public Cluster(Node parent)
{
this.clusterPair = null;
this.items = new List<Vector>();
this.mean = null;
this.parent = parent;
}
// The special constructor is used to deserialize values.
public Sample(SerializationInfo info, StreamingContext context)
{
x = (Vector)info.GetValue("x", typeof(Vector));
y = (Vector)info.GetValue("y", typeof(Vector));
label = (double)info.GetValue("label", typeof(double));
id = (int)info.GetValue("id", typeof(int));
}
public Sample(double[] input, double[] output, double label, int id)
{
this.x = new Vector(input);
this.y = new Vector(output);
this.id = id;
this.label = label;
}
// The special constructor is used to deserialize values.
public Cluster(SerializationInfo info, StreamingContext context)
{
clusterPair = (ClusterPair)info.GetValue("clusterPair", typeof(ClusterPair));
items = (List<Vector>)info.GetValue("items", typeof(List<Vector>));
mean = (Vector)info.GetValue("mean", typeof(Vector));
meanMDF = (ILArray<double>)info.GetValue("meanMDF", typeof(ILArray<double>));
parent = (Node)info.GetValue("parent", typeof(Node));
}
public void GetDistance_GetCorrectDistance()
{
Vector v1 = new Vector(new double[] { 1, 3, 4 });
Vector v2 = new Vector(new double[] { 7, 2, 3 });
double distance = Math.Round(v1.GetDistance(v2), 3);
Assert.AreEqual(distance, 6.164);
}
public void Equals_VectorNotEqualsToVector()
{
Vector vector1 = new Vector(new double[] { 1.0, 2.0, 0, 3.0 });
Vector vector2 = new Vector(new double[] { 1.0, 0.0, 0, 3.0 });
bool equals = vector1.EqualsToVector(vector2);
Assert.IsFalse(equals);
}
public void GetMDFDistance_GetCorrectDistance()
{
Vector v = new Vector(new double[] { 1, 3, 4 });
v.ValuesMDF = new double[] { 1, 3, 4 };
ILArray<double> a2 = new double[] { 7, 2, 3 };
double distance = Math.Round(v.GetMDFDistance(a2), 3);
Assert.AreEqual(distance, 6.164);
}
public void AddItem(Vector vector, double label)
{
Vector newItem = new Vector(vector.Values.ToArray());
newItem.Label = label;
newItem.Id = this.items.Count + 1;
this.items.Add(newItem);
// update mean
this.UpdateMean(newItem);
}
public void Add_AddToVectorCorrectVector()
{
Vector vector = new Vector(new double[] { 1.0, 2.0, 3.0});
Vector vector2 = new Vector(new double[] { 2.0, 3.0, 4.0 });
vector.Add(vector2);
Assert.AreEqual(vector.Values[0], 3.0);
Assert.AreEqual(vector.Values[1], 5.0);
Assert.AreEqual(vector.Values[2], 7.0);
}
public void GetIdOfClosestVector_GetCorrectId()
{
List<Vector> vectors = new List<Vector>();
vectors.Add(new Vector(new double[] { 1.0, 2.0, 3.0 }, 1.0, 0));
vectors.Add(new Vector(new double[] { 2.0, 3.0, 4.0 }, 1.0, 1));
vectors.Add(new Vector(new double[] { 2.0, 2.0, 2.0 }, 1.0, 2));
vectors.Add(new Vector(new double[] { 7.0, 8.0, 6.0 }, 1.0, 3));
Vector vector = new Vector(new double[] { 2.0, 3.0, 5.0 });
int result = vector.GetIdOfClosestVector(vectors);
Assert.AreEqual(result, 1);
}
public void GetMeanOfDataWithLabel_ReturnCorrectMean()
{
Params.inputDataDimension = 3;
Samples samples = new Samples();
samples.Items.Add(new Sample(new double[] { 1, 2, 3 }, 1, 0));
samples.Items.Add(new Sample(new double[] { 2, 3, 4 }, 1, 0));
samples.Items.Add(new Sample(new double[] { 1, 2, 3 }, 2, 0));
samples.Items.Add(new Sample(new double[] { 2, 3, 4 }, 2, 0));
samples.Items.Add(new Sample(new double[] { 5, 6, 7 }, 2, 0));
samples.Items.Add(new Sample(new double[] { 5, 6, 7 }, 3, 0));
Vector result1 = samples.GetMeanOfDataWithLabel(1.0);
Vector result2 = new Vector(new double[] { 1.5, 2.5, 3.5 });
Assert.AreEqual(result1.Values[0], result1.Values[0]);
Assert.AreEqual(result1.Values[1], result1.Values[1]);
Assert.AreEqual(result1.Values[2], result1.Values[2]);
}
public bool EqualsToVector(Vector vector)
{
if (this.values.Length != vector.values.Length) throw new InvalidOperationException("Not the same count of attributes");
for (int i = 0; i < this.values.Length; i++)
{
