Table C.1
Comparison of the four different nested sampling implementations on the noise-free benchmark retrieval from section 4.1.
| Method | Live points | Likelihood calls | log Z | Ntotal | Neff | ε (in %) | Wall time (in h) |
|---|---|---|---|---|---|---|---|
| dynesty (1) | 5000 | 18 730 100 | 401.497 ± 0.080 | 206 506 | 41 755.9 | 20.2 | 1548.1 |
| MultiNest | 4000 | 17 786 920 | 407.182 ± 0.080 | 30 540 | — | — | 326.2 |
| nautilus | 4000 | 1 042 944 | 401.844 ± 0.010 | 331 008 | 10 004.4 | 3.1 | 19.5 |
| UltraNest (2) | 400 | > 34 319 279 | > 213.0 | — | — | — | > 813.5 |
Notes. (1) While the logZ estimate from dynesty is similar to the other methods (including FMPE-IS / NPE-IS), the 1D and 2D marginal posteriors deviate strongly from all other methods. (2) We aborted the UltraNest run when the number of likelihood evaluations exceeded the size of our ML training set and the sampler was still far from convergence (based on its log Z estimate).
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.