Table B.1.
Computation times for the six modeling methods, including best-fit optimization and posterior distribution sampling.
| Modeling method label | Software package | Involved steps | Longest instance | Total | Hardware configuration |
|---|---|---|---|---|---|
| Sérsic+Shapelets | LENSTRONOMY | PSO + MCMC | 17 hours | 38 hours | Intel Core i7 CPU (2 cores) |
| Adaptive+Matérn | VKL | Nested sampling | 4 days | 4.2 days | Intel Xeon Platinum 8260 CPU (96 cores) |
| Cluster+Exp | QLENS | Nested sampling | 1.5 hours | 1.5 hours | Intel Xeon Platinum 8160 CPU (192 cores) |
| Cluster+Exp+Lumweight | QLENS | Nested sampling | 2 hours | 2 hours | Intel Xeon Platinum 8160 CPU (192 cores) |
| Sparsity+Wavelets | HERCULENS | Grad. descent + SVI | 6 minutes | 3.4 hours | NVIDIA A10 GPU |
| Correlated Field | HERCULENS | Grad. descent + geoVI | 45 minutes | 4.5 hours | NVIDIA A10 GPU |
Notes. Some of the final models were obtained by combining several model instances, so we quote the runtime for a single instance (the most expensive) in addition to the total runtime (assuming single instances were not performed in parallel). For additional details and software package references, see Sect. 3 and Appendix B.
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.