/**
* Finds the set of clusters which have the strongest "community structure". The more edges
* removed the smaller and more cohesive the clusters.
*
* @param graph the graph
*/
public Set<Set<V>> transform(Graph<V, E> graph) {
if (mNumEdgesToRemove < 0 || mNumEdgesToRemove > graph.getEdgeCount()) {
throw new IllegalArgumentException("Invalid number of edges passed in.");
}
edges_removed.clear();
for (int k = 0; k < mNumEdgesToRemove; k++) {
BetweennessCentrality<V, E> bc = new BetweennessCentrality<V, E>(graph);
E to_remove = null;
double score = 0;
for (E e : graph.getEdges()) {
Integer weight = mWeight.get(e);
if (bc.getEdgeScore(e) / weight > score) {
to_remove = e;
score = bc.getEdgeScore(e) / weight;
}
}
edges_removed.put(to_remove, graph.getEndpoints(to_remove));
graph.removeEdge(to_remove);
}
WeakComponentClusterer<V, E> wcSearch = new WeakComponentClusterer<V, E>();
Set<Set<V>> clusterSet = wcSearch.transform(graph);
for (Map.Entry<E, Pair<V>> entry : edges_removed.entrySet()) {
Pair<V> endpoints = entry.getValue();
graph.addEdge(entry.getKey(), endpoints.getFirst(), endpoints.getSecond());
}
return clusterSet;
}
protected void initialize(Graph<V, E> graph) {
this.graph = graph;
this.vertex_scores = new HashMap<V, Double>();
this.edge_scores = new HashMap<E, Double>();
this.vertex_data = new HashMap<V, BetweennessData>();
for (V v : graph.getVertices()) this.vertex_scores.put(v, 0.0);
for (E e : graph.getEdges()) this.edge_scores.put(e, 0.0);
}
/**
* Calculates betweenness scores based on the all-pairs weighted shortest paths in the graph.
*
* <p>NOTE: This version of the algorithm may not work correctly on all graphs; we're still
* working out the bugs. Use at your own risk.
*
* @param graph the graph for which the scores are to be calculated
* @param edge_weights the edge weights to be used in the path length calculations
*/
public BetweennessCentrality(Graph<V, E> graph, Transformer<E, ? extends Number> edge_weights) {
// reject negative-weight edges up front
for (E e : graph.getEdges()) {
double e_weight = edge_weights.transform(e).doubleValue();
if (e_weight < 0)
throw new IllegalArgumentException("Weight for edge '" + e + "' is < 0: " + e_weight);
}
initialize(graph);
computeBetweenness(new MapBinaryHeap<V>(new BetweennessComparator()), edge_weights);
}
/**
* Searches the graph and searches for the shortest path from a given start node to the
* destination. Returns {@code null} if there are no vertices, edges, or path to the goal,
* otherwise a list with the order of vertices on the shortest path.
*
* @param graph The graph to search
* @param source The vertex to start the search from
* @param destination The goal vertex
* @return A list with the order of vertices on the shortest path, null if no path exists in the
* graph.
*/
public List<V> search(Graph<V, E> graph, V source, V destination) {
// Check if it is even possible to find a path, return null
// if the graph has no vertices or edges
if (graph.getVertexCount() == 0) {
System.out.println("No nodes in the graph, " + "no shortest path can be found");
return null;
} else if (graph.getEdgeCount() == 0) {
System.out.println("No edges in graph, no path " + "can be found");
return null;
}
// Keep record of distance to each vertex, map each vertex
// in the graph to it's distance
HashMap<V, Number> distanceTable = new HashMap<>();
// Unvisited node queue, uses a pair <Vertex, Double> and ordered
// by the distance to the vertex
PriorityQueue<Pair<V, Number>> queue = new PriorityQueue<>(new QueueComparator());
// Map of nodes on the path, parents is value, key is child
HashMap<V, V> parent = new HashMap<>();
Number maxValue;
E edgeTest = graph.getEdges().iterator().next();
// This is so ugly, I hate Java Numbers
int numberType = 0;
if (edgeTest.getWeight() instanceof Integer) {
numberType = 1;
} else if (edgeTest.getWeight() instanceof Double) {
numberType = 2;
}
// Place each vertex in the map, initialize distances and put
// the pairings into the queue.
for (V vertex : graph.getVertices()) {
if (numberType == 1) {
maxValue = Integer.MAX_VALUE;
if (vertex.equals(source)) {
distanceTable.put(source, 0);
queue.add(new Pair<>(vertex, 0));
} else {
distanceTable.put(vertex, Integer.MAX_VALUE);
queue.add(new Pair<>(vertex, maxValue));
}
} else if (numberType == 2) {
maxValue = Double.MAX_VALUE;
if (vertex.equals(source)) {
distanceTable.put(source, 0.0);
queue.add(new Pair<>(vertex, 0.0));
} else {
distanceTable.put(vertex, Double.MAX_VALUE);
queue.add(new Pair<>(vertex, maxValue));
}
}
}
parent.put(source, null);
while (!queue.isEmpty()) {
Pair<V, Number> topPair = queue.remove();
V vertex = topPair.getLeft();
// Goal test, return the list of nodes on the path
// if we reach the destination
if (vertex.equals(destination)) {
return tracePath(parent, destination);
}
Collection<V> neighbours = graph.getNeighbors(vertex);
for (V neighbour : neighbours) {
E edge = graph.findEdge(vertex, neighbour);
assert (edge != null);
// Test for type of number used for weight, work accordingly
// Did I mention I hate the Java Number class.
