SBI++ is a complete methodology based on simulation-based (likelihood-free) inference that is customized for astronomical applications. Specifically, the code retains the fast inference speed of ∼1 sec for objects in the observational training set distribution, and additionally permits parameter inference outside of the trained noise and data at ~1 min per object. The package includes scripts for training and implementing SBI++ and is dependent on sbi (ascl:2306.002).
https://ui.adsabs.harvard.edu/abs/2023ApJ...952L..10W ; if using the code as is, please also cite https://ui.adsabs.harvard.edu/abs/2020JOSS....5.2505T and https://ui.adsabs.harvard.edu/abs/2019arXiv190507488G