3 - CLASSIFICATION: Classification probabilities

Overview

DES used 3 classifiers to select Type Ia Supernovae. Here are released the probabilities of the 1635 used SNIa for the DES-SN5YR Hubble Diagram.

Classifiers

We made use of the following classifiers:

Classifier agreement

_images/classifier_agreement.png _images/training_sample_comparison.png

Release Format

Probabilities with classifiers that use Redshift information

A csv file is released with the candidate ID (CID) as the single event identifier, in combination with the probabilities.

Column definitions

  • PROB_SNNV19 (*) - Probability of being Ia from SuperNNova trained on sims generated using core-collapse templates from Vincenzi et al. 2019 (Nominal)

  • PROB_SNNDESCC - Probability of being Ia from SuperNNova trained on sims generated using core-collapse templates from Jones et al. 2017

  • PROB_SNNJ17 - Probability of being Ia from SuperNNova trained on sims generated using core-collapse templates from DES data

  • PROB_SCONE - Probability of being Ia from SCONE trained on sims generated using core-collapse templates Vincenzi et al. 2019

  • PROB_SNIRFV19 - Probability of being Ia from SNIRF trained on sims generated using core-collapse templates from Vincenzi et al. 2019

  • PROBCC_BEAMS - BEAMS Probability of being core-collapse (see eq. 6 in Vincenzi et al. 2024)

(*) PROB_SNNV19 is our Nominal

Probabilities of being SN Ia using only light-curves (no redshift information)

We also release the classification from Möller et al. 2024 that does not use redshift information.

This is in a csv file again with the candidate ID (CID) as the single event identifier, in combination with the probabilities.

Column definitions

  • PROB_SNN_noz_singlemodel: Prob of being Ia from SuperNNova trained on sims generated using core-collapse templates from Vincenzi et al 2019. This model uses only light-curves for classification, no redshift information. (Nominal: single model)

  • PROB_SNN_noz_ensemble: Prob of being Ia from SuperNNova trained on sims generated using core-collapse templates from Vincenzi et al 2019. This model uses only light-curves for classification, no redshift information. These probabilities are an average of 5 independet models (Ensemble).