if (numberType == 1) {
Integer alternateDistance = (Integer) edge.getWeight();
if (alternateDistance < (Integer) distanceTable.get(neighbour)) {
distanceTable.put(neighbour, alternateDistance);
parent.put(neighbour, vertex);
queue.add(new Pair<>(neighbour, alternateDistance));
}
} else if (numberType == 2) {
Double alternateDistance = (Double) edge.getWeight();
if (alternateDistance < (Double) distanceTable.get(neighbour)) {
distanceTable.put(neighbour, alternateDistance);
parent.put(neighbour, vertex);
queue.add(new Pair<>(neighbour, alternateDistance));
}
}
}
}
// Exhausted all possible paths from source, could not find a path
// to the goal.
return null;
}
public void testMixedSaveLoadSave() throws IOException {
Graph<Number, Number> graph1 = new SparseMultigraph<Number, Number>();
for (int i = 0; i < 5; i++) {
graph1.addVertex(i);
}
int j = 0;
List<Number> id = new ArrayList<Number>(graph1.getVertices());
GreekLabels<Number> gl = new GreekLabels<Number>(id);
Number[] edges = {0, 1, 2, 3, 4, 5};
graph1.addEdge(j++, 0, 1, EdgeType.DIRECTED);
graph1.addEdge(j++, 0, 2, EdgeType.DIRECTED);
graph1.addEdge(j++, 1, 2, EdgeType.DIRECTED);
graph1.addEdge(j++, 1, 3);
graph1.addEdge(j++, 1, 4);
graph1.addEdge(j++, 4, 3);
Map<Number, Number> nr = new HashMap<Number, Number>();
for (int i = 0; i < edges.length; i++) {
nr.put(edges[i], new Float(Math.random()));
}
assertEquals(graph1.getEdgeCount(), 6);
// System.err.println(" mixed graph1 = "+graph1);
// for(Number edge : graph1.getEdges()) {
// System.err.println("edge "+edge+" is directed? "+graph1.getEdgeType(edge));
// }
// for(Number v : graph1.getVertices()) {
// System.err.println(v+" outedges are "+graph1.getOutEdges(v));
// System.err.println(v+" inedges are "+graph1.getInEdges(v));
// System.err.println(v+" incidentedges are "+graph1.getIncidentEdges(v));
// }
String testFilename = "mtest.net";
String testFilename2 = testFilename + "2";
// assign arbitrary locations to each vertex
Map<Number, Point2D> locations = new HashMap<Number, Point2D>();
for (Number v : graph1.getVertices()) {
locations.put(v, new Point2D.Double(v.intValue() * v.intValue(), 1 << v.intValue()));
}
Function<Number, Point2D> vld = Functions.forMap(locations);
PajekNetWriter<Number, Number> pnw = new PajekNetWriter<Number, Number>();
pnw.save(graph1, testFilename, gl, Functions.forMap(nr), vld);
Graph<Number, Number> graph2 = pnr.load(testFilename, graphFactory);
Function<Number, String> pl = pnr.getVertexLabeller();
List<Number> id2 = new ArrayList<Number>(graph2.getVertices());
Function<Number, Point2D> vld2 = pnr.getVertexLocationTransformer();
assertEquals(graph1.getVertexCount(), graph2.getVertexCount());
assertEquals(graph1.getEdgeCount(), graph2.getEdgeCount());
// test vertex labels and locations
for (int i = 0; i < graph1.getVertexCount(); i++) {
Number v1 = id.get(i);
Number v2 = id2.get(i);
assertEquals(gl.apply(v1), pl.apply(v2));
assertEquals(vld.apply(v1), vld2.apply(v2));
}
// test edge weights
Function<Number, Number> nr2 = pnr.getEdgeWeightTransformer();
for (Number e2 : graph2.getEdges()) {
Pair<Number> endpoints = graph2.getEndpoints(e2);
Number v1_2 = endpoints.getFirst();
Number v2_2 = endpoints.getSecond();
Number v1_1 = id.get(id2.indexOf(v1_2));
Number v2_1 = id.get(id2.indexOf(v2_2));
Number e1 = graph1.findEdge(v1_1, v2_1);
assertNotNull(e1);
assertEquals(nr.get(e1).floatValue(), nr2.apply(e2).floatValue(), 0.0001);
}
pnw.save(graph2, testFilename2, pl, nr2, vld2);
compareIndexedGraphs(graph1, graph2);
pnr.setVertexLabeller(null);
Graph<Number, Number> graph3 = pnr.load(testFilename2, graphFactory);
compareIndexedGraphs(graph2, graph3);
File file1 = new File(testFilename);
File file2 = new File(testFilename2);
Assert.assertTrue(file1.length() == file2.length());
file1.delete();
file2.delete();
}
protected void computeBetweenness(Queue<V> queue, Transformer<E, ? extends Number> edge_weights) {
for (V v : graph.getVertices()) {
