public static IEnumerable <Cluster <Word> > GenerateCognateSets(this CogProject project, Meaning meaning)
{
var words = new HashSet <Word>();
var noise = new HashSet <Word>();
foreach (VarietyPair vp in project.VarietyPairs)
{
WordPair wp;
if (vp.WordPairs.TryGetValue(meaning, out wp))
{
if (wp.AreCognatePredicted)
{
words.Add(wp.Word1);
words.Add(wp.Word2);
noise.Remove(wp.Word1);
noise.Remove(wp.Word2);
}
else
{
if (!words.Contains(wp.Word1))
{
noise.Add(wp.Word1);
}
if (!words.Contains(wp.Word2))
{
noise.Add(wp.Word2);
}
}
}
}
double min = double.MaxValue, max = double.MinValue;
var distanceMatrix = new Dictionary <UnorderedTuple <Word, Word>, double>();
Word[] wordArray = words.ToArray();
for (int i = 0; i < wordArray.Length; i++)
{
for (int j = i + 1; j < wordArray.Length; j++)
{
Word w1 = wordArray[i];
Word w2 = wordArray[j];
double score = 0;
WordPair wp;
if (w1.Variety != w2.Variety && w1.Variety.VarietyPairs[w2.Variety].WordPairs.TryGetValue(meaning, out wp) && wp.AreCognatePredicted &&
wp.GetWord(w1.Variety) == w1 && wp.GetWord(w2.Variety) == w2)
{
score = wp.CognicityScore;
}
double distance = 1.0 - score;
min = Math.Min(min, distance);
max = Math.Max(max, distance);
distanceMatrix[UnorderedTuple.Create(w1, w2)] = distance;
}
}
var clusterer = new FlatUpgmaClusterer <Word>((w1, w2) => distanceMatrix[UnorderedTuple.Create(w1, w2)], (max + min) / 2);
return(clusterer.GenerateClusters(words).Concat(new Cluster <Word>(noise, true)));
}
public void Cluster()
{
var matrix = new[,]
{
{0.00, 0.50, 0.67, 0.80, 0.20},
{0.50, 0.00, 0.40, 0.70, 0.60},
{0.67, 0.40, 0.00, 0.80, 0.80},
{0.80, 0.70, 0.80, 0.00, 0.30},
{0.20, 0.60, 0.80, 0.30, 0.00}
};
var fupgma = new FlatUpgmaClusterer<char>((o1, o2) => matrix[o1 - 'A', o2 - 'A'], 0.5);
Cluster<char>[] clusters = fupgma.GenerateClusters(new[] {'A', 'B', 'C', 'D', 'E'}).ToArray();
var expected = new[]
{
new Cluster<char>('B', 'C'),
new Cluster<char>('A', 'E', 'D')
};
Assert.That(clusters, Is.EquivalentTo(expected).Using(new ClusterEqualityComparer<char>()));
}
public void Cluster()
{
var matrix = new[, ]
{
{ 0.00, 0.50, 0.67, 0.80, 0.20 },
{ 0.50, 0.00, 0.40, 0.70, 0.60 },
{ 0.67, 0.40, 0.00, 0.80, 0.80 },
{ 0.80, 0.70, 0.80, 0.00, 0.30 },
{ 0.20, 0.60, 0.80, 0.30, 0.00 }
};
var fupgma = new FlatUpgmaClusterer <char>((o1, o2) => matrix[o1 - 'A', o2 - 'A'], 0.5);
Cluster <char>[] clusters = fupgma.GenerateClusters(new[] { 'A', 'B', 'C', 'D', 'E' }).ToArray();
var expected = new[]
{
new Cluster <char>('B', 'C'),
new Cluster <char>('A', 'E', 'D')
};
Assert.That(clusters, Is.EquivalentTo(expected).Using(new ClusterEqualityComparer <char>()));
}
private void ClusterVarieties()
{
if (!_projectService.AreAllVarietiesCompared)
{
return;
}
Func <Variety, Variety, double> getDistance = null;
switch (_similarityMetric)
{
case SimilarityMetric.Lexical:
getDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].LexicalSimilarityScore;
break;
case SimilarityMetric.Phonetic:
getDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].PhoneticSimilarityScore;
break;
