Edge weights

Layout clustering may benefit from being influenced by edge attribute values. For visual examples, see the edge weight demo notebook. By default, each edge contributes the same weight, resulting in nodes with multiple edges between them getting pulled closer to one another. However, other common scenarios merit a different approach.

  • An edge contains naturally weighted data attributes, such as a user preference score: nodes connected by higher user preferences should cluster together more strongly
  • An edge represents multiple edges, such as if they're bundled together, and thus have an attribute such as count: higher-count edges should be weighted more heavily
  • An edge is labed with algorithm scores, such as relevance: nodes relevant to one another should cluster together more than less related ones

Override default uniform edge weights with two settings:

  • Bind edge_weight to a numeric edge attribute. The column's values will be normalized between 0 and 1.
  • Set URL parameter edgeInfluence to a value between 0 and 10. Value 0 is the default: ignore edge weight when considering an edge during layout, and instead treat them all uniformly. Value 10 means only consider edge weight, and to a strong amount.
Try the edge weight tutorial notebook to see it in action.