if (this.values[i] != vector.values[i]) return false;
}
return true;
}
public double GetDistance(Vector vector)
{
if (this.values.Count() != vector.values.Count()) throw new InvalidOperationException("Not the same count of attributes");
ILArray<double> result = this.values - vector.values;
result = result * result;
return Math.Sqrt(result.Sum());
}
public static Vector GetVarianceOfVectors(List<Vector> vectors, Vector mean)
{
Vector result = new Vector(vectors[0].values.Length, vectors[0].valuesMDF.Length, 0.0);
foreach (var item in vectors)
{
Vector diff = new Vector(item.valuesMDF.ToArray());
diff.SubtractMDF(mean);
diff.MultiplyMDF(diff);
result.AddMDF(diff);
}
result.DivideMDF(vectors.Count - 1);
return result;
}
public void Add(Vector vector)
{
if (this.values.Length != vector.values.Length) throw new InvalidOperationException("Not the same count of attributes");
this.values = this.values + vector.values;
}
/// <summary>
/// Update mean with amnesic average
/// </summary>
/// <param name="vector">Vector, which should be add to mean</param>
protected void UpdateMean(Vector vector)
{
#warning this must be remade according to F. Amnesic average with parameters t1, t2
double t = (double)this.items.Count;
//double multiplier1 = (t - 1) / t;
//double multiplier2 = 1 / t;
double multiplier1 = (t - 1 - this.GetAmnesicParameter(t)) / t;
double multiplier2 = (1 + this.GetAmnesicParameter(t)) / t;
this.mean.Multiply(multiplier1);
Vector incrementPart = new Vector(vector.Values.ToArray());
incrementPart.Multiply(multiplier2);
this.mean.Add(incrementPart);
}
public static Vector GetMeanOfVectorsMDF(List<Vector> vectors)
{
Vector result = new Vector(vectors[0].values.Length, vectors[0].valuesMDF.Length, 0.0);
foreach (Vector item in vectors)
{
result.AddMDF(item);
}
result.DivideMDF((double)vectors.Count);
return result;
}
private double GetLabelOfClosestY(Vector YMean)
{
double minDistance = double.MaxValue;
double resultLabel = 0;
foreach (var item in this.labelOutputVectors)
{
double distance = item.Value.GetDistance(YMean);
if (distance < minDistance)
{
minDistance = distance;
resultLabel = item.Key;
}
}
return resultLabel;
}
public void SubtractMDF(Vector vector)
{
if (this.valuesMDF.Length != vector.valuesMDF.Length) throw new InvalidOperationException("Not the same count of attributes");
this.valuesMDF = this.valuesMDF - vector.valuesMDF;
}
public void UpdateMeanMdf(Vector newItem)
{
ILArray<double> newsample = newItem.ValuesMDF.ToArray();
this.meanMDF = (this.meanMDF * ((double)(this.countOfSamples - 1) / (double)this.countOfSamples)) + ((1 / (double)this.countOfSamples) * newsample);
}
public Vector GetOutputVectorFromBmp(Bitmap bmp)
{
if(Params.outputDataDimension != bmp.Width * bmp.Height)
{
throw new InvalidOperationException("Unable to create output from bmp of another dimensionality.");
}
List<double> attributes = new List<double>();
for (int i = 0; i < bmp.Width; i++)
{
for (int j = 0; j < bmp.Height; j++)
{
Color color = bmp.GetPixel(j,i);
attributes.Add((int)((color.R * .3) + (color.G * .59) + (color.B * .11)));
}
}
Vector output = new Vector(attributes.ToArray());
return output;
}
private void EvaluateSwap()
{
if (this.clusterPairs.Count < Params.q)
{
throw new InvalidOperationException("Small count of microclusters, impossible to swap");
}
// random selection is deleted, because of unification of results
// select q random samples
//List<Vector> centres = new List<Vector>();
//Random random = new Random();
//List<int> randoms = new List<int>();
//for (int i = 0; i < Params.q; i++)
//{
// bool randomIsNotUnique = true;
// int r = 0;
// while (randomIsNotUnique)
// {
// r = random.Next(this.clusterPairs.Count);
// randomIsNotUnique = randoms.Contains(r);
// }
// //Console.WriteLine("Center like cluster: " + r.ToString());
// randoms.Add(r);
// Vector vector = new Vector(this.clusterPairs[r].X.Mean.Values.ToArray());