}
var clusterer = new FlatUpgmaClusterer <Variety>(getDistance, 1.0 - _similarityScoreThreshold);
_currentClusters.Clear();
_currentClusters.AddRange(clusterer.GenerateClusters(_varieties.Select(v => v.DomainVariety)).Where(c => c.DataObjects.Any(v => v.Regions.Count > 0)).OrderByDescending(c => c.DataObjects.Count));
foreach (GeographicalVarietyViewModel variety in _varieties)
{
if (variety.Regions.Count > 0)
{
int index = _currentClusters.FindIndex(c => c.DataObjects.Contains(variety.DomainVariety));
variety.ClusterIndex = index;
}
else
{
variety.ClusterIndex = -1;
}
}
}
/* Commented out for now, because the CommandLine library doesn't like finding two Usage attributes and we already have one on VerbBase
* [Usage(ApplicationAlias = "cog-cmdline")]
* public new static IEnumerable<Example> Examples
* {
* get
* {
* yield return new Example("UPGMA clustering (specify a threshhold value)", new ClusterVerb { Method = "upgma", Threshhold = 0.2 });
* yield return new Example("DBSCAN clustering (specify epsilon and min-words values)", new ClusterVerb { Method = "dbscan", Epsilon = 0.2, MinWords = 2 });
* yield return new Example("LSDBC clustering (specify alpha and K values)", new ClusterVerb { Method = "lsdbc", Alpha = 0.2, K = 3 });
* }
* }
*/
protected override ReturnCode DoWork(TextReader inputReader, TextWriter outputWriter, TextWriter errorWriter)
{
ReturnCode retcode = ReturnCode.Okay;
SetupProject();
string lowerMethod = Method.ToLowerInvariant();
switch (lowerMethod)
{
case "dbscan":
case "lsdbc":
case "upgma":
break;
default:
Errors.Add($"Invalid clustering method {Method}. Valid values are \"upgma\", \"dbscan\", and \"lsdbc\" (not case-sensitive, e.g. \"Upgma\" also works.)");
return(ReturnCode.InputError);
}
foreach (string line in ReadLines(inputReader))
{
// Format: word1 word2 score (where score is a floating-point number with 1.0 = 100% similarity)
string[] words = line.Split(' ');
if (words.Length < 3)
{
Errors.Add(line, "Each line should contain two words and one score, separated by spaces.");
continue;
}
double score;
if (!double.TryParse(words[2], NumberStyles.Float, CultureInfo.InvariantCulture, out score))
{
Errors.Add(line, $"Could not parse score \"{words[2]}\". Scores should be a number between 0 and 1.");
continue;
}
if (score < 0.0)
{
Errors.Add(line, $"Invalid score \"{words[2]}\". Scores should not be negative, but should be a number between 0 and 1.");
continue;
}
if (score > 1.0)
{
Errors.Add(line, $"Invalid score \"{words[2]}\". Scores should not be greater than 1, but should be a number between 0 and 1.");
continue;
}
double distance = 1.0 - score;
KeepScoreForUpgmaClusterer(words[0], words[1], distance); // TODO: Since we always call this, maybe we should rename it
}
IFlatClusterer <string> clusterer;
switch (Method.ToLowerInvariant())
{
case "dbscan":
// public DbscanClusterer(Func<T, IEnumerable<T>> getNeighbors, double minPoints)
KeepScoreForDbscanClusterer();
clusterer = new DbscanClusterer <string>(word => _distanceGraph[word]
.TakeWhile(scoreWordTuple => scoreWordTuple.Item1 <= Epsilon)
