude of peak in {r} band,
Gaussian Processes (GP)
(magnitudesrgp)
829-847 F19.16 mJy FrGP Flux of peak in {r} band,
Gaussian Processes (GP) (fluxesrgp)
849-866 F18.12 d TrGP Time of peak in {r} band in MJD,
Gaussian Processes (GP)
(timesrgp)
--------------------------------------------------------------------------------
Acknowledgements:
Mikhail Hushchyn, mhushchyn(at)hse.ru
Mariia Demianenko, demianenko(at)mpia.de
(End) Patricia Vannier [CDS] 29-Jun-2023