SAGEConv settings referring to the paper

chpoonag · April 5, 2024, 1:11am

Thanks for reading this.

I want to reproduce the results in the GraphSAGE paper, but got confused with the SAGEConv. ensure whether these settings are correct. (I thought i just need to change the aggregator_type?)

eg

GraphSAGE_gcn :
SAGEConv(
aggregator_type=“gcn”,
feat_drop=0.0, bias=True, norm=None, activation=None
)

GraphSAGE_pool:
SAGEConv(
aggregator_type=“pool”,
feat_drop=0.0, bias=True, norm=None, activation=None
)

Thanks again!

BarclayII · April 10, 2024, 1:05pm

For reproducing GraphSAGE, I would recommend using an existing example:

github.com

dmlc/dgl/blob/master/examples/core/graphsage/node_classification.py

"""
This script trains and tests a GraphSAGE model based on the information of 
a full graph.

This flowchart describes the main functional sequence of the provided example.
main
│
├───> Load and preprocess full dataset
│
├───> Instantiate SAGE model
│
├───> train
│     │
│     └───> Training loop
│           │
│           └───> SAGE.forward
└───> test
      │
      └───> Evaluate the model
"""

This file has been truncated. show original

system · May 10, 2024, 1:05pm

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.