Table 1.
Test set ℒ1 for a LightGBM model that uses only the features which are checked.
| CNN 1-3 | CNN 4 | Luminosity | Distance | Test ℒ1 |
|---|---|---|---|---|
| ✓ | ✓ | ✓ | ✓ | 2.29 |
| ✓ | ✓ | ✓ | 2.37 | |
| ✓ | ✓ | ✓ | 2.41 | |
| ✓ | ✓ | ✓ | 2.32 | |
| ✓ | ✓ | ✓ | 2.67 | |
| ✓ | ✓ | 3.38 |
Notes. CNN 1-3 refers to the first three CNNs (from the left) in Fig 3, all of which are only trained on the Galaxy Zoo probabilities. CNN 4 is the ResNet50 which was retrained on M/L. When leaving out the distance feature (redshift and pixelsize), we also replace all luminosity features by the corresponding flux features. The baseline ℒ1 (i.e. minimizing ℒ1 when predicting a single value) is 6.52.
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.