Published January 9, 2024 | Version v1
Dataset Open

Probabilistic classification of Fermi LAT gamma-ray sources (4FGL-DR4, no coordinate features)

  • 1. ROR icon Friedrich-Alexander-Universität Erlangen-Nürnberg

Description

These are catalogs with probabilistic classification of the Fermi-LAT 4FGL-DR4 sources used in https://arxiv.org/abs/2401.04565

The classification is performed similarly to https://zenodo.org/doi/10.5281/zenodo.8140548
with the following differences:

  1. Coordinate features (Glat and Glon) are not used as input features.
  2. Some versions of the catalogs below do not have sources with uncertain associations in the training (bcu and spp classes in 4FGL-DR4).

Files ending with "prob_cat" contain the class probabilities and corresopnding statistical uncertainties for all 4FGL-DR4 sources together with input features and source names. sin(GLAT), cos(GLON), sin(GLON) are included for convenience but are not used in training.

Files ending with "summary" contain definition of the classes in terms of the groups of physical classes in 4FGL-DR4 and the predicted source counts for associated and unassociated sources.

The files 

4FGL-DR4_5classes_GMM_no_bcu_spp_no_coord_features_prob_cat_v1.csv
4FGL-DR4_5classes_GMM_no_bcu_spp_no_coord_features_summary_v1.csv

correspond to the main analysis in https://arxiv.org/abs/2401.04565: 5 classes and no bcu, spp, or unk sources in training.

The files 

4FGL-DR4_5classes_GMM_no_bcu_spp_no_coord_features_weighted_prob_cat_v1.csv
4FGL-DR4_5classes_GMM_no_bcu_spp_no_coord_features_weighted_summary_v1.csv

are similar to the main analysis results but include a weighting in training to account for the covariate shift, e.g., the differences in the distribution of associated and unassociated sources. More details on the weighting can be found in https://arxiv.org/abs/2307.09584. The weighted training is used as a check of systematic uncertainties in Appendix A of https://arxiv.org/abs/2401.04565.

The files 

4FGL-DR4_6classes_GMM_no_coord_features_prob_cat_v1.csv
4FGL-DR4_6classes_GMM_no_coord_features_summary_v1.csv
4FGL-DR4_6classes_GMM_no_coord_features_weighted_prob_cat_v1.csv
4FGL-DR4_6classes_GMM_no_coord_features_weighted_summary_v1.csv

contain respectively unweighted and weighted training with 6 classes, where bcu and spp sources are included in training (unk sources are not included in training). The groups are similar to the groups in https://arxiv.org/abs/2307.09584, but the coordinate features are not used in training. These catalogs are also used to estimate systematic uncertainties in Appendix A of https://arxiv.org/abs/2401.04565.

Files

4FGL-DR4_5classes_GMM_no_bcu_spp_no_coord_features_prob_cat_v1.csv

Additional details

Related works

Continues
Dataset: 10.5281/zenodo.8140548 (DOI)
Is derived from
Publication: 10.1093/rasti/rzad053 (DOI)
Publication: 10.1093/mnras/stad940 (DOI)
Is supplement to
Preprint: https://arxiv.org/abs/2401.04565 (URL)

Funding

Multi-class classification and population studies of unassociated Fermi-LAT gamma-ray sources with machine learning MA 8279/3-1
Deutsche Forschungsgemeinschaft