Platform For AI:Modularity

Use cases

• Evaluate clustering quality: After running a community detection algorithm (such as Louvain or K-means on graphs), use the Modularity component to score how well the detected communities are separated.

• Compare community detection results: Run Modularity on multiple partitioning outcomes to identify which produces the most meaningful community structure.

• Detect meaningful groups: Determine whether a graph (such as a social network or transaction graph) contains significant community structure before applying further analysis.

Prerequisites

Before you begin, ensure that you have:

• An active PAI workspace with access to Machine Learning Designer

• An input edge table with source vertex, source group, target vertex, and target group columns

Configure the component

PAI -name Modularity
    -project algo_public
    -DinputEdgeTableName=Modularity_func_test_edge
    -DfromVertexCol=flow_out_id
    -DfromGroupCol=group_out_id
    -DtoVertexCol=flow_in_id
    -DtoGroupCol=group_in_id
    -DoutputTableName=Modularity_func_test_result;

Example

This example creates a graph with two well-separated communities and computes its modularity score.

In the SQL Script component's right-side pane, clear the Use Script Mode and Whether the system adds a create table statement checkboxes before running these steps.

1. Add a SQL Script component and paste the following SQL into the editor to create the input edge table. The test graph has eight nodes split into two communities: nodes 1–4 (group 3) and nodes 5–8 (group 7). Edges connect nodes within each group densely, with only one cross-community edge between nodes 4 and 6.

   Corresponding graph data structure:

   drop table if exists Modularity_func_test_edge;
   create table Modularity_func_test_edge as
   select * from
   (
       select '1' as flow_out_id,'3' as group_out_id,'2' as flow_in_id,'3' as group_in_id
       union all
       select '1' as flow_out_id,'3' as group_out_id,'3' as flow_in_id,'3' as group_in_id
       union all
       select '1' as flow_out_id,'3' as group_out_id,'4' as flow_in_id,'3' as group_in_id
       union all
       select '2' as flow_out_id,'3' as group_out_id,'3' as flow_in_id,'3' as group_in_id
       union all
       select '2' as flow_out_id,'3' as group_out_id,'4' as flow_in_id,'3' as group_in_id
       union all
       select '3' as flow_out_id,'3' as group_out_id,'4' as flow_in_id,'3' as group_in_id
       union all
       select '4' as flow_out_id,'3' as group_out_id,'6' as flow_in_id,'7' as group_in_id
       union all
       select '5' as flow_out_id,'7' as group_out_id,'6' as flow_in_id,'7' as group_in_id
       union all
       select '5' as flow_out_id,'7' as group_out_id,'7' as flow_in_id,'7' as group_in_id
       union all
       select '5' as flow_out_id,'7' as group_out_id,'8' as flow_in_id,'7' as group_in_id
       union all
       select '6' as flow_out_id,'7' as group_out_id,'7' as flow_in_id,'7' as group_in_id
       union all
       select '6' as flow_out_id,'7' as group_out_id,'8' as flow_in_id,'7' as group_in_id
       union all
       select '7' as flow_out_id,'7' as group_out_id,'8' as flow_in_id,'7' as group_in_id
   )tmp
   ;

2. Add another SQL Script component, paste the following commands into the editor, then connect it to the component from the previous step.

   drop table if exists ${o1};
   PAI -name Modularity
       -project algo_public
       -DinputEdgeTableName=Modularity_func_test_edge
       -DfromVertexCol=flow_out_id
       -DfromGroupCol=group_out_id
       -DtoVertexCol=flow_in_id
       -DtoGroupCol=group_in_id
       -DoutputTableName=${o1};

3. Run the pipeline. After it completes, right-click the SQL Script component from the previous step and choose View Data > SQL Script Output to view the result.

   | val                 |
   | ------------------- |
   | 0.42307692766189575 |

   The val column is the modularity score (Q value). A value of approximately 0.42 confirms a strong community structure in this graph, consistent with the two visually distinct groups in the test data.

What's next

• To detect communities automatically before scoring them, combine the Modularity component with a community detection component such as Label Propagation or Louvain.

• To learn how to build and run pipelines in Machine Learning Designer, see Use Machine Learning Designer to build a pipeline.