Open Access
Table B.1.
Optimised hyper-parameters for our fiducial five-feature GNN model.
| Hyper-parameter | Range/Values considered | Best-fit |
|---|---|---|
| Learning rate | 10−6 − 10−2 | 0.00099 . |
| Weight decay | 10−3 − 10−1 | 0.0029 |
| Learning rate factor | [0.1, 0.3, 0.5, 0.7, 0.9] | 0.5 |
| Embedding size | [64, 100, 128] | 128 |
| Dropout probability (p) | 0 | Fixed |
| Batch size | [32, 64, 128] | 128 |
| Optimiser | Adam | Fixed. |
| Scheduler | ReduceLROnPlateau | Fixed |
| GNN layers | 1-4 | 3 |
| GNN layer type | [GCNConv, GATConv, SAGEConv, TransformerConv ] | TransformerConv |
| Pooling Type | [Mean, Add, Max, Attention] | Max |
| Linear Layers | 3 | Fixed |
| Epochs | 500 | Fixed |
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.