/**
* Remove all edges connecting dead cells with others Call this after loading this entity from db
* OR after config is changed. (i.e., after ratio or actualW or actualH is changed)
*/
public void updateGraph() {
cellContainer = new Cell[rowCount][colCount];
initGraph();
int row, col;
for (DeadPoint dp : deadPoints) {
col = getCorrespondingCol(dp.getX());
row = getCorrespondingRow(dp.getY());
if (valid(row, col)) g.removeVertex(cellContainer[row][col]);
}
}
/**
* Disable the cell containing the given point
*
* @param x
* @param y
*/
public void disableCell(int x, int y) {
int col = getCorrespondingCol(x);
int row = getCorrespondingRow(y);
// If a cell is already dead, it wouldn't be disabled again; Do this to reduce the number of
// dead points created
if (valid(row, col) && !cellContainer[row][col].isDead()) {
cellContainer[row][col].disableCell();
deadPoints.add(new DeadPoint(x, y, this));
// removing this cell from the graph and all of its touching edges
g.removeVertex(cellContainer[row][col]);
}
}
public static void main(String[] args) throws IOException {
System.out.println("Enter parameters: N p_w p_r steps:");
try {
BufferedReader reader = new BufferedReader(new InputStreamReader(System.in));
String paras[] = reader.readLine().split(" ");
N = Integer.parseInt(paras[0]);
p_w = Float.parseFloat(paras[1]);
p_r = Float.parseFloat(paras[2]);
steps = Integer.parseInt(paras[3]);
} catch (IOException e) {
System.out.println("Error reading from user");
}
long start = System.currentTimeMillis();
int simulations = 1;
for (int t = 1; t <= simulations; t++) {
// Initialization, generate an empty network of N nodes
EIM dynamic = new EIM();
G = new SimpleWeightedGraph<Integer, DefaultWeightedEdge>(DefaultWeightedEdge.class);
for (int i = 0; i < N; i++) G.addVertex(i);
String para = "EIM_" + Integer.toString(t);
String folder = "files/" + "Networks" + "/";
File f = new File(folder);
if (f.exists() == false) {
f.mkdirs();
}
// distribution of social space
socialposition = dynamic.uniform(N);
for (int s = 0; s < steps; s++) {
int v = random.nextInt(N);
// LA process - weighted random walk + link with social distance
dynamic.localattach(v);
// GA process
// no edges, create one random link
if (G.edgesOf(v).size() == 0) {
dynamic.globalattach(v);
} else {
float prob = random.nextFloat();
if (prob < p_r) dynamic.globalattach(v);
}
// ND process
int d = random.nextInt(N);
if (G.edgesOf(d).size() > 0) {
float prob = random.nextFloat();
if (prob < p_d) {
Set<DefaultWeightedEdge> edges = new HashSet(G.edgesOf(d));
G.removeAllEdges(edges);
}
}
if (s % 100000 == 0) {
System.out.print("Steps:" + Integer.toString(s) + " ");
System.out.print("Edges:");
System.out.print(G.edgeSet().size() + " ");
System.out.println(
"Avg_degree: "
+ Float.toString((float) G.edgeSet().size() * 2 / G.vertexSet().size()));
}
}
// delete isolate nodes
ArrayList<Integer> nodelist = new ArrayList<Integer>();
for (int node : G.vertexSet()) nodelist.add(node);
for (int node : nodelist) {
if (G.degreeOf(node) == 0) G.removeVertex(node);
}
System.out.print("Nodes:");
System.out.println(G.vertexSet().size());
System.out.print("Edges:");
System.out.println(G.edgeSet().size());
// get largest connected component
Statistics stat = new Statistics();
Graph G_LCC = stat.largestConnectedComponent(G);
ExportGraph export = new ExportGraph();
export.exportWPairs((SimpleWeightedGraph<Integer, DefaultWeightedEdge>) G_LCC, para, folder);
System.out.print("Nodes in LCC:");
System.out.println(G_LCC.vertexSet().size());
System.out.print("Edges in LCC:");
System.out.println(G_LCC.edgeSet().size());
System.out.println("Avg_degree: " + Double.toString(stat.avg_degree(G_LCC)));
System.out.println("Avg_clustering: " + Double.toString(stat.avg_clustering(G_LCC)));
System.out.println(
"Degree assortativity: " + Double.toString(stat.assortativityCoefficient(G_LCC)));
}
long elapsedTimeMillis = System.currentTimeMillis() - start;
float elapsedTimeHour = elapsedTimeMillis / (60 * 60 * 1000F);
System.out.print("Elapsed time: ");
System.out.print(elapsedTimeHour);
System.out.println(" hours.");
}
/**
* Simulate next step for the given particle.
