Summary:
Add support for explicitly accessing graph embeddings by introducing an embed() method to GNNs.
Motivation:
For model analysis and interpretability, it's often useful to inspect the learned graph embeddings in addition to the predictions.
Being able to extract embeddings facilitates debugging, visualization (e.g., t-SNE), and downstream tasks like clustering or similarity analysis.
- Compute and log embeddings during or after training.
- Swap out or reuse embeddings across experiments.
Summary:
Add support for explicitly accessing graph embeddings by introducing an
embed()method to GNNs.Motivation:
For model analysis and interpretability, it's often useful to inspect the learned graph embeddings in addition to the predictions.
Being able to extract embeddings facilitates debugging, visualization (e.g., t-SNE), and downstream tasks like clustering or similarity analysis.