C

LabelPropagationClustering

Show Usages
Detects the communities in the specified input graph by applying a label propagation algorithm.
Inheritance Hierarchy

Remarks

This algorithm iteratively assigns a label to each node. The label of a node is set to the most frequent label among its neighbors. If there are multiple candidates the algorithm randomly chooses one of them. After applying the algorithm, all nodes with the same label belong to the same community.

The implementation is based on the description in: "Near linear time algorithm to detect community structures in large-scale networks" by U. N. Raghavan, R. Albert, and S. Kumara in Phys. Rev. E 76, 036106, 2007

For a given node and neighbor, the weight of the neighbor's label is nodeWeight(neighbor) * edgeWeight((node, neighbor)). In addition, the overall weight of a specific label is the sum of such weights for all neighbors with matching label. The weights are specified in nodeWeights and edgeWeights, respectively. If no values are provided uniform weights of 1.0 are applied.

If no initialLabels are provided each node starts with a unique label. Otherwise, at the start of the algorithm, each node is associated with the label specified with initialLabels. Negative values are ignored, i.e., at the start there may be nodes without label. This implies that there may be results, where some of the nodes are not associated with labels, too. Such nodes are not associated with any cluster.

If edgeDirectedness is specified, the algorithm only considers a subset of the neighbors when determining the label of a node. More precisely, it only considers the neighbors connected by edges to a node n for which

  • the edge's directedness is greater than 0.0 and the edge is incoming (i.e. n is the edge's target),
  • the edge's directedness is smaller than 0.0 and the edge is outgoing (i.e. n is the edge's source),
  • or the edge's directedness is 0.0 (i.e. the edge is considered to be undirected).

Other Clustering Algorithms

yFiles for HTML supports a number of other clustering algorithms:

Examples

Calculating clusters of a graph using the label propagation method
// prepare the label propagation algorithm
const algorithm = new LabelPropagationClustering({
  edgeWeights: (edge) =>
    edge.style instanceof PolylineEdgeStyle
      ? edge.style.stroke!.thickness
      : 0,
  nodeWeights: (node) => node.layout.width,
})
// run the algorithm
const result = algorithm.run(graph)

// highlight the nodes of the clusters with different styles
for (const node of graph.nodes) {
  const componentId = result.nodeClusterIds.get(node)!
  graph.setStyle(node, clusterStyles.get(componentId)!)
}
Excluding self-loop edges
const algorithm = new LabelPropagationClustering({
  // Ignore edges without target arrow heads
  subgraphEdges: {
    excludes: (edge: IEdge): boolean =>
      edge.sourceNode === edge.targetNode,
  },
})

See Also

Developer's Guide

API

labelPropagation

Members

No filters for this type

Constructors

LabelPropagationClustering

()

Parameters

Properties

edgeDirectedness

: ItemMapping<IEdge, number>
Gets or sets a mapping that stores the directedness of the edges.
If a value is provided here, the algorithm only considers a subset of the neighbors when determining the label of a node. More precisely, it only considers the neighbors connected by edges to a node n for which
  • the edge's directedness is greater than 0.0 and the edge is incoming (i.e. n is the edge's target),
  • the edge's directedness is smaller than 0.0 and the edge is outgoing (i.e. n is the edge's source),
  • or the edge's directedness is 0.0 (i.e. the edge is considered to be undirected).
conversionfinal

edgeWeights

: ItemMapping<IEdge, number>
Gets or sets a mapping for edge weights.
For a given node and neighbor, the weight of the neighbor's label is nodeWeight(neighbor) * edgeWeight((node, neighbor)). In addition, the overall weight of a specific label is the sum of such weights for all neighbors with matching label. The weights are specified in nodeWeights and edgeWeights, respectively. If no values are provided, uniform weights of 1.0 are applied.
conversionfinal

initialLabels

: ItemMapping<INode, number>
Gets or sets a mapping that stores the initial integer labels of each node.

At the start of the algorithm, each node is associated with the label specified here. Negative values are ignored, i.e., at the start there may be nodes without label. This implies that there may be results where some of the nodes are not associated with labels, too. Such nodes are not associated with any cluster.

If nothing is set each node starts with a unique label.

The labels which are considered in this algorithm are internally assigned integer values. They have nothing to do with the marking objects which are commonly referred to as "label" in yFiles for HTML.
conversionfinal

nodeWeights

: ItemMapping<INode, number>
Gets or sets a mapping for node weights.
For a given node and neighbor, the weight of the neighbor's label is nodeWeight(neighbor) * edgeWeight((node, neighbor)). In addition, the overall weight of a specific label is the sum of such weights for all neighbors with matching label. The weights are specified in nodeWeights and edgeWeights, respectively. If no values are provided, uniform weights of 1.0 are applied.
conversionfinal

subgraphEdges

: ItemCollection<IEdge>
Gets or sets the collection of edges which define a subset of the graph for the algorithms to work on.

If nothing is set, all edges of the graph will be processed.

If only the excludes are set, all edges in the graph except those provided in the excludes are processed.

Note that edges which start or end at nodes which are not in the subgraphNodes are automatically not considered by the algorithm.

ItemCollection<T> instances may be shared among algorithm instances and will be (re-)evaluated upon (re-)execution of the algorithm.

