/// <summary>
/// Computes a normalized version P of a given betweenness preference matrix B.
/// </summary>
/// <param name="x">The node for which the normalized matrix is computed</param>
/// <param name="aggregate_net">The weighted aggregate network</param>
/// <param name="B">The betweenness preference matrix that shall be normalized</param>
/// <returns>A normalized version of the betweenness preference matrix B</returns>
public static double[,] NormalizeMatrix(string x, WeightedNetwork aggregate_net, double[,] B)
{
// Normalize the matrix ( i.e. this is equation (3) )
double[,] P = new double[aggregate_net.GetIndeg(x), aggregate_net.GetOutdeg(x)];
double sum = 0d;
for (int s = 0; s < aggregate_net.GetIndeg(x); s++)
{
for (int d = 0; d < aggregate_net.GetOutdeg(x); d++)
{
sum += B[s, d];
}
}
if (sum > 0d)
{
for (int s = 0; s < aggregate_net.GetIndeg(x); s++)
{
for (int d = 0; d < aggregate_net.GetOutdeg(x); d++)
{
P[s, d] = B[s, d] / sum;
}
}
}
return(P);
}
/// <summary>
/// Computes the scalar betwenness preference of a node based on its normalized betweenness preference matrix
/// </summary>
/// <param name="aggregate_net">The temporal network for which to compute betweenness preference</param>
/// <param name="x">The node for which to compute betweenness preference</param>
/// <param name="P">The betweenness preference matrix based on which betw. pref. will be computed</param>
/// <returns>The betweenness preference, defined as the mutual information of the source and target of two-paths</returns>
public static double GetBetweennessPref(WeightedNetwork aggregate_net, string x, double[,] P, bool normalized = false)
{
// If the network is empty, just return zero
if (aggregate_net.VertexCount == 0)
{
return(0d);
}
// Compute the mutual information (i.e. betweenness preference)
double I = 0;
int indeg = aggregate_net.GetIndeg(x);
int outdeg = aggregate_net.GetOutdeg(x);
double[] marginal_s = new double[indeg];
double[] marginal_d = new double[outdeg];
// Marginal probabilities P_d = \sum_s'{P_{s'd}}
for (int d = 0; d < outdeg; d++)
{
double P_d = 0d;
for (int s_prime = 0; s_prime < indeg; s_prime++)
{
P_d += P[s_prime, d];
}
marginal_d[d] = P_d;
}
// Marginal probabilities P_s = \sum_d'{P_{sd'}}
for (int s = 0; s < indeg; s++)
{
double P_s = 0d;
for (int d_prime = 0; d_prime < outdeg; d_prime++)
{
P_s += P[s, d_prime];
}
marginal_s[s] = P_s;
}
double H_s = Entropy(marginal_s);
double H_d = Entropy(marginal_d);
// Here we just compute equation (4) of the paper ...
for (int s = 0; s < indeg; s++)
{
for (int d = 0; d < outdeg; d++)
{
if (P[s, d] != 0) // 0 * Log(0) = 0
// Mutual information
{
I += P[s, d] * Math.Log(P[s, d] / (marginal_s[s] * marginal_d[d]), 2d);
}
}
}
return(normalized?I / Math.Min(H_s, H_d):I);
}
/// <summary>
/// Computes the baseline betweenness preference matrix of a node under the assumption
/// that the temporal network does not contain a betweenness preference correlation. This corresponds to
/// equation (5) in the paper.
/// </summary>
/// <param name="v">The node to compute the baseline betweenness preference for</param>
/// <param name="aggregate_net">The weighted, aggregate ego network of node x based on which the matrix will be computed</param>
/// <param name="index_pred">Indices of predecessor nodes in the betweenness preference matrix</param>
/// <param name="index_succ">Indices of successor nodes in the betweenness preference matric</param>
/// <param name="normalize">Whether or not to normalize the betweenness preference matrix (i.e. whether B or P shall be returned)</param>
/// <returns>Depending on the normalization, a betweenness preference matrix B or the normalized version P will be returned</returns>
public static double[,] GetUncorrelatedBetweennessPrefMatrix(WeightedNetwork aggregate_net, string v, out Dictionary <string, int> index_pred, out Dictionary <string, int> index_succ)
{
// Use a mapping of indices to node labels
index_pred = new Dictionary <string, int>();
index_succ = new Dictionary <string, int>();
// Create an empty matrix
double[,] P = new double[aggregate_net.GetIndeg(v), aggregate_net.GetOutdeg(v)];
// Create the index-to-node mapping
int i = 0;
foreach (string u in aggregate_net.GetPredecessors(v))
{
index_pred[u] = i++;
}
i = 0;
foreach (string w in aggregate_net.GetSuccessors(v))
{
index_succ[w] = i++;
}
// Sum over the weights of all source nodes
double sum_source_weights = 0d;
foreach (string s_prime in aggregate_net.GetPredecessors(v))
{
sum_source_weights += aggregate_net.GetWeight(s_prime, v);
}
// Normalization factor for d
double sum_dest_weights = 0d;
foreach (string d_prime in aggregate_net.GetSuccessors(v))
{
sum_dest_weights += aggregate_net.GetWeight(v, d_prime);
}
double min_p = double.MaxValue;
// Equation (5) in the paper
foreach (string s in aggregate_net.GetPredecessors(v))
{
foreach (string d in aggregate_net.GetSuccessors(v))
{
P[index_pred[s], index_succ[d]] = (aggregate_net.GetWeight(s, v) / sum_source_weights) * (aggregate_net.GetWeight(v, d) / sum_dest_weights);
min_p = Math.Min(P[index_pred[s], index_succ[d]], min_p);
}
}
return(P);
}
