Expert parameters for mode AUTOMATIC

Fine-tuning the clustering

The clustering mode AUTOMATIC takes the specified clusters as initial clusters into account. If some or all nodes have no specified cluster, it calculates clusters from the topology of the graph.

This mode takes several topology criteria into account:

Biconnectivity
Minimum number of nodes per cluster
Maximum number of nodes per cluster
High degree nodes to form star clusters

Clustering by biconnectivity

If a set of nodes has many connections among each other, they must form a cluster. The informal property "many connections" can be mathematically described by the term biconnectivity: A set of nodes is biconnected if removing only one node or only one link keeps the nodes connected. For example, a connection ring is always biconnected. It is a suitable criterion for clusters. The algorithm detects by default first the biconnected parts of the graph, and forms a regular cluster (not a star cluster) for each part. If it is not suitable, the detection of biconnected components can also be disabled.

Example of disabling the detection of biconnected parts (CL algorithm)

To disable the detection of biconnected components:

In CSS

Add to the GraphLayout section:

clusterByBiconnectivity: "false";

In Java

Use the method:

void setClusterByBiconnectivity(boolean enable)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Minimum cluster size

It is possible to specify the minimal size of regular clusters (not of star clusters). If the initial clusters are too small, they are merged into larger clusters.

Example of specifying the minimum cluster size (CL algorithm)

To specify the minimum cluster size:

In CSS

Add to the GraphLayout section:

minimumClusterSize: "10";

In Java

Use the method:

void setMinimumClusterSize(int size)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Maximum cluster size

It is possible to specify the maximal size of regular clusters (not of star clusters). If the initial clusters are too large, they are split into smaller clusters. This value is only a hint. If the minimum cluster size and the maximum cluster size are close to each other, the maximum cluster size cannot always be respected.

Example of specifying the maximum cluster size (CL algorithm)

To specify the maximum cluster size:

In CSS

Add to the GraphLayout section:

maximumClusterSize: "30";

In Java

Use the method:

void setMaximumClusterSize(int size)

Splitting clusters is an iterative process. It is possible to specify the maximum number of iterations to obtain the maximum cluster size; the more iterations, the better the quality, and the slower the layout.

Example of specifying the maximum number of max cluster iterations (CL algorithm)

To specify the maximum number of iterations:

In CSS

Add to the GraphLayout section:

maximumNumberOfIterationsToReachMaxClusterSize: "15";

In Java

Use the method:

void setMaximumNumberOfIterationsToReachMaxClusterSize(int numIterations)

These parameters have no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Minimum star cluster size

Besides regular clusters, the layout algorithm can also detect star clusters. It detects nodes with high degree that have many unclustered neighbors and forms them into clusters with the high degree node as star center

Example of specifying the minimum star cluster size (CL algorithm)

To specify the minimum degree to form a node and its neighbor to a star cluster:

In CSS

Add to the GraphLayout section:

highDegreeStarClusterSize: "10";

In Java

Use the method:

void setHighDegreeStarClusterSize(int size)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Fine-tuning the crossing reduction

If the ordering of the nodes within each cluster is not specified by cluster indices (see Cluster membership and order of the nodes in a cluster (CL)), then the AUTOMATIC mode reorders the nodes within each cluster to avoid link crossings. It is only a heuristic and cannot always guarantee the minimum number of link crossings. There are two types of links: intracluster links connect nodes that belong to the same cluster, and intercluster links connect nodes of different clusters. During the crossing reduction, intracluster links compete with intercluster links. Some node orderings of a cluster produce more crossings of links inside the cluster, while other node orderings produce more crossings of links connecting different clusters. The behavior can be controlled by crossing weights.

Intercluster link crossing reduction

The crossing reduction step that considers intercluster links is the slowest part of the crossing reduction. It is enabled by default, but it can be disabled. If disabled, only intracluster links are considered to calculate the ordering of nodes within each cluster.

