/// <summary>
/// Evaluates all given <see cref="ClusterSet{TInstance}" />s according to the given
/// <see cref="IInternalEvaluationCriterion{TInstance}" />.
/// </summary>
/// <param name="clustering">The cluster-sets to be evaluated.</param>
/// <param name="criterion">The criterion used to evaluate the cluster-sets.</param>
/// <param name="maxClusters">The maximum number of clusters allowed for a cluster-set for it to be evaluated.</param>
/// <returns>A list containing the evaluations for each cluster-set.</returns>
/// <typeparam name="TInstance">The type of instance considered.</typeparam>
public static ICollection <ClusterSetEvaluation <TInstance> > EvaluateClustering <TInstance>(
this ClusteringResult <TInstance> clustering, IInternalEvaluationCriterion <TInstance> criterion,
uint maxClusters) where TInstance : IComparable <TInstance>
{
var evals = new List <ClusterSetEvaluation <TInstance> >();
// checks only one cluster allowed
if (maxClusters == 1)
{
evals.Add(new ClusterSetEvaluation <TInstance>(clustering[0], double.NaN));
return(evals);
}
// evaluates all cluster-sets
foreach (var clusterSet in clustering.Reverse())
{
if (clusterSet.Count < 2 || clusterSet.Count > maxClusters)
{
continue;
}
var eval = criterion.Evaluate(clusterSet);
evals.Add(new ClusterSetEvaluation <TInstance>(clusterSet, eval));
}
return(evals);
}
/// <inheritdoc />
public ClusteringResult <TInstance> GetClustering(IEnumerable <Cluster <TInstance> > clusters,
double dissimilarity = 0d)
{
// initializes elements
var currentClusters = clusters.ToArray();
this._clusters = new Cluster <TInstance> [currentClusters.Length * 2 - 1];
this._dissimilarities = new double[currentClusters.Length * 2 - 1][];
this._curClusterCount = 0;
// calculates initial dissimilarities
foreach (var cluster in currentClusters)
{
this._clusters[this._curClusterCount] = cluster;
this.UpdateDissimilarities();
this._curClusterCount++;
}
var clustering = new ClusteringResult <TInstance>(currentClusters.Length)
{
[0] = new ClusterSet <TInstance>(currentClusters, dissimilarity)
};
var numSteps = currentClusters.Length;
for (var i = 1; i < numSteps; i++)
{
// gets minimal dissimilarity between a pair of existing clusters
var minDissimilarity = this.GetMinDissimilarity(out var clusterIdx1, out var clusterIdx2);
// gets a copy of previous clusters, removes new cluster elements
var cluster1 = this._clusters[clusterIdx1];
var cluster2 = this._clusters[clusterIdx2];
var newClusters = new Cluster <TInstance> [currentClusters.Length - 1];
var idx = 0;
foreach (var cluster in currentClusters)
{
if (!cluster.Equals(cluster1) && !cluster.Equals(cluster2))
{
newClusters[idx++] = cluster;
}
}
this._clusters[clusterIdx1] = null;
this._clusters[clusterIdx2] = null;
// creates a new cluster from the union of closest clusters (save reference to parents)
var newCluster = new Cluster <TInstance>(cluster1, cluster2, minDissimilarity);
// adds cluster to list and calculates distance to all others
newClusters[idx] = this._clusters[this._curClusterCount] = newCluster;
this.UpdateDissimilarities();
this._curClusterCount++;
// updates global list of clusters
currentClusters = newClusters;
clustering[i] = new ClusterSet <TInstance>(currentClusters, minDissimilarity);
}
this._clusters = null;
this._dissimilarities = null;
return(clustering);
}
/// <summary>
/// Evaluates all given <see cref="ClusterSet{TInstance}" />s according to the given
/// <see cref="IInternalEvaluationCriterion{TInstance}" />. The maximum number of clusters allowed in a cluster-set for
/// it to be evaluated corresponds to sqrt(N/2), where N is the total number of instances clustered.
/// </summary>
/// <param name="clustering">The cluster-sets to be evaluated.</param>
/// <param name="criterion">The criterion used to evaluate the cluster-sets.</param>
/// <returns>A list containing the evaluations for each cluster-set.</returns>
/// <typeparam name="TInstance">The type of instance considered.</typeparam>
public static ICollection <ClusterSetEvaluation <TInstance> > EvaluateClustering <TInstance>(
this ClusteringResult <TInstance> clustering, IInternalEvaluationCriterion <TInstance> criterion)
where TInstance : IComparable <TInstance> =>
EvaluateClustering(clustering, criterion, (uint)Math.Sqrt(clustering.Count / 2d));
/// <summary>
/// Evaluates all given <see cref="ClusterSet{TInstance}" />s according to the given
/// <see cref="IExternalEvaluationCriterion{TInstance,TClass}" />. The maximum number of clusters allowed in a
/// cluster-set for it to be evaluated corresponds to sqrt(N/2), where N is the total number of instances clustered.
/// </summary>
/// <param name="clustering">The cluster-sets to be evaluated.</param>
/// <param name="criterion">The criterion used to evaluate the cluster-sets.</param>
/// <param name="instanceClasses">The instances' classes.</param>
/// <returns>A list containing the evaluations for each cluster-set.</returns>
/// <typeparam name="TInstance">The type of instance considered.</typeparam>
/// <typeparam name="TClass">The type of class considered.</typeparam>
public static ICollection <ClusterSetEvaluation <TInstance> > EvaluateClustering <TInstance, TClass>(
this ClusteringResult <TInstance> clustering, IExternalEvaluationCriterion <TInstance, TClass> criterion,
IDictionary <TInstance, TClass> instanceClasses) where TInstance : IComparable <TInstance> =>
EvaluateClustering(clustering, criterion, instanceClasses, (uint)Math.Sqrt(clustering.Count / 2d));
