liveBook · Manning

Node masking

As explained in Graph Neural Networks in Action

A technique that hides selected node values so a model must infer them from the remaining graph information.

Node masking hides values at selected graph nodes and turns them into prediction targets. In a spatiotemporal graph, masking the final timestep leaves earlier node features and relationships available for inference. section 6.3.1

A practical masking workflow:

  • Build graph layers with intra-timestep edges, then connect corresponding nodes across adjacent timesteps. section 6.3.1
  • Reserve the final timestep as unknown data and mask its node values. section 6.3.1
  • Use information flowing from unmasked earlier nodes to estimate the masked features with a GAT model. section 6.3.1

How Graph Neural Networks in Action covers it

How deep it goes
Explains masking within a five-step graph pipeline and connects it to GAT training, stopping short of masking implementation details.
Examples
Uses final-timestep nodes in a layered graph and illustrates inference from unmasked nodes at earlier timesteps.
What it assumes
Assumes graph nodes, edges, timesteps, and GATs, while demonstrating how their relationships support masked-node prediction.
What you can do after
Apply it via PyTorch Geometric-style graph construction and train a GAT to estimate masked node features.
How it connects
Connects node masking with spatiotemporal graphs, temporal dependencies, node prediction, and graph attention networks.
Cover of Graph Neural Networks in Action
liveBookmanning.com
Read Graph Neural Networks in Action on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Graph Neural Networks in Action. The section links point to the parts of the book it draws on.