public void KMeansClusteringWorksOnTwoDimensions()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 2);
var p2 = new IdentifiableDataPoint(1, 2);
var p3 = new IdentifiableDataPoint(2, 2);
var p4 = new IdentifiableDataPoint(3, 2);
p1.AddAttribute("Gender", 1);
p1.AddAttribute("Income", 1);
p2.AddAttribute("Gender", 0);
p2.AddAttribute("Income", 0.1429);
p3.AddAttribute("Gender", 1);
p3.AddAttribute("Income", 0.2858);
p4.AddAttribute("Gender", 1);
p4.AddAttribute("Income", 1);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
dataSet.AddItem(p4);
var kmeans2 = new KMeans(dataSet, new [] { 0, 1, 2 }, new EuclideanMetric());
var result2 = kmeans2.Calculate();
Assert.AreEqual(2, result2.Clusters[0].Members.Count);
Assert.AreEqual(1, result2.Clusters[1].Members.Count);
Assert.AreEqual(1, result2.Clusters[2].Members.Count);
}
public void AddAttributeShouldCheckUpperBound()
{
var datapoint = new IdentifiableDataPoint(0, 1);
datapoint.AddAttribute("Age", 27);
Assert.Throws <NumberOfDimensionsExceededException>(() => datapoint.AddAttribute("Status", 1));
}
public void IndexerShouldReturnCorrectValue()
{
var datapoint = new IdentifiableDataPoint(0, 5);
datapoint.AddAttribute("Attr1", 42);
datapoint.AddAttribute("Attr2", 24);
datapoint.AddAttribute("Attr3", 25);
Assert.AreEqual(42, datapoint["Attr1"]);
Assert.AreEqual(24, datapoint["Attr2"]);
Assert.AreEqual(25, datapoint["Attr3"]);
}
public void AttributesWithTheSameNameShouldBeAllowed()
{
var datapoint = new IdentifiableDataPoint(0, 5);
datapoint.AddAttribute("Attr1", 41);
datapoint.AddAttribute("Attr1", 42);
Assert.AreEqual("Attr1", datapoint.Attributes[0]);
Assert.AreEqual("Attr1", datapoint.Attributes[1]);
Assert.AreEqual(41, datapoint[0]);
Assert.AreEqual(42, datapoint[1]);
}
public void ShouldSaveAttributesInOrder()
{
var datapoint = new IdentifiableDataPoint(0, 3);
datapoint.AddAttribute("Attr1", 27);
datapoint.AddAttribute("Attr2", 23);
datapoint.AddAttribute("Attr3", 24);
Assert.AreEqual("Attr1", datapoint.Attributes[0]);
Assert.AreEqual("Attr2", datapoint.Attributes[1]);
Assert.AreEqual("Attr3", datapoint.Attributes[2]);
}
public void KMeansCentroidsArePlacedRightInFiveDimensions()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 5);
var p2 = new IdentifiableDataPoint(1, 5);
var p3 = new IdentifiableDataPoint(2, 5);
var p4 = new IdentifiableDataPoint(3, 5);
p1.AddAttribute("Gender", 1);
p1.AddAttribute("Income", 1);
p1.AddAttribute("Age", 0.16);
p1.AddAttribute("Purchase", 0.5);
p1.AddAttribute("Control", 1);
p2.AddAttribute("Gender", 0);
p2.AddAttribute("Income", 0.1429);
p2.AddAttribute("Age", 0.16);
p2.AddAttribute("Purchase", 1);
p2.AddAttribute("Control", 0);
p3.AddAttribute("Gender", 1);
p3.AddAttribute("Income", 0.2858);
p3.AddAttribute("Age", 0.16);
p3.AddAttribute("Purchase", 1);
p3.AddAttribute("Control", 1);
p4.AddAttribute("Gender", 1);
p4.AddAttribute("Income", 1);
p4.AddAttribute("Age", 0.16);
p4.AddAttribute("Purchase", 1);
p4.AddAttribute("Control", 0.5);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
dataSet.AddItem(p4);
var kmeans2 = new KMeans(dataSet, new [] { 0, 1, 2 }, new EuclideanMetric());
var result2 = kmeans2.Calculate();
double[] _centroid1 = { 1, 0.2858, 0.16, 1, 1 };
double[] _centroid2 = { 1, 1, 0.16, 0.75, 0.75 };
double[] _centroid3 = { 0, 0.1429, 0.16, 1, 0 };
Assert.AreEqual(_centroid2, result2.Clusters[0].Centroid.Coordinates);
Assert.AreEqual(_centroid3, result2.Clusters[1].Centroid.Coordinates);
Assert.AreEqual(_centroid1, result2.Clusters[2].Centroid.Coordinates);
}
private void ParseMultipleBinaryField(IField field, CsvReader csv, IdentifiableDataPoint profile)
{
string label = csv.GetField <string>(field.Index);
if (RemoveWhiteSpace)
{
label = label.Trim();
}
int fieldCount = field.Values.Count();
foreach (IFieldValue possibleFieldValue in field.Values)
{
double value = 0;
string originalValue = "No";
if (possibleFieldValue.Name.Equals(label))
{
value = field.Weight / fieldCount;
originalValue = "Yes";
}
string name = String.Format("{0}: {1}", field.Category, possibleFieldValue.Name);
profile.AddAttribute(name, value, originalValue);
}
}
private void ParseMultipleChoiceBinaryField(IField field, CsvReader csv, IdentifiableDataPoint profile)
{
string label = csv.GetField <string>(field.Index);
if (RemoveWhiteSpace)
{
