package convolutions
Type Members
- class SamplingConvolution extends Serializable with Logging
A convolution operation on graph data that aggregates features from sampled neighbors using min-hash sampling with a seed.
A convolution operation on graph data that aggregates features from sampled neighbors using min-hash sampling with a seed.
For each vertex in the input GraphFrame, this class samples up to a maximum number of neighbors using a min-hash approach based on hashing the destination vertex IDs with a provided seed. The feature embeddings of these sampled neighbors are averaged, resulting in an aggregated neighbor embedding. Optionally, this aggregated embedding can be concatenated with the original vertex features to produce an updated feature vector.
The graph is expected to have a standard structure with ID columns in vertices and edges (via GraphFrame conventions).