Table 7
Performance by brightness (held-out test set).
| Model | Bright (%) | Medium (%) | Faint (%) |
|---|---|---|---|
| RF | 100.0/100.0 | 99.80/99.78 | 99.82/99.84 |
| XGBoost | 100.0/100.0 | 99.80/99.78 | 99.91/99.89 |
| HardCBM-Cal | 99.47/95.76 | 94.38/94.26 | 96.74/95.95 |
| HardCBM | 98.77/91.11 | 91.88/91.92 | 93.84/91.26 |
Notes. Values are accuracy/macro F1 (%). Weighted F1 scores are also available: HardCBM bright 98.72%, medium 91.73%, faint 93.35%. Note that the bright subset has a highly skewed class distribution (only six RRAB out of 567 sources), which depresses macro F1 relative to accuracy.
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.