// initialize the betweenness data for this new vertex
for (V s : graph.getVertices()) this.vertex_data.put(s, new BetweennessData());
// if (v.equals(new Integer(0)))
// System.out.println("pause");
vertex_data.get(v).numSPs = 1;
vertex_data.get(v).distance = 0;
Stack<V> stack = new Stack<V>();
// Buffer<V> queue = new UnboundedFifoBuffer<V>();
// queue.add(v);
queue.offer(v);
while (!queue.isEmpty()) {
// V w = queue.remove();
V w = queue.poll();
stack.push(w);
BetweennessData w_data = vertex_data.get(w);
for (E e : graph.getOutEdges(w)) {
// TODO (jrtom): change this to getOtherVertices(w, e)
V x = graph.getOpposite(w, e);
if (x.equals(w)) continue;
double wx_weight = edge_weights.transform(e).doubleValue();
// for(V x : graph.getSuccessors(w))
// {
// if (x.equals(w))
// continue;
// FIXME: the other problem is that I need to
// keep putting the neighbors of things we've just
// discovered in the queue, if they're undiscovered or
// at greater distance.
// FIXME: this is the problem, right here, I think:
// need to update position in queue if distance changes
// (which can only happen with weighted edges).
// for each outgoing edge e from w, get other end x
// if x not already visited (dist x < 0)
// set x's distance to w's dist + edge weight
// add x to queue; pri in queue is x's dist
// if w's dist + edge weight < x's dist
// update x's dist
// update x in queue (MapBinaryHeap)
// clear x's incoming edge list
// if w's dist + edge weight = x's dist
// add e to x's incoming edge list
BetweennessData x_data = vertex_data.get(x);
double x_potential_dist = w_data.distance + wx_weight;
if (x_data.distance < 0) {
// queue.add(x);
// vertex_data.get(x).distance =
// vertex_data.get(w).distance + 1;
x_data.distance = x_potential_dist;
queue.offer(x);
}
// note:
// (1) this can only happen with weighted edges
// (2) x's SP count and incoming edges are updated below
if (x_data.distance > x_potential_dist) {
x_data.distance = x_potential_dist;
// invalidate previously identified incoming edges
// (we have a new shortest path distance to x)
x_data.incomingEdges.clear();
// update x's position in queue
((MapBinaryHeap<V>) queue).update(x);
}
// if (vertex_data.get(x).distance ==
// vertex_data.get(w).distance + 1)
//
// if (x_data.distance == x_potential_dist)
// {
// x_data.numSPs += w_data.numSPs;
//// vertex_data.get(x).predecessors.add(w);
// x_data.incomingEdges.add(e);
// }
}
for (E e : graph.getOutEdges(w)) {
V x = graph.getOpposite(w, e);
if (x.equals(w)) continue;
double e_weight = edge_weights.transform(e).doubleValue();
BetweennessData x_data = vertex_data.get(x);
double x_potential_dist = w_data.distance + e_weight;
if (x_data.distance == x_potential_dist) {
x_data.numSPs += w_data.numSPs;
// vertex_data.get(x).predecessors.add(w);
x_data.incomingEdges.add(e);
}
}
}
while (!stack.isEmpty()) {
V x = stack.pop();
// for (V w : vertex_data.get(x).predecessors)
for (E e : vertex_data.get(x).incomingEdges) {
V w = graph.getOpposite(x, e);
double partialDependency =
vertex_data.get(w).numSPs
/ vertex_data.get(x).numSPs
* (1.0 + vertex_data.get(x).dependency);
vertex_data.get(w).dependency += partialDependency;
// E w_x = graph.findEdge(w, x);
// double w_x_score = edge_scores.get(w_x).doubleValue();
// w_x_score += partialDependency;
// edge_scores.put(w_x, w_x_score);
double e_score = edge_scores.get(e).doubleValue();
edge_scores.put(e, e_score + partialDependency);
}
if (!x.equals(v)) {
double x_score = vertex_scores.get(x).doubleValue();
x_score += vertex_data.get(x).dependency;
vertex_scores.put(x, x_score);
}
}
}
if (graph instanceof UndirectedGraph) {
for (V v : graph.getVertices()) {
double v_score = vertex_scores.get(v).doubleValue();
v_score /= 2.0;
vertex_scores.put(v, v_score);
}
for (E e : graph.getEdges()) {
double e_score = edge_scores.get(e).doubleValue();
e_score /= 2.0;
edge_scores.put(e, e_score);
}
}
vertex_data.clear();
}