The edges provided here must be part of the graph which is passed to the run method.
conversionfinal

Examples

Calculating the clusters based on label propagation on a subset of the graph
// prepare the label propagation algorithm
const algorithm = new LabelPropagationClustering({
  // Ignore edges without target arrow heads
  subgraphEdges: {
    excludes: (edge: IEdge): boolean =>
      edge.style instanceof PolylineEdgeStyle &&
      edge.style.targetArrow instanceof Arrow &&
      edge.style.targetArrow.type === ArrowType.NONE,
  },
})
// run the algorithm
const result = algorithm.run(graph)

// highlight the nodes of the clusters with different styles
for (const node of graph.nodes) {
  const componentId = result.nodeClusterIds.get(node)!
  graph.setStyle(node, clusterStyles.get(componentId)!)
}

subgraphNodes

: ItemCollection<INode>
Gets or sets the collection of nodes which define a subset of the graph for the algorithms to work on.

If nothing is set, all nodes of the graph will be processed.

If only the excludes are set, all nodes in the graph except those provided in the excludes are processed.

ItemCollection<T> instances may be shared among algorithm instances and will be (re-)evaluated upon (re-)execution of the algorithm.

The nodes provided here must be part of the graph which is passed to the run method.
conversionfinal

Examples

Calculating the clusters based on label propagation on a subset of the graph
// prepare the label propagation algorithm
const algorithm = new LabelPropagationClustering({
  subgraphNodes: {
    // only consider elliptical nodes in the graph
    includes: (node: INode): boolean =>
      node.style instanceof ShapeNodeStyle &&
      node.style.shape === ShapeNodeShape.ELLIPSE,
    // but ignore the first node, regardless of its shape
    excludes: graph.nodes.first()!,
  },
})
// run the algorithm
const result = algorithm.run(graph)

// highlight the nodes of the clusters with different styles
for (const node of graph.nodes) {
  const componentId = result.nodeClusterIds.get(node)!
  graph.setStyle(node, clusterStyles.get(componentId)!)
}

Methods

run

(graph: IGraph): LabelPropagationClusteringResult
Detects the communities in the specified input graph by applying a label propagation algorithm.

This algorithm iteratively assigns a label to each node. The label of a node is set to the most frequent label among its neighbors. If there are multiple candidates the algorithm randomly chooses one of them. After applying the algorithm, all nodes with the same label belong to the same community.

For a given node and neighbor, the weight of the neighbor's label is nodeWeight(neighbor) * edgeWeight((node, neighbor)). In addition, the overall weight of a specific label is the sum of such weights for all neighbors with matching label. The weights are specified in nodeWeights and edgeWeights, respectively. If no values are provided uniform weights of 1.0 are applied.

If no initialLabels are provided each node starts with a unique label. Otherwise, at the start of the algorithm, each node is associated with the label specified with initialLabels. Negative values are ignored, i.e., at the start there may be nodes without a label. This implies that there may be results where some of the nodes are not associated with labels, too. Such nodes are not associated with any cluster.

If edgeDirectedness is specified, the algorithm only considers a subset of the neighbors when determining the label of a node. More precisely, it only considers the neighbors connected by edges to a node n for which

  • the edge's directedness is greater than 0.0 and the edge is incoming (i.e. n is the edge's target),
  • the edge's directedness is smaller than 0.0 and the edge is outgoing (i.e. n is the edge's source),
  • or the edge's directedness is 0.0 (i.e. the edge is considered to be undirected).
A node with a self-loop is considered to be a neighbor of itself. Hence, the node's label is also included when calculating the most frequent label among the neighbors. If you want to prevent this, you can explicitly exclude self-loop edges. Alternatively, you can specify suitable edge weights. Setting the weight of self-loops to 0.0 results in ignoring them.
If none of the neighbors of a node is associated with a label with positive weight, the node keeps its current label.
If no edgeDirectedness is specified, the algorithm assumes that all edges are undirected (i.e. have directedness 0.0).
The labels which are considered in this algorithm are internally assigned integer values. They have nothing to do with the marking objects which are commonly referred to as "label" in yFiles for HTML.
The result obtained from this algorithm is a snapshot which is no longer valid once the graph has changed, e.g. by adding or removing nodes or edges.
final

Parameters

graph: IGraph
The undirected input graph.

Return Value

LabelPropagationClusteringResult
The resulting clusters.

Throws

Exception ({ name: 'InvalidOperationError' })
If the algorithm can't create a valid result due to an invalid graph structure or wrongly configured properties.

Examples

Calculating clusters of a graph using the label propagation method
// prepare the label propagation algorithm
const algorithm = new LabelPropagationClustering({
  edgeWeights: (edge) =>
    edge.style instanceof PolylineEdgeStyle
      ? edge.style.stroke!.thickness
      : 0,
  nodeWeights: (node) => node.layout.width,
})
// run the algorithm
const result = algorithm.run(graph)

// highlight the nodes of the clusters with different styles
for (const node of graph.nodes) {
  const componentId = result.nodeClusterIds.get(node)!
  graph.setStyle(node, clusterStyles.get(componentId)!)
}
Excluding self-loop edges
const algorithm = new LabelPropagationClustering({
  // Ignore edges without target arrow heads
  subgraphEdges: {
    excludes: (edge: IEdge): boolean =>
      edge.sourceNode === edge.targetNode,
  },
})