private static Map getPredictionStatEdgeSampling()

in src/main/java/com/twitter/sbf/core/MHAlgorithm.java [629:688]
56 lines of code
9 McCabe index (conditional complexity)

  private static Map<String, PredictionStat> getPredictionStatEdgeSampling(
      Graph graph,
      SparseBinaryMatrix matrix,
      RandomAdaptor rng,
      int recallSamples,
      int precisionSamples,
      int verticesToSample) {
    PredictionStat recallStat = new PredictionStat();
    for (int i = 0; i < precisionSamples; i++) {
      // sample first vertex according to degree distribution
      int v1 = graph.getDegreeDistribution(rng).sample();
      if (graph.getDegree(v1) > 0) {
        int v2 = graph.getNeighbors(v1)[rng.nextInt(graph.getDegree(v1))];
        float edgeWeight = graph.getWeightOfEdge(v1, v2);
        boolean predictEdge = Util.hasCommonElement(matrix.getRow(v1), matrix.getRow(v2));
        recallStat.incActualPositive();
        recallStat.incWeightActualPositive(edgeWeight);
        if (predictEdge) {
          recallStat.incTruePositive();
          recallStat.incWeightTruePositive(edgeWeight);
        }
      }
    }

    double[] colSizesNormalized = new double[matrix.getNumCols()];
    int[] colIds = new int[matrix.getNumCols()];
    double totalSize = 0;
    for (int i = 0; i < colIds.length; i++) {
      colSizesNormalized[i] = matrix.getColumn(i).size() * (matrix.getColumn(i).size() - 1);
      colIds[i] = i;
      totalSize += colSizesNormalized[i];
    }
    PredictionStat precisionStat = new PredictionStat();
    for (int i = 0; i < colIds.length; i++) {
      colSizesNormalized[i] = colSizesNormalized[i] / totalSize;
      precisionStat.add(
          clusterPrecision(
              graph, matrix, i, (int) Math.ceil(colSizesNormalized[i] * recallSamples), rng
          )
      );
    }

    PredictionStat orphansStat = new PredictionStat();
    for (int i = 0; i < verticesToSample; i++) {
      int vId = rng.nextInt(graph.getNumVertices());
      int[] row = matrix.getRow(vId);
      if (row.length > 0) {
        orphansStat.incEvalVertices();
        if (isOrphan(graph, matrix, vId, row)) {
          orphansStat.incEvalVerticesWithZeroTruePos();
        }
      }
    }

    return ImmutableMap.of(
        "precision", precisionStat,
        "recall", recallStat,
        "orphans", orphansStat
    );
  }