/// <summary>
/// Overload operator + for DPoint
/// </summary>
/// <param name="d1">Vector d1</param>
/// <param name="d2">vector d2</param>
/// <returns>sum of d1 & d2</returns>
public static DPoint operator +(DPoint d1, DPoint d2)
{
int d =0 ;
d = d1.MyNode.Count;
DPoint dp = new DPoint();
dp.MyNode = new List<Node>();
for (int i =0; i<d; ++i)
{
dp.MyNode.Add(new Node(d1.MyNode[i].Index, d1.MyNode[i].Value + d2.MyNode[i].Value));
}
return dp;
}
/// <summary>
/// Compute Weight of 2 DPoint dp1 and dp2
/// </summary>
/// <param name="dp1">DPoint 1</param>
/// <param name="dp2">DPoint 2</param>
/// <returns>weight of 2 vertex dp1 and dp2</returns>
public double Weight(DPoint dp1, DPoint dp2)
{
return ( Math.Exp( (-1) * Math.Pow(Math.Abs(Distance(dp1,dp2)),2) / ( 2*Xichma*Xichma)) );
}
/// <summary>
/// Euclidean distance of DPoint dp1 and dp2
/// </summary>
/// <param name="dp1">DPoint 1</param>
/// <param name="dp2">DPoint 2</param>
/// <returns>distance is Positive if dp1 and dp2 have same label. Distance is Negative if dp1 and dp2 have different label </returns>
public static double Distance(DPoint dp1, DPoint dp2)
{
double x1 = dp1.MyNode[0].Value;
double y1 = dp1.MyNode[1].Value;
double x2 = dp2.MyNode[0].Value;
double y2 = dp2.MyNode[1].Value;
if (dp1.Label * dp2.Label > 0) //Same label
return (Math.Sqrt(Math.Pow(x1 - x2, 2) + Math.Pow(y1 - y2, 2)));
else //Different label
return (-1) * (Math.Sqrt(Math.Pow(x1 - x2, 2) + Math.Pow(y1 - y2, 2)));
}
/// <summary>
/// Explore all Cluster of Data
/// </summary>
public void exploreCluster()
{
int N = _data.Count;
kMax = alpha * (int) Log2(N);
double distortionMin = double.MaxValue;
//Explore Cluster by Sampling technique
for (int t = 1; t <= nrSampling; ++t)
{
int[] cIndex = new int[N];
Dictionary<int, DPoint[]> cluster = new Dictionary<int, DPoint[]>();
List<double> distortion = new List<double>(); //distortion for nrSampling
cIndex = samplingData(kMax);
//Create cluster and calculate Distortion
for (int i = 1; i <= cIndex.Max(); ++i)
{
List<DPoint> temp = new List<DPoint>();
DPoint avg = new DPoint();
double avgDistortion = 0;
//Create cluster
for (int j = 0; j < N; ++j)
if (cIndex[j] == i)
temp.Add(_data.Data[j]);
cluster.Add(i, temp.ToArray());
if (temp.Count == 0) continue;
//Calculate Distortion
avg = temp[0]/temp.Count;
for (int j = 1; j < temp.Count; ++j)
avg = avg + temp[j]/temp.Count;
//Calculate avgDistortion
foreach (DPoint dp in temp)
avgDistortion += Distance(dp, avg);
distortion.Add(avgDistortion);
}
double sum = distortion.Sum();
if ( sum < distortionMin)
{
distortionMin = sum;
myClusters = cluster;
}
}
}
/// <summary>
/// Euclidean distance of DPoint dp1 and dp2
/// </summary>
/// <param name="dp1">DPoint 1</param>
/// <param name="dp2">DPoint 2</param>
/// <returns>distance is Positive if dp1 and dp2 have same label. Distance is Negative if dp1 and dp2 have different label </returns>
public static double Distance(DPoint dp1, DPoint dp2)
{
double x1 = dp1.MyNode[0].Value;
double y1 = dp1.MyNode[1].Value;
double x2 = dp2.MyNode[0].Value;
double y2 = dp2.MyNode[1].Value;
return(Math.Sqrt(Math.Pow(x1 - x2, 2) + Math.Pow(y1 - y2, 2)));
}
/// <summary>
/// Constructor
/// </summary>
/// <param name="_data">your data here</param>
public Dataset(DPoint[] _data)
{
Data = _data;
Count = Data.Length;
}
/// <summary>
/// Read a Dataset from a Stream
/// </summary>
/// <param name="stream">Stream to Read</param>
/// <returns>The Dataset</returns>
private static Dataset Read(Stream stream)
{
StreamReader reader = new StreamReader(stream);
//LibSVM Variable
List<DPoint> data = new List<DPoint>();
int max_index = 0;
while (!reader.EndOfStream)
{
DPoint dataLine = new DPoint();
List<Node> X = new List<Node>();
string[] tmp = reader.ReadLine().Split(' ');
for (int i = 0; i < tmp.Length; ++i)
if (tmp[i] == "")
tmp = tmp.Where(w => w != tmp[i]).ToArray();
int m = tmp.Length - 1;
Node[] x = new Node[m];
dataLine.Label = double.Parse(tmp[0]);
for (int i = 0; i < m; ++i)
{
string[] t = tmp[i + 1].Split(':');
x[i] = new Node();
x[i].Index = int.Parse(t[0]);
x[i].Value = double.Parse(t[1]);
}
dataLine.MyNode = x.ToList();
if (m > 0) max_index = Math.Max(max_index, x[m - 1].Index);
data.Add(dataLine);
}
return new Dataset(data.ToArray());
}
public static DPoint operator /(DPoint d1, double num)
{
int d = d1.MyNode.Count;
DPoint dp = new DPoint();
dp.MyNode = new List<Node>();
for (int i = 0; i < d; ++i)
{
dp.MyNode.Add(new Node(d1.MyNode[i].Index, (double)d1.MyNode[i].Value/num));
}
return dp;
}