public void Compute()
{
var clusterer = new NeighborJoiningClusterer <TSeq>((seq1, seq2) =>
{
var pairwiseAligner = new PairwiseAlignmentAlgorithm <TSeq, TItem>(_scorer, seq1, seq2, _itemsSelector);
pairwiseAligner.Compute();
return(1.0 - pairwiseAligner.GetAlignments().First().NormalizedScore);
});
IUndirectedGraph <Cluster <TSeq>, ClusterEdge <TSeq> > unrootedTree = clusterer.GenerateClusters(_sequences);
IBidirectionalGraph <Cluster <TSeq>, ClusterEdge <TSeq> > rootedTree = unrootedTree.ToRootedTree();
var profiles = new Dictionary <Cluster <TSeq>, Profile <TSeq, TItem> >();
var nodeStack = new Stack <Cluster <TSeq> >();
Cluster <TSeq> root = rootedTree.Roots().First();
double maxWeight = double.MinValue;
if (root.DataObjects.Count == 1)
{
profiles[root] = CreateProfile(root.DataObjects.First(), 0);
maxWeight = 0;
}
nodeStack.Push(root);
foreach (ClusterEdge <TSeq> edge in rootedTree.OutEdges(root))
{
maxWeight = Math.Max(maxWeight, CalcSequenceWeights(rootedTree, edge, 0, nodeStack, profiles));
}
foreach (Profile <TSeq, TItem> profile in profiles.Values)
{
profile.Weights[0] += 1.0 - maxWeight;
}
var scorer = new ProfileScorer <TSeq, TItem>(_scorer);
while (nodeStack.Count > 0)
{
Cluster <TSeq> node = nodeStack.Pop();
var curProfiles = new Stack <Profile <TSeq, TItem> >();
foreach (ClusterEdge <TSeq> childEdge in rootedTree.OutEdges(node))
{
curProfiles.Push(profiles[childEdge.Target]);
profiles.Remove(childEdge.Target);
}
if (node.DataObjects.Count == 1)
{
curProfiles.Push(profiles[node]);
profiles.Remove(node);
}
while (curProfiles.Count > 1)
{
Profile <TSeq, TItem> profile1 = curProfiles.Pop();
Profile <TSeq, TItem> profile2 = curProfiles.Pop();
var profileAligner = new PairwiseAlignmentAlgorithm <Profile <TSeq, TItem>, AlignmentCell <TItem>[]>(scorer, profile1, profile2, GetProfileItems);
profileAligner.Compute();
Alignment <Profile <TSeq, TItem>, AlignmentCell <TItem>[]> profileAlignment = profileAligner.GetAlignments().First();
var sequences = new List <Tuple <TSeq, AlignmentCell <TItem>, IEnumerable <AlignmentCell <TItem> >, AlignmentCell <TItem> > >();
for (int i = 0; i < profile1.Alignment.SequenceCount; i++)
{
int seq = i;
sequences.Add(Tuple.Create(profile1.Alignment.Sequences[i], profile1.Alignment.Prefixes[i], Enumerable.Range(0, profileAlignment.ColumnCount)
.Select(col => profileAlignment[0, col].IsNull ? new AlignmentCell <TItem>() : profileAlignment[0, col][0][seq]), profile1.Alignment.Suffixes[i]));
}
for (int j = 0; j < profile2.Alignment.SequenceCount; j++)
{
int seq = j;
sequences.Add(Tuple.Create(profile2.Alignment.Sequences[j], profile2.Alignment.Prefixes[j], Enumerable.Range(0, profileAlignment.ColumnCount)
.Select(col => profileAlignment[1, col].IsNull ? new AlignmentCell <TItem>() : profileAlignment[1, col][0][seq]), profile2.Alignment.Suffixes[j]));
}
var newAlignment = new Alignment <TSeq, TItem>(profileAlignment.RawScore, profileAlignment.NormalizedScore, sequences);
curProfiles.Push(new Profile <TSeq, TItem>(newAlignment, profile1.Weights.Concat(profile2.Weights)));
}
profiles[node] = curProfiles.Pop();
}
Alignment <TSeq, TItem> alignment = profiles[root].Alignment;
if (UseInputOrder)
{
var reorderedSequences = new List <Tuple <TSeq, AlignmentCell <TItem>, IEnumerable <AlignmentCell <TItem> >, AlignmentCell <TItem> > >();
foreach (TSeq sequence in _sequences)
{
for (int i = 0; i < alignment.SequenceCount; i++)
{
int seq = i;
if (sequence.Equals(alignment.Sequences[seq]))
{
reorderedSequences.Add(Tuple.Create(sequence, alignment.Prefixes[seq], Enumerable.Range(0, alignment.ColumnCount).Select(col => alignment[seq, col]), alignment.Suffixes[seq]));
break;
}
}
}
_result = new Alignment <TSeq, TItem>(alignment.RawScore, alignment.NormalizedScore, reorderedSequences);
}
else
{
_result = alignment;
}
}
public IBidirectionalGraph <HierarchicalGraphVertex, HierarchicalGraphEdge> GenerateHierarchicalGraph(HierarchicalGraphType graphType,
ClusteringMethod clusteringMethod, SimilarityMetric similarityMetric)
{
switch (clusteringMethod)
{
case ClusteringMethod.Upgma:
Func <Variety, Variety, double> upgmaGetDistance = null;
switch (similarityMetric)
{
case SimilarityMetric.Lexical:
upgmaGetDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].LexicalSimilarityScore;
break;
case SimilarityMetric.Phonetic:
upgmaGetDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].PhoneticSimilarityScore;
break;
}
var upgma = new UpgmaClusterer <Variety>(upgmaGetDistance);
IBidirectionalGraph <Cluster <Variety>, ClusterEdge <Variety> > upgmaTree = upgma.GenerateClusters(_projectService.Project.Varieties);
return(BuildHierarchicalGraph(upgmaTree));
case ClusteringMethod.NeighborJoining:
Func <Variety, Variety, double> njGetDistance = null;
switch (similarityMetric)
{
case SimilarityMetric.Lexical:
njGetDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].LexicalSimilarityScore;
break;
case SimilarityMetric.Phonetic:
njGetDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].PhoneticSimilarityScore;
break;
}
var nj = new NeighborJoiningClusterer <Variety>(njGetDistance);
IUndirectedGraph <Cluster <Variety>, ClusterEdge <Variety> > njTree = nj.GenerateClusters(_projectService.Project.Varieties);
switch (graphType)
{
case HierarchicalGraphType.Dendrogram:
IBidirectionalGraph <Cluster <Variety>, ClusterEdge <Variety> > rootedTree = njTree.ToRootedTree();
return(BuildHierarchicalGraph(rootedTree));
case HierarchicalGraphType.Tree:
return(BuildHierarchicalGraph(njTree));
}
break;
}
return(null);
}