Table 8
ML SGL searches in KiDS (details in Sec. 2).
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 |
|---|---|---|---|---|---|---|---|---|---|---|
| Lens Search | band(s) | Preselection | p | Sample | MLcand. | % MLcand. | TP (HQ) | FP | % FP | %P |
| Petrillo+2017 | r | LRG | 0.8 | 21 789 | 761 | 3.49 | 56 | 705 | 92.64 | 7.36 |
| Petrillo+2018 | r i g | LRG | 0.8 | 21 789 | 390 | 1.79 | 60 | 330 | 84.61 | 15.38 |
| Petrillo+2019 | r i g | LRG | 0.8 | 88 327 | 1689 | 1.91 | 89 | 1600 | 94.73 | 5.27 |
| Li+2020 | r | BG | 0.98 | 3 808 963 | 3552 | 0.09 | 133 | 3419 | 96.26 | 3.74 |
| Li+2020 | r i g | LRG | 0.75 | 126884 | 2848 | 2.24 | 153 | 2695 | 94.63 | 5.37 |
| Li+2021 | r i g | BG | 0.9 | 1 432 348 | 3740 | 0.26 | 295 | 3548 | 94.87 | 7.89 |
| Li+2021 | r i g | LRG | 0.8 | 72 010 | 1299 | 1.80 | 192 | 1004 | 77.29 | 14.79 |
| This work | r i g u | z < 0.8 | 0.8 | 5 538 525 | 51 638 | 0.93 | 231 (56) | 51 396 | 99.55 | 0.45 |
| This work | r i g u | LRG | 0.8 | 32709 | 13 560 | 41.46 | 62 (17) | 13498 | 99.88 | 0.46 |
| This work | r i g u | BG | 0.8 | 865 022 | 40 738 | 4.71 | 192 (52) | 40 546 | 99.88 | 0.47 |
| This work | r i g u | BG | 0.90 | 865 022 | 32858 | 3.80 | 158 (47) | 32 700 | 99.86 | 0.48 |
| This work | r i g u | BG | 0.98 | 865 022 | 19 985 | 2.31 | 122 (38) | 19 863 | 99.82 | 0.62 |
Notes. Column 2 specifies the bands given to the ML model, column 3 describes the type of sample of galaxies (BG = bright galaxies, LRG = luminous red galaxies, z < 0.8 is the redshift cut at 0.8) to create the column 5 given to the ML model. Column 4 is the prediction probability threshold used to divide the lens candidates class from the non-lens class. Column 6 is the number of SGL candidates identified by each model, with column 7 the percentage of candidates in relation to the sample size. TP (Col. 8) is the number of confirmed candidates within parenthesis the number of high-confidence ones, and FP (Col. 9) is the number of false positives with % FP (Col. 10) the percentage of false positives in the ML candidates sample. Finally, Col. 11 displays the precision (% P, Eq. (6)) achieved by each approach. Note that all data for this work is based on the 221 deg2 search described in Sec. 6.
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.