/**
* Calculate the distance between two clusters, forcing them to have the same words in them, and to not be related
* to each other.
*
* @param cluster1 the first cluster
* @param cluster2 the second cluster
* @return The intra-cluster distance, or Double.NEGATIVE_INFINITY if the clusters should never be clustered
* together.
*/
protected double intraClusterDistance(Cluster cluster1, Cluster cluster2)
{
LogMath logMath = LogMath.getLogMath();
double maxSim = Double.NegativeInfinity;
foreach (Node node1 in cluster1.getElements())
{
foreach (Node node2 in cluster2.getElements())
{
if (!node1.getWord().getSpelling().Equals(
node2.getWord().getSpelling()))
{
return(Double.NegativeInfinity);
}
if (node1.hasAncestralRelationship(node2))
{
return(Double.NegativeInfinity);
}
double overlap = getOverlap(node1, node2);
if (overlap > 0.0)
{
overlap = logMath.logToLinear((float)overlap);
overlap += node1.getPosterior() + node2.getPosterior();
if (overlap > maxSim)
{
maxSim = overlap;
}
}
}
}
return(maxSim);
}
private float[] computePosterios(float[] componentScores, int numStreams)
{
float[] posteriors = componentScores;
int step = componentScores.Length / numStreams;
int startIdx = 0;
for (int i = 0; i < numStreams; i++)
{
float max = posteriors[startIdx];
for (int j = startIdx + 1; j < startIdx + step; j++)
{
if (posteriors[j] > max)
{
max = posteriors[j];
}
}
for (int j = startIdx; j < startIdx + step; j++)
{
posteriors[j] = (float)logMath.logToLinear(posteriors[j] - max);
}
startIdx += step;
}
return(posteriors);
}
/**
* Calculate the distance between two clusters
*
* @param c1 the first cluster
* @param c2 the second cluster
* @return the inter cluster similarity, or Double.NEGATIVE_INFINITY if these clusters should never be clustered
* together.
*/
protected double interClusterDistance(Cluster c1, Cluster c2)
{
if (areClustersInRelation(c1, c2))
{
return(Double.NegativeInfinity);
}
float totalSim = LogMath.LOG_ZERO;
float wordPairCount = (float)0.0;
HashSet <String> wordsSeen1 = new HashSet <String>();
LogMath logMath = LogMath.getLogMath();
foreach (Node node1 in c1.getElements())
{
String word1 = node1.getWord().getSpelling();
if (wordsSeen1.Contains(word1))
{
continue;
}
wordsSeen1.Add(word1);
HashSet <String> wordsSeen2 = new HashSet <String>();
foreach (Node node2 in c2.getElements())
{
String word2 = node2.getWord().getSpelling();
if (wordsSeen2.Contains(word2))
{
continue;
}
wordsSeen2.Add(word2);
float sim = (float)computePhoneticSimilarity(node1, node2);
sim = logMath.linearToLog(sim);
sim += (float)wordSubClusterProbability(c1, word1);
sim += (float)wordSubClusterProbability(c2, word2);
totalSim = logMath.addAsLinear(totalSim, sim);
wordPairCount++;
}
}
return(totalSim - logMath.logToLinear(wordPairCount));
}