Example of disabling intercluster link crossing reduction (CL algorithm)

To disable the intercluster link crossing reduction:

In CSS

Add to the GraphLayout section:

interClusterCrossingReduction: "false";

In Java

Use the method:

void setInterClusterCrossingReduction(boolean enable)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Intracluster link crossing penalty

The crossing reduction tries to minimize the sum of penalties of crossings. If two intracluster links cross, it contributes to the sum with the intracluster penalty. Increase this penalty if intracluster links must have a higher weight when calculating the node ordering for each cluster.

Example of specifying the intracluster penalty (CL algorithm)

To specify the intracluster link crossing penalty:

In CSS

Add to the GraphLayout section:

intraClusterLinkCrossingPenalty: "2.0";

In Java

Use the method:

void setIntraClusterLinkCrossingPenalty(float penalty)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Intercluster link crossing penalty

If two intercluster links cross, it contributes to the penalty sum with the intercluster penalty. Increase this penalty if intercluster links must have a higher weight when calculating the node ordering for each cluster.

Example of specifying the intercluster penalty (CL algorithm)

To specify the intercluster link crossing penalty:

In CSS

Add to the GraphLayout section:

interClusterLinkCrossingPenalty: "2.0";

In Java

Use the method:

void setInterClusterLinkCrossingPenalty(float penalty)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Mixed cluster link crossing penalty

If an intracluster link crosses an intercluster link, it contributes to the penalty sum with the mixed cluster penalty.

Example of specifying the mixed cluster penalty (CL algorithm)

To specify the mixed cluster link crossing penalty:

In CSS

Add to the GraphLayout section:

mixedClusterLinkCrossingPenalty: "2.0";

In Java

Use the method:

void setMixedClusterLinkCrossingPenalty(float penalty)

This parameter has no effect if the clustering mode is BY_CLUSTER_IDS or BY_SUBGRAPHS .

Layout of cluster graph

The nodes of each cluster are placed in a circle. Then each circle is placed in the plane. The circular layout uses by default a tree layout in radial mode to place the circles.

Example of handling layout of cluster graphs

The layout parameters of the tree layout can be changed by accessing the tree layout:

In Java

To access the layout of the cluster graph, use the method:

IlvGraphLayout getLayoutOfClusterGraph()

By default, the layout is an instance of IlvTreeLayout . You can set any global tree layout parameter on this layout instance.

You can also specify a different layout instance to place the clusters:

In CSS

Add to the GraphLayout section:

layoutOfClusterGraph: "@#MyLayout";

and a new rule that specifies the details of MyLayout , for instance:

Subobject#MyLayout {
   class : "ilog.views.graphlayout.uniformlengthedges.IlvUniformLengthEdgesLayout" ;
   respectNodeSizes : "true" ;
}

In Java

Use the method:

void setLayoutOfClusterGraph(IlvGraphLayout layout)

Handling of disconnected graphs

The layout algorithm can use the generic mechanism to lay out connected components. (For more information about this mechanism, see Layout of connected components). This generic mechanism is available at the circular layout instance itself, but as well at the tree layout instance that lays out the cluster graph.

Example of handling of disconnected graphs

You can enable it on the circular layout itself:

In CSS

Add to the GraphLayout section:

layoutOfConnectedComponentsEnabled: "true";
layoutOfConnectedComponents: "@#SomeGridLayout";

In Java

Call on the circular layout instance:

circularLayout.setLayoutOfConnectedComponentsEnabled(true);
circularLayout.setLayoutOfCOnnectedComponents(someGridLayout);

If it is disabled on the circular layout, the layout of the cluster graph handles disconnected components. You can enable the generic handling of disconnected components as well on the layout of the cluster graph (if it is supported there, which is true for the default tree layout) via

In Java

circularLayout.getLayoutOfClusterGraph().setLayoutOfConnectedComponentsEnabled(true);
circularLayout.getLayoutOfClusterGraph().setLayoutOfCOnnectedComponents(someGridLayout);

but it has the same effect as if you specify it directly on the circular layout instance.