graphprovider.md

Ofnil's GraphProvider

Workflow of Ofnil GraphProvider

Input data of Ofnil NeighborSampledDataLoader is a tuple containing two structs, TopologyFeatureViewInfo and List[TableFeatureViewInfo]]. Ofnil uses those two inputs to generate corresponding topology information and vertex/edge feature views.

After that, Ofnil calls GraphProvider to get information about the input graph. As Pytorch will use this information, we have designed a class called IterableDatasetWrapper to store data according to Pytorch format. According to the definition in PyTorch docs, all subclasses should overwrite the __iter__ in python. That's why an iterator is needed for making a GraphProvider plugin. The graph provider should provide the number of nodes, the type of nodes, edges, and the relationship between the root node and its neighbors to NeighborSampledDataLoader for sample generation.

How to create a GraphProvider plugin

Example：Build a GraphProvider for Neo4j

• Step 1: Import the corresponding database module, GraphProvider abstract method and TopologyFeatureViewInfo.

  from neo4j import GraphDatabase
  from ofnil.infra.graph.graph_provider import GraphProvider
  from ofnil.ofnil import TopologyFeatureViewInfo
  

  Step 2: Create the constructor and destructor for GraphProvider class.

  def __init__(self, topo_info, seed_type):
      super().__init__(seed_type) # The label for nodes within the graph.
  
      self.driver = GraphDatabase.driver(
          uri=,
          auth=,
      ) # The driver API from the corresponding database
  
      self.seed_query = (
  
      ) # The query to get seed_type from a graph.
  
      self.seed_session # The session within the chosen driver
      self.seed_stream  # This variable used to store the excuting result from seed_session.
  
  def __del__(self):
      self.seed_session.close()
      self.driver.close()
  

  Step 3: Create an iterator for GraphProvider class to get an iterable-style dataset.

   def __iter__(self):
      self.seed_stream = self.seed_session.run(self.seed_query)
      for record in self.seed_stream:
      yield record["seed"]
  

  Step 4: Create methods that provide needed information about the graph.

      def get_num_seeds(self):
  
      def get_num_vertices(self, vertex_type: str = None):
  
      def get_num_edges(self):