public double[][] compute(SingleDTGraph data) {
SparseVector[] featureVectors = computeFeatureVectors(data);
double[][] kernel =
KernelUtils.initMatrix(data.getInstances().size(), data.getInstances().size());
long tic = System.currentTimeMillis();
kernel = KernelUtils.computeKernelMatrix(featureVectors, kernel);
compTime += System.currentTimeMillis() - tic;
return kernel;
}
private static Map<Long, Map<Boolean, Map<Integer, Pair<SingleDTGraph, List<Double>>>>>
createDataSetCache(
RDFDataSet tripleStore,
LargeClassificationDataSet data,
long[] seeds,
double fraction,
int minSize,
int maxClasses,
int[] depths,
boolean[] inference) {
Map<Long, Map<Boolean, Map<Integer, Pair<SingleDTGraph, List<Double>>>>> cache =
new HashMap<Long, Map<Boolean, Map<Integer, Pair<SingleDTGraph, List<Double>>>>>();
for (long seed : seeds) {
cache.put(seed, new HashMap<Boolean, Map<Integer, Pair<SingleDTGraph, List<Double>>>>());
data.createSubSet(seed, fraction, minSize, maxClasses);
for (boolean inf : inference) {
cache.get(seed).put(inf, new HashMap<Integer, Pair<SingleDTGraph, List<Double>>>());
for (int depth : depths) {
System.out.println("Getting Statements...");
Set<Statement> stmts =
RDFUtils.getStatements4Depth(
tripleStore, data.getRDFData().getInstances(), depth, inf);
System.out.println("# Statements: " + stmts.size());
stmts.removeAll(new HashSet<Statement>(data.getRDFData().getBlackList()));
System.out.println("# Statements: " + stmts.size() + ", after blackList");
System.out.println("Building Graph...");
SingleDTGraph graph =
RDFUtils.statements2Graph(
stmts, RDFUtils.REGULAR_LITERALS, data.getRDFData().getInstances(), true);
System.out.println(
"Built Graph with "
+ graph.getGraph().nodes().size()
+ ", and "
+ graph.getGraph().links().size()
+ " links");
cache
.get(seed)
.get(inf)
.put(
depth,
new Pair<SingleDTGraph, List<Double>>(
graph, new ArrayList<Double>(data.getTarget())));
}
}
}
return cache;
}
public SparseVector[] computeFeatureVectors(SingleDTGraph data) {
SparseVector[] featureVectors = new SparseVector[data.getInstances().size()];
for (int i = 0; i < featureVectors.length; i++) {
featureVectors[i] = new SparseVector();
}
init(data.getGraph(), data.getInstances());
WeisfeilerLehmanIterator<DTGraph<StringLabel, StringLabel>> wl =
new WeisfeilerLehmanDTGraphIterator(reverse, noDuplicateSubtrees);
List<DTGraph<StringLabel, StringLabel>> gList =
new ArrayList<DTGraph<StringLabel, StringLabel>>();
gList.add(rdfGraph);
long tic = System.currentTimeMillis();
wl.wlInitialize(gList);
double weight = 1.0;
computeFVs(rdfGraph, instanceVertices, weight, featureVectors, wl.getLabelDict().size() - 1, 0);
for (int i = 0; i < iterations; i++) {
wl.wlIterate(gList);
computeFVs(
rdfGraph, instanceVertices, weight, featureVectors, wl.getLabelDict().size() - 1, i + 1);
}
compTime = System.currentTimeMillis() - tic;
if (this.normalize) {
featureVectors = KernelUtils.normalize(featureVectors);
}
return featureVectors;
}