*
* @param nextPts
* @param particleNum
*/
public void simulateForwardStep(ArrayList<Point> nextPts, Particle thisParticle) {
HashMap<Vertex, Point> thisObservedPosMap = new HashMap<Vertex, Point>();
HashMap<Vertex, AntPath> thisAntPathMap = new HashMap<Vertex, AntPath>();
thisParticle.currentFalsePositives = new ArrayList<Point>();
for (Point pt : nextPts) {
Vertex vx = new ObsVertex();
thisObservedPosMap.put(vx, pt);
}
for (AntPath ap : thisParticle.getPaths()) {
Vertex vx = new PathVertex();
thisAntPathMap.put(vx, ap);
}
/*
* Compute probability graph for this particle.
*/
SimpleWeightedGraph<Vertex, DefaultWeightedComparableEdge> probabilityGraph =
new SimpleWeightedGraph<Vertex, DefaultWeightedComparableEdge>(
DefaultWeightedComparableEdge.class);
ValueSortedMap<DefaultWeightedComparableEdge, Double> edgeMap =
new ValueSortedMap<DefaultWeightedComparableEdge, Double>(true);
computeLogProbabilityGraph(probabilityGraph, edgeMap, thisObservedPosMap, thisAntPathMap);
// displayLogProbabilityGraph(probabilityGraph,thisObservedPosMap,thisAntPathMap,vertexSums);
/*
* Generate new ant locations based on probability graph. We do this by sampling the most likely event,
* removing this event from the probability graph, updating the prob. graph, then repeating until all
* observations/causes are accounted for.
*
* To do this (relatively) efficiently, we maintain a sorted hash-map of possible observation/cause pairs.
* To avoid having to rescale the entire hash-map every step, we maintain the current sum of weights.
*
* We're also keeping track of the posterior log-probability of the simulated step.
*/
double logprob = 0;
int nfp = 0;
int nfn = 0;
int doOutput = 0;
while (edgeMap.size() > 0) {
/*
* Sample an event from the probability graph.
*/
DefaultWeightedComparableEdge event = sampleEdge(probabilityGraph, edgeMap);
edgeMap.remove(event);
double edgeWeight = probabilityGraph.getEdgeWeight(event);
double thisLogProb = edgeWeight;
logprob = logprob + thisLogProb;
Vertex v1 = probabilityGraph.getEdgeSource(event);
Vertex v2 = probabilityGraph.getEdgeTarget(event);
if (doOutput > 0) {
System.err.println("edgeWeight: " + edgeWeight);
doOutput = outputInterestingStuff(v1, v2, probabilityGraph);
}
if (v1.getClass().equals(ObsVertex.class)) {
assert (v2.getClass().equals(PathVertex.class));
Vertex tv = v1;
v1 = v2;
v2 = tv;
}
/*
* Update the probability graph.
*/
if (!v1.equals(falsePositive)) {
Set<DefaultWeightedComparableEdge> es = probabilityGraph.edgesOf(v1);
for (DefaultWeightedComparableEdge ed : es) edgeMap.remove(ed);
boolean tt = probabilityGraph.removeVertex(v1);
assert (tt);
} else {
nfp += 1;
thisParticle.currentFalsePositives.add(thisObservedPosMap.get(v2));
}
if (!v2.equals(falseNegative)) {
Set<DefaultWeightedComparableEdge> es = probabilityGraph.edgesOf(v2);
for (DefaultWeightedComparableEdge ed : es) edgeMap.remove(ed);
boolean tt = probabilityGraph.removeVertex(v2);
assert (tt);
} else nfn += 1;
// Utils.computeVertexSums(probabilityGraph,vertexSums);
/*
* Update AntPath trajectory.
*/
if (!v1.equals(falsePositive)) {
AntPath ap = thisAntPathMap.get(v1);
assert (ap != null);
Point obs = thisObservedPosMap.get(v2);
Point newPos = new Point();
double thislp = sampleConditionalPos(ap, obs, newPos) + thisLogProb;
ap.updatePosition(newPos, obs, thislp);
}
/*
* Compute likelihoods.
*/
}
}
/**
* Remove all nodes which are NOT in cells from graph g
*
* @param cells
*/
private void filterGraph(ArrayList<Cell> cells) {
for (int row = 0; row < rowCount; ++row)
for (int col = 0; col < colCount; ++col)
if (!cells.contains(cellContainer[row][col])) g.removeVertex(cellContainer[row][col]);
}