label = label.Trim();
}
string[] array = label.Split(ValueDelimiter).Select(l => l.Trim()).ToArray();
int fieldCount = field.Values.Count();
foreach (IFieldValue possibleFieldValue in field.Values)
{
double value = 0;
string originalValue = "No";
if (array.Contains(possibleFieldValue.Name))
{
value = field.Weight / fieldCount;
originalValue = "Yes";
}
string name = String.Format("{0}: {1}", field.Category, possibleFieldValue.Name);
profile.AddAttribute(name, value, originalValue);
}
}
public void KMeansShouldCalculateCorrectCentroidsInOneDimension()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 1);
var p2 = new IdentifiableDataPoint(1, 1);
var p3 = new IdentifiableDataPoint(2, 1);
var p4 = new IdentifiableDataPoint(3, 1);
p1.AddAttribute("Gender", 1);
p2.AddAttribute("Gender", 0);
p3.AddAttribute("Gender", 1);
p4.AddAttribute("Gender", 1);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
dataSet.AddItem(p4);
var kmeans2 = new KMeans(dataSet, new [] { 0, 1 }, new EuclideanMetric());
var result2 = kmeans2.Calculate();
double[] centroid1 = { 1 };
double[] centroid2 = { 0 };
Assert.AreEqual(centroid1, result2.Clusters[0].Centroid.Coordinates);
Assert.AreEqual(centroid2, result2.Clusters[1].Centroid.Coordinates);
}
public void KMeansShouldClusterDataPointsInOneDimensionCorrectly()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 1);
var p2 = new IdentifiableDataPoint(1, 1);
var p3 = new IdentifiableDataPoint(2, 1);
var p4 = new IdentifiableDataPoint(3, 1);
p1.AddAttribute("Gender", 1);
p2.AddAttribute("Gender", 0);
p3.AddAttribute("Gender", 1);
p4.AddAttribute("Gender", 1);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
dataSet.AddItem(p4);
var kmeans2 = new KMeans(dataSet, new [] { 0, 1 }, new EuclideanMetric());
var result2 = kmeans2.Calculate();
var cluster1Members = result2.Clusters[0].Members.Select(e => e.Member).ToArray();
var cluster2Members = result2.Clusters[1].Members.Select(e => e.Member).ToArray();
Assert.AreEqual(3, cluster1Members.Length);
Assert.Contains(p1, cluster1Members);
Assert.Contains(p3, cluster1Members);
Assert.Contains(p4, cluster1Members);
Assert.AreEqual(1, cluster2Members.Length);
Assert.Contains(p2, cluster2Members);
}
private void ParseScalarField(IField field, CsvReader csv, IdentifiableDataPoint profile)
{
string label = csv.GetField <string>(field.Index);
double?translatedField = field.Values.GetDoubleValueFor(label);
if (!translatedField.HasValue)
{
throw new InvalidFieldValueException(csv.Row, field.Index);
}
double value = translatedField.Value * field.Weight;
profile.AddAttribute(field.Category, value, label);
}
public void CentroidsHaveDistinctValues()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 2);
var p2 = new IdentifiableDataPoint(1, 2);
var p3 = new IdentifiableDataPoint(2, 2);
p1.AddAttribute("Gender", 1);
p1.AddAttribute("Income", 1);
p2.AddAttribute("Gender", 1);
p2.AddAttribute("Income", 1);
p3.AddAttribute("Gender", 0);
p3.AddAttribute("Income", 0);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
Assert.Throws <InvalidOperationException>(() => new KMeans(dataSet, new [] { 0, 1 }, new EuclideanMetric()));
}
private void ParseNumericField(IField field, CsvReader csv, IdentifiableDataPoint dataItem)
{
double valueInDataField;
string val = csv.GetField(field.Index);
if (!double.TryParse(val, NumberStyles.Any, parseCulture, out valueInDataField))
{
throw new InvalidNumericValueException(csv.Row, field.Index);
}
string originalValue = valueInDataField.ToString();
double difference = field.MaxValue - field.MinValue;
double normalizedValue = (valueInDataField - field.MinValue) / difference;
double finalValue = normalizedValue * field.Weight;
dataItem.AddAttribute(field.Category, finalValue, originalValue);
}
public void CalculateWithZeroClusters()
{
var dataSet = new IdentifiableDataPointCollection();
var p1 = new IdentifiableDataPoint(0, 1);
var p2 = new IdentifiableDataPoint(1, 1);
var p3 = new IdentifiableDataPoint(2, 1);
var p4 = new IdentifiableDataPoint(3, 1);
p1.AddAttribute("Gender", 1);
p2.AddAttribute("Gender", 0);
p3.AddAttribute("Gender", 1);
p4.AddAttribute("Gender", 1);
dataSet.AddItem(p1);
dataSet.AddItem(p2);
dataSet.AddItem(p3);
dataSet.AddItem(p4);
Assert.Throws <ArgumentException>(() => new KMeans(dataSet, 0, new EuclideanMetric()));
}