PLAsTiCC / README.md
chaenykim's picture
Update README.md
927a8f7 verified
metadata
task_categories:
  - feature-extraction
tags:
  - astro
size_categories:
  - 1M<n<10M

Astronomical Time-Series Dataset

This is the full dataset of astronomical time-series from the 2018 Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC) Kaggle competition. There are 18 types of astronomical sources represented, including transient phenomena (e.g. supernovae, kilonovae) and variable objects (e.g. active galactic nuclei, Mira variables).

The original Kaggle competition can be found here. This note from the competition describes the dataset in detail. Astronomers may be interested in this paper describing the simulations used to generate the data.

Dataset Structure

Data Fields

  • object_id: unique object identifier
  • times_wv: 2D array of shape (N, 2) containing the observation times (modified Julian days, MJD) and filter (wavelength) for each observation, N=number of observations\
  • target: 2D array of shape (N, 2) containing the flux (arbitrary units) and flux error for each observation\
  • label: integer representing the class of the object (see below)\
  • redshift: true redshift of the object\
  • ddf: 1 if the object was in the deep drilling fields (DDF) survey area of LSST, 0 if wide-fast-deep (WFD)\
  • hostgal_specz: spectroscopic redshift of the host galaxy\
  • hostgal_photoz: photometric redshift of the host galaxy\
  • hostgal_photoz_err: uncertainty on the photometric redshift

Data Splits

The original PLAsTiCC challenge had a training set that was biased to be lower redshift, brighter, and higher signal-to-noise than the test set. This was created to emulate a spectroscopically confirmed subset of observations that typically would be used to train a machine learning classifier. The test set represents a realistic simulation of all LSST observations -- fainter and noisier than the training set. In this dataset, the original PLAsTiCC training set was split into 90/10 training/validation and the original test set was uploaded unchanged.

  • train: 90% of the PLAsTiCC training set
  • validation: 10% of the PLAsTiCC training set
  • test: full PLAsTiCC test set

Additional Information

Class Descriptions

 6: microlens-single
 15: tidal disruption event (TDE)
 16: eclipsing binary (EB)
 42: type II supernova (SNII)
 52: peculiar type Ia supernova (SNIax)
 53: Mira variable
 62: type Ibc supernova(SNIbc)
 64: kilonova (KN)
 65: M-dwarf
 67: peculiar type Ia supernova (SNIa-91bg)
 88: active galactic nuclei (AGN)
 90: type Ia supernova (SNIa)
 92: RR-Lyrae (RRL)
 95: superluminous supernova (SLSN-I)
 991: microlens-binary
 992: intermediate luminosity optical transient (ILOT)
 993: calcium-rich transient (CaRT)
 994: pair instability supernova (PISN)
 995: microlens-string

Citation Information

@ARTICLE{2018arXiv181000001T,
        author = {{The PLAsTiCC team} and {Allam}, Tarek, Jr. and {Bahmanyar}, Anita  and {Biswas}, Rahul and {Dai}, Mi and {Galbany}, Llu{\'\i}s and {Hlo{\v{z}}ek},      Ren{\'e}e and {Ishida}, Emille E.~O. and {Jha}, Saurabh W. and {Jones}, David O.     and {Kessler}, Richard and {Lochner}, Michelle and {Mahabal}, Ashish A. and {Malz},  Alex I. and {Mandel}, Kaisey S. and {Mart{\'\i}nez-Galarza}, Juan Rafael and         {McEwen}, Jason D. and {Muthukrishna}, Daniel and {Narayan}, Gautham and {Peiris},   Hiranya and {Peters}, Christina M. and {Ponder}, Kara and {Setzer}, Christian N.     and {The LSST Dark Energy Science Collaboration} and {LSST Transients}, The and      {Variable Stars Science Collaboration}},
         title = "{The Photometric LSST Astronomical Time-series Classification Challenge (PLAsTiCC): Data set}",
       journal = {arXiv e-prints},
      keywords = {Astrophysics - Instrumentation and Methods for Astrophysics, Astrophysics - Solar and Stellar Astrophysics},
          year = 2018,
         month = sep,
           eid = {arXiv:1810.00001},
         pages = {arXiv:1810.00001},
           doi = {10.48550/arXiv.1810.00001},
 archivePrefix = {arXiv},
        eprint = {1810.00001},
  primaryClass = {astro-ph.IM},
        adsurl = {https://ui.adsabs.harvard.edu/abs/2018arXiv181000001T},
       adsnote = {Provided by the SAO/NASA Astrophysics Data System}
 }