//This will calculate the errors
//first it must ask what column to use as a label
private void externalEvalCalculateButton_Click(object sender, EventArgs e)
{
//start by parsing label file
DelimitedFile delimitedLabelFile = new DelimitedFile(externalEvalLabelText.Text);
int labelCol = Prompt.ShowDialog("Enter the Column to use", "Select Attribute", 1, delimitedLabelFile.Data[0].Length);
LabelList labels = new LabelList(delimitedLabelFile.GetColumn(labelCol - 1));
//get the Partion file
Partition clusterFile = new Partition(externalEvalClusterText.Text);
int countOfPoints = clusterFile.DataCount;
//create a count mapping
//[actual cluster label, number in found clusters]
int[,] clusterMatching = new int[labels.UniqueLabels.Count, clusterFile.Clusters.Count];
foreach (Cluster c in clusterFile.Clusters)
{
foreach (ClusteredItem k in c.Points)
{
int actualMatching = labels.LabelIndices[k.Id];
int foundMatching = k.ClusterId;
clusterMatching[actualMatching, foundMatching]++;
}
}
//One-To-One Mapping like Darla's
String greedyError = ExternalEval.GreedyErrorEval(clusterFile, labels, clusterMatching);
externalEvalResultText.Text = greedyError;
}
public static double RandIndex(String labelFile, String clusterFileName)
{
//CALCULATING THE RAND INDEX
//start by parsing label file
DelimitedFile delimitedLabelFile = new DelimitedFile(labelFile);
int labelCol = delimitedLabelFile.Data[0].Length;
LabelList labels = new LabelList(delimitedLabelFile.GetColumn(labelCol - 1));
//get the Partion file
Partition clusterFile = new Partition(clusterFileName);
int[] assignments = new int[labels.LabelIndices.Length];
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
for (int j = 0; j < clusterFile.Clusters[cluster].Points.Count; j++)
{
int clusterid = clusterFile.Clusters[cluster].Points[j].ClusterId;
int id = clusterFile.Clusters[cluster].Points[j].Id;
assignments[id] = clusterid;
}
}
// compare two arrays, assigments and labels.LabelIndices
int a = 0;
int b = 0;
for (int i = 0; i < assignments.Length; i++)
{
for (int j = i + 1; j < assignments.Length; j++)
{
//Check for case a -> i and j are in same cluster in assignments and LabelIndices
if (labels.LabelIndices[i] == labels.LabelIndices[j] && assignments[i] == assignments[j])
{
a++;
}
else if (labels.LabelIndices[i] != labels.LabelIndices[j] && assignments[i] != assignments[j])
{
b++;
}
}
}
int denominator = assignments.Length * (assignments.Length - 1) / 2;
double randIndex = (a + b) / (double)denominator;
//return "Group A: " + a + " Group B: " + b + " RandIndex: " + randIndex;
return(randIndex);
}
//This will calculate the errors
//first it must ask what column to use as a label
private void externalEvalCalculateButton_Click(object sender, EventArgs e)
{
//start by parsing label file
DelimitedFile delimitedLabelFile = new DelimitedFile(externalEvalLabelText.Text);
int labelCol = Prompt.ShowDialog("Enter the Column to use", "Select Attribute", 1, delimitedLabelFile.Data[0].Length);
LabelList labels = new LabelList(delimitedLabelFile.GetColumn(labelCol - 1));
//get the Partion file
Partition clusterFile = new Partition(externalEvalClusterText.Text);
//Calculate the Error
ExternalEval error = new ExternalEval(clusterFile, labels);
externalEvalResultText.Text = error.TextResults;
}
private void button19_Click_1(object sender, EventArgs e)
{
int somDim = (int)somWidth.Value;
NetMining.Data.PointSet data = new PointSet(textBox4.Text);
HexagonalSelfOrganizingMap hSOM = new HexagonalSelfOrganizingMap(data, somDim, 0.3);
hSOM.runLargeEpochs(0.2, 1);
hSOM.runLargeEpochs(0.05, 2);
hSOM.runLargeEpochs(0.01, 2);
//hSOM.runLargeEpochs(0.01, 4);
//hSOM.runLargeEpochs(0.005, 6);
//Setup out labels
DelimitedFile f = new DelimitedFile(SOMLabelTextbox.Text);
int labelIndex = Prompt.ShowDialog("Select Label Index", "Label File", 1, f.Data[0].Length) - 1;
String[] labels = f.GetColumn(labelIndex);
//Now build our array of indicies
List <String> labelNames = new List <string>();
int[] labelIndexArr = new int[data.Count];
for (int i = 0; i < data.Count; i++)
{
if (!labelNames.Contains(labels[i]))
{
labelNames.Add(labels[i]);
}
labelIndexArr[i] = labelNames.IndexOf(labels[i]);
}
var bmp = hSOM.GetUMatrix(10, labelNames.Count, labelIndexArr);
bmp[0].Save("test" + somDim.ToString() + ".bmp");
bmp[1].Save("count" + somDim.ToString() + ".bmp");
for (int i = 2; i < bmp.Count; i++)
{
bmp[i].Save("count" + somDim.ToString() + "class_" + (i - 1).ToString() + ".bmp");
}
MessageBox.Show("Done!");
}
public static double Purity(String labelFile, String clusterFileName)
{
//start by parsing label file
DelimitedFile delimitedLabelFile = new DelimitedFile(labelFile);
int labelCol = delimitedLabelFile.Data[0].Length;
LabelList labels = new LabelList(delimitedLabelFile.GetColumn(labelCol - 1));
//get the Partion file