.Select(scoreWordTuple => scoreWordTuple.Item2), MinWords);
break;
case "lsdbc":
// public LsdbcClusterer(double alpha, Func<T, IEnumerable<Tuple<T, double>>> getKNearestNeighbors)
KeepScoreForLsdbcClusterer();
clusterer = new LsdbcClusterer <string>(Alpha, word => _distanceGraph[word].Take(K)
.Select(tuple => new Tuple <string, double>(tuple.Item2, tuple.Item1)));
break;
default:
clusterer = new FlatUpgmaClusterer <string>((w1, w2) =>
_distanceDict[new UnorderedTuple <string, string>(w1, w2)], Threshhold);
break;
}
IEnumerable <Cluster <string> > clusters = clusterer.GenerateClusters(_allWords);
PrintResults(outputWriter, clusters);
return(retcode);
}
private void ClusterVarieties()
{
if (!_projectService.AreAllVarietiesCompared)
return;
Func<Variety, Variety, double> getDistance = null;
switch (_similarityMetric)
{
case SimilarityMetric.Lexical:
getDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].LexicalSimilarityScore;
break;
case SimilarityMetric.Phonetic:
getDistance = (v1, v2) => 1.0 - v1.VarietyPairs[v2].PhoneticSimilarityScore;
break;
}
var clusterer = new FlatUpgmaClusterer<Variety>(getDistance, 1.0 - _similarityScoreThreshold);
_currentClusters.Clear();
_currentClusters.AddRange(clusterer.GenerateClusters(_varieties.Select(v => v.DomainVariety)).Where(c => c.DataObjects.Any(v => v.Regions.Count > 0)).OrderByDescending(c => c.DataObjects.Count));
foreach (GeographicalVarietyViewModel variety in _varieties)
{
if (variety.Regions.Count > 0)
{
int index = _currentClusters.FindIndex(c => c.DataObjects.Contains(variety.DomainVariety));
variety.ClusterIndex = index;
}
else
{
variety.ClusterIndex = -1;
}
}
}
public static IEnumerable<Cluster<Word>> GenerateCognateSets(this CogProject project, Meaning meaning)
{
var words = new HashSet<Word>();
var noise = new HashSet<Word>();
foreach (VarietyPair vp in project.VarietyPairs)
{
WordPair wp;
if (vp.WordPairs.TryGetValue(meaning, out wp))
{
if (wp.Cognacy)
{
words.Add(wp.Word1);
words.Add(wp.Word2);
noise.Remove(wp.Word1);
noise.Remove(wp.Word2);
}
else
{
if (!words.Contains(wp.Word1))
noise.Add(wp.Word1);
if (!words.Contains(wp.Word2))
noise.Add(wp.Word2);
}
}
}
double min = double.MaxValue;
var distanceMatrix = new Dictionary<UnorderedTuple<Word, Word>, double>();
Word[] wordArray = words.ToArray();
for (int i = 0; i < wordArray.Length; i++)
{
for (int j = i + 1; j < wordArray.Length; j++)
{
Word w1 = wordArray[i];
Word w2 = wordArray[j];
double score = 0;
WordPair wp;
if (w1.Variety != w2.Variety && w1.Variety.VarietyPairs[w2.Variety].WordPairs.TryGetValue(meaning, out wp) && wp.Cognacy
&& wp.GetWord(w1.Variety) == w1 && wp.GetWord(w2.Variety) == w2)
{
score = wp.PredictedCognacyScore;
}
double distance = 1.0 - score;
min = Math.Min(min, distance);
distanceMatrix[UnorderedTuple.Create(w1, w2)] = distance;
}
}
var clusterer = new FlatUpgmaClusterer<Word>((w1, w2) => distanceMatrix[UnorderedTuple.Create(w1, w2)], (1.0 + min) / 2);
return clusterer.GenerateClusters(words).Concat(new Cluster<Word>(noise, true));
}
