Probabilistic classification of Fermi LAT gamma-ray sources (effect of covariate shift)

Version 1:

These are data products connected to https://arxiv.org/abs/2307.09584, where an analysis of the effect of covariate shift on the probabilistic classification of the Fermi LAT gamma-ray sources from the 4FGL-DR3 catalog is performed.

The files 

4FGL-DR3_6class_GMM_nmin100_prob_cat.csv4FGL-DR3_6class_GMM_nmin100_weighted_prob_cat.csv

contain probabilistic classification into 6 classes (determined in https://arxiv.org/abs/2301.07412) with random forest and neural networks methods. The catalog in "4FGL-DR3_6class_GMM_nmin100_weighted_prob_cat.csv" is constructed including weights for associated sources used in training in order to account for the difference in the distribution of associated (training dataset) and  unassociated (target dataset) sources. The catalog in "4FGL-DR3_6class_GMM_nmin100_prob_cat.csv" is constructed with unweighted training samples.

The files

4FGL-DR3_6class_GMM_nmin100_summary.csv4FGL-DR3_6class_GMM_nmin100_weighted_summary.csv

contain the corresponding summaries of the definition of classes and predicted numbers of sources for the RF and NN algorithms for associated sources (averaged over cases when the sources are in the testing samples) and unassociated sources.

Detailed description of the construction of the catalogs can be found in https://arxiv.org/abs/2307.09584.

Version 2: update for the Fermi LAT 4FGL-DR4 catalog.

The filenames slightly change.Probabilistic catalogs with unweighted and weighted training respectively:4FGL-DR4_6classes_GMM_prob_cat.csv4FGL-DR4_6classes_GMM_weighted_prob_cat.csvThe corresponding summary files:4FGL-DR4_6classes_GMM_summary.csv4FGL-DR4_6classes_GMM_weighted_summary.csv

Version 3: catalogs corresponding to the published version of the paper. The filenames and the format are the same as in Version 2.