/**
* If there are unclustered documents, appends the "Other Topics" group to the <code>clusters
* </code>.
*
* @see #buildOtherTopics(List, List, String)
*/
public static void appendOtherTopics(
List<Document> allDocuments, List<Cluster> clusters, String label) {
final Cluster otherTopics = buildOtherTopics(allDocuments, clusters, label);
if (!otherTopics.getDocuments().isEmpty()) {
clusters.add(otherTopics);
}
}
/**
* Assigns sequential identifiers to the provided <code>clusters</code> (and their sub-clusters).
* If a cluster already has an identifier, the identifier will not be changed.
*
* @param clusters Clusters to assign identifiers to.
* @throws IllegalArgumentException if the provided clusters contain non-unique identifiers
*/
public static void assignClusterIds(Collection<Cluster> clusters) {
final ArrayList<Cluster> flattened = Lists.newArrayListWithExpectedSize(clusters.size());
flatten(flattened, clusters);
synchronized (clusters) {
final HashSet<Integer> ids = Sets.newHashSet();
// First, find the start value for the id and check uniqueness of the ids
// already provided.
int maxId = Integer.MIN_VALUE;
for (final Cluster cluster : flattened) {
if (cluster.id != null) {
if (!ids.add(cluster.id)) {
throw new IllegalArgumentException("Non-unique cluster id found: " + cluster.id);
}
maxId = Math.max(maxId, cluster.id);
}
}
// We'd rather start with 0
maxId = Math.max(maxId, -1);
// Assign missing ids
for (final Cluster c : flattened) {
if (c.id == null) {
c.id = ++maxId;
}
}
}
}
/*
* Recursive descent into subclusters.
*/
private static void flatten(ArrayList<Cluster> flattened, Collection<Cluster> clusters) {
for (Cluster c : clusters) {
flattened.add(c);
final List<Cluster> subclusters = c.getSubclusters();
if (!subclusters.isEmpty()) {
flatten(flattened, subclusters);
}
}
}
/** A recursive routine for collecting unique documents from this cluster and subclusters. */
private static Set<Document> collectAllDocuments(Cluster cluster, Set<Document> docs) {
if (cluster == null) {
return docs;
}
docs.addAll(cluster.getDocuments());
final List<Cluster> subclusters = cluster.getSubclusters();
for (final Cluster subcluster : subclusters) {
collectAllDocuments(subcluster, docs);
}
return docs;
}
/**
* Builds an "Other Topics" cluster that groups those documents from <code>allDocument</code> that
* were not referenced in any cluster in <code>clusters</code>.
*
* @param allDocuments all documents to check against
* @param clusters list of clusters with assigned documents
* @param label label for the "Other Topics" group
* @return the "Other Topics" cluster
*/
public static Cluster buildOtherTopics(
List<Document> allDocuments, List<Cluster> clusters, String label) {
final Set<Document> unclusteredDocuments = Sets.newLinkedHashSet(allDocuments);
final Set<Document> assignedDocuments = Sets.newHashSet();
for (Cluster cluster : clusters) {
collectAllDocuments(cluster, assignedDocuments);
}
unclusteredDocuments.removeAll(assignedDocuments);
final Cluster otherTopics = new Cluster(label);
otherTopics.addDocuments(unclusteredDocuments);
otherTopics.setOtherTopics(true);
return otherTopics;
}
/**
* Locate the first cluster that has id equal to <code>id</code>. The search includes all the
* clusters in the input and their sub-clusters. The first cluster with matching identifier is
* returned or <code>null</code> if no such cluster could be found.
*/
public static Cluster find(int id, Collection<Cluster> clusters) {
for (Cluster c : clusters) {
if (c != null) {
if (c.id != null && c.id == id) {
return c;
}
if (!c.getSubclusters().isEmpty()) {
final Cluster sub = find(id, c.getSubclusters());
if (sub != null) {
return sub;
}
}
}
}
return null;
}
public Double apply(Cluster cluster) {
return cluster.isOtherTopics() ? 1.0 : -1.0;
}
public String apply(Cluster cluster) {
return cluster.getLabel();
}
public Double apply(Cluster cluster) {
return cluster.getAttribute(SCORE);
}
public Integer apply(Cluster cluster) {
return cluster.size();
}