// vector.Id = i;
// centres.Add(vector);
//}
#region instead of random
List<Vector> centres = new List<Vector>();
for (int i = 0; i < Params.q; i++)
{
Vector vector = new Vector(this.clusterPairs[i].X.Mean.Values.ToArray());
vector.Id = i;
centres.Add(vector);
}
#endregion
// evaluate to which center vector belongs
foreach (ClusterPair clPair in clusterPairs)
{
clPair.CurrentToPrev();
clPair.CurrentCenter = clPair.X.Mean.GetIdOfClosestVector(centres);
}
// while assignment is not changed, do
bool next = true;
int whil = 1;
while (next)
{
//Console.WriteLine("While cycle " + whil.ToString());
for (int i = 0; i < centres.Count; i++)
{
//select all vectors assignet to center
List<Vector> assignedVectors = clusterPairs.Where(cp => cp.CurrentCenter == centres[i].Id).Select(cp => cp.X.Mean).ToList<Vector>();
// update center
if (assignedVectors.Count > 0)
{
Vector v = new Vector(Params.inputDataDimension, 0.0);
double dis1 = centres[i].GetDistance(v);
centres[i] = Vector.GetMeanOfVectors(assignedVectors);
double dis2 = centres[i].GetDistance(v);
//Console.WriteLine("Center " + i.ToString() + " Distance 1: " + dis1.ToString() + " Distance 2: " + dis2.ToString());
}
centres[i].Id = i;
}
// update assignments
// evaluate to which center vector belongs
foreach (ClusterPair clPair in clusterPairs)
{
clPair.CurrentToPrev();
clPair.CurrentCenter = clPair.X.Mean.GetIdOfClosestVector(centres);
//Console.WriteLine("Change cl. " + clPair.Id.ToString() + " from " + clPair.PreviousCenter.ToString() + " to " + clPair.CurrentCenter.ToString());
}
next = this.AssigmentIsChanged();
whil++;
}
}
public void Substract_RestulVectorIsCorrect()
{
Vector vector = new Vector(new double[] { 1.0, 2.0, 3.0 });
Vector vector2 = new Vector(new double[] { 2.0, 4.0, 6.0 });
vector.Subtract(vector2);
Assert.AreEqual(vector.Values[0], -1.0);
Assert.AreEqual(vector.Values[1], -2.0);
Assert.AreEqual(vector.Values[2], -3.0);
}
/// <summary>
/// Update MDF mean with amnesic average
/// </summary>
/// <param name="vector">Vector, which should be add to mean</param>
protected void UpdateMeanMDF(ILArray<double> vector)
{
double t = (double)this.items.Count;
//double multiplier1 = (t - 1) / t;
//double multiplier2 = 1 / t;
double multiplier1 = (t - 1 - this.GetAmnesicParameter(t)) / t;
double multiplier2 = (1 + this.GetAmnesicParameter(t)) / t;
Vector incrementPart = new Vector(vector.ToArray());
this.meanMDF = (this.meanMDF * multiplier1) + (vector * multiplier2);
}
public void Multiply(Vector multiplier)
{
this.values = this.values * multiplier.values;
}
public void AddItemNonLeaf(Vector newItem)
{
newItem.Id = this.items.Count + 1;
this.items.Add(newItem);
// update mean
this.UpdateMean(newItem);
//node.CountGSOManifold();
this.UpdateMeanAndCovMatrixMDF(newItem.ValuesMDF.ToArray());
}
public void MultiplyMDF(Vector multiplier)
{
this.valuesMDF = this.valuesMDF * multiplier.valuesMDF;
}
public static int GetIdOfClosestVector(Vector vector, List<Vector> vectors)
{
int result = 0;
double minDistance = double.MaxValue;
foreach (Vector item in vectors)
{
double distance = vector.GetDistance(item);
if (distance < minDistance)
{
result = item.Id;
minDistance = distance;
}
}
return result;
}
public Vector GetOutputFromKnownSamples(Sample sample)
{
if (!this.outputs.ContainsKey(sample.Label))
{
this.outputs[sample.Label] = new MappedValue(){ Mean = new Vector(sample.X.Values.ToArray()), Count = 1};
}
else
{
this.outputs[sample.Label].Count++;
int count = this.outputs[sample.Label].Count;
this.outputs[sample.Label].Mean.Multiply(((double)count-1.0)/(double)count);
Vector addPart = new Vector(sample.X.Values.ToArray());
addPart.Multiply(1 / (double)count);
this.outputs[sample.Label].Mean.Add(addPart);
}
return new Vector(this.outputs[sample.Label].Mean.Values.ToArray());
}
public void AddItemWithoutUpdatingStats(Vector vector)
{
this.items.Add(vector);
}
