This repository contains the code used to investigate the application of positive and unlabeled (PU) techniques to identify repeater candidates in the CHIME/FRB 2021 and 2023 catalogs. It is an extension of a previously prepared paper (the code for which is here), with newer data, expanded analysis, and more extensive comparison with previous results.
- For feature extraction, we use a modified version of the code provided by Zhu-Ge et al. for their paper Machine learning classification of CHIME fast radio bursts: II. Unsupervised Methods. Their original code can be found here.
- Classic Elkanoto, Weighted Elkanoto, and Bagging PU are implemented with the
pulearnlibrary - Modified logistic regression, as described by Jasky et al. in A modified logistic regression for positive and unlabeled learning, is implemented with a custom fork of the original repository provided by the authors.
- PUExtraTrees, as described in Positive-Unlabeled Learning using Random Forests via Recursive Greedy Risk Minimization, is implemented with a custom fork of the original repository provided by the authors.
Report bugs in the code by opening a new issue in this GitHub repository.