New version of FastRP that utilizes message passing. This results in better memory usage and removes the need of pulling all edges into a global accumulator for UDF processing.
Testing Method:
We tested the new implementation by comparing the cosine similarities between vertex embeddings produced between the current FastRP and the proposed FastRP. The new implementation here produces very similar/identical results.
New version of FastRP that utilizes message passing. This results in better memory usage and removes the need of pulling all edges into a global accumulator for UDF processing.
Testing Method: We tested the new implementation by comparing the cosine similarities between vertex embeddings produced between the current FastRP and the proposed FastRP. The new implementation here produces very similar/identical results.