Contained in this repository are the data files and R code used for the analyses in "Classifying X-ray Binaries: A Probabilistic Approach" by G. Gopalan, S. Vrtilek, and L. Bornn. Our hope is that one may use this code to both recreate the analysis in this paper and apply it to other data sets.
- Data Sets: There are four main folders of CCI data used in the paper: i) The 24 training systems are in the folder "Training". ii) The 6 validation systems from Table 1 are in "Validation". iii) The 6 burster systems from Table 2 are in "Burster_Test". iv) The 5 unclassified systems from Table 3 are in "Unclass".
Note on the formatting of input data: the first column is the system name, the second is the second is the system type, the third is the date, the fourth is the intensity, the fifth is color 1, and the sixth is color 2.
-
Main Code: The flow of the code is as follows. i) First, "observational_study_design.R" reads in the CCI files in "Training" and subsamples the data to create a training set as discussed in Section 2. The resultant file is output to "dat.RData". ii) Second, "rcpp_bbh_inv_samples.r" is run to generate the samples necessary for the MCMC algorithm we employ. The resultant output is stored in "rcpp_bbh_inverse.RData" iii) Third, "bbh_ess_sampler.r" is run to sample from the posterior distribution of model parameters and latent variables using elliptical slice sampling (Murray, Adams, and MacKay 2010). The resultant output is stored in "bbh_ess_10000reps.RData" iv) Fourth, "predictive_distribution_sampler.r" is run to generate the posterior predictive samples of compact object type of all the observations in a file specified in the first command line argument (where the third through fifth coumns contain CCI values) and the output file name is specified in the second command line argument. The resultant output is stored in "system_name.RData". v) Fifth, "compute+predictions.r" computes the probability estimates and associated standard errors that appear in the Tables.
-
Output: The ".RData" files in the "Validation_Output" folder store the posterior predictive draws of the compact object type for all observations in the system (denoted as Y_pred in the paper) and are located in the Validation_Output folder.
NOTES:
- This code was run on the Harvard Odyssey supercomputer and so it will take substantially longer on a personal computer; if possible the code should be run on a supercomputer. A supercomputer is not necessary, however.
- More instructions for running the code are found in the comments within each R file.