Partition clusterFile = new Partition(clusterFileName);
int[] majority = new int[clusterFile.Clusters.Count];
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
int[] assignments = new int[labels.UniqueLabels.Count];
for (int j = 0; j < clusterFile.Clusters[cluster].Points.Count; j++)
{
int clusterid = clusterFile.Clusters[cluster].Points[j].ClusterId;
int id = clusterFile.Clusters[cluster].Points[j].Id;
assignments[labels.LabelIndices[id]]++;
}
// now find the max of assignments
int maxAssign = 0;
for (int k = 0; k < assignments.Length; k++)
{
if (assignments[k] > maxAssign)
{
maxAssign = assignments[k];
}
}
majority[cluster] = maxAssign;
}
// add up majority[] and divide by number of vertices
int total = 0;
for (int i = 0; i < majority.Length; i++)
{
total += majority[i];
}
return((double)total / labels.LabelIndices.Length);
}
public static String CheckForNoise(String labelFile, String clusterFileName)
{
// need to calculate ns, ms and cs, as described in Yang and Leskovec ICDM2012
//start by parsing label file
DelimitedFile delimitedLabelFile = new DelimitedFile(labelFile);
int labelCol = delimitedLabelFile.Data[0].Length;
LabelList labels = new LabelList(delimitedLabelFile.GetColumn(labelCol - 1));
//get the Partion file
Partition clusterFile = new Partition(clusterFileName);
int[] assignments = new int[labels.LabelIndices.Length];
// initialize assignments array to -1
// ultimately, nodes that have been removed as part of a critical attack set will stay at -1 assignment
for (int i = 0; i < assignments.Length; i++)
{
assignments[i] = -1;
}
int noiseThreshold;
//if (assignments.Length == 550) noiseThreshold = 500;
//else if (assignments.Length == 770) noiseThreshold = 700;
//else noiseThreshold = 1100;
if (assignments.Length == 220)
{
noiseThreshold = 200;
}
else if (assignments.Length == 440)
{
noiseThreshold = 400;
}
else
{
noiseThreshold = 800;
}
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
for (int j = 0; j < clusterFile.Clusters[cluster].Points.Count; j++)
{
int clusterid = clusterFile.Clusters[cluster].Points[j].ClusterId;
int id = clusterFile.Clusters[cluster].Points[j].Id;
assignments[id] = clusterid;
}
}
int[] ns = new int[clusterFile.Clusters.Count];
int[] ms = new int[clusterFile.Clusters.Count];
int[] cs = new int[clusterFile.Clusters.Count];
Boolean[] isAllNoise = new Boolean[clusterFile.Clusters.Count];
// if we're doing this without reassign, we need new nodes and edges valuse
int edges = 0;
int nodes = 0;
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
ns[cluster] = clusterFile.Clusters[cluster].Points.Count;
isAllNoise[cluster] = true;
for (int j = 0; j < clusterFile.Clusters[cluster].Points.Count; j++) // for each vertex in this cluster
{
nodes++;
if (clusterFile.Clusters[cluster].Points[j].Id < noiseThreshold)
{
isAllNoise[cluster] = false;
}
for (int k = 0; k < clusterFile.Graph.Nodes[clusterFile.Clusters[cluster].Points[j].Id].Edge.Length; k++) // for each edge k adjacent to j
{
edges++;
int edge = clusterFile.Graph.Nodes[clusterFile.Clusters[cluster].Points[j].Id].Edge[k];
if (cluster == assignments[edge])
{
ms[cluster]++;
//if (cluster == 7) Console.WriteLine("ms " + edge);
}
else
{
cs[cluster]++;
//if (cluster == 7) Console.WriteLine("cs " + edge);
}
}
}
}
String report = "";
double[] internalDensity = new double[clusterFile.Clusters.Count];
double[] averageDegree = new double[clusterFile.Clusters.Count];
double[] expansion = new double[clusterFile.Clusters.Count];
double[] cutRatio = new double[clusterFile.Clusters.Count];
double[] conductance = new double[clusterFile.Clusters.Count];
double[] separability = new double[clusterFile.Clusters.Count];
double WAinternalDensity = 0;
double WAaverageDegree = 0;
double WAexpansion = 0;
double WAcutRatio = 0;
double WAconductance = 0;
double WAseparability = 0;
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
double totalPossibleInternalEdges = ((ns[cluster] * (ns[cluster] - 1)) / 2);
internalDensity[cluster] = totalPossibleInternalEdges == 0 ? 0 : (double)ms[cluster] / totalPossibleInternalEdges;
averageDegree[cluster] = 2.0 * ms[cluster] / ns[cluster];
expansion[cluster] = (double)cs[cluster] / ns[cluster];
cutRatio[cluster] = (double)cs[cluster] / (ns[cluster] * (assignments.Length - ns[cluster]));
conductance[cluster] = (double)cs[cluster] / (2 * ms[cluster] + cs[cluster]);
separability[cluster] = (double)ms[cluster] / cs[cluster];
}
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
WAinternalDensity += internalDensity[cluster] * ns[cluster];
WAaverageDegree += averageDegree[cluster] * ns[cluster];
WAexpansion += expansion[cluster] * ns[cluster];
WAcutRatio += cutRatio[cluster] * ns[cluster];
WAconductance += conductance[cluster] * ns[cluster];
WAseparability += separability[cluster] * ns[cluster];
}
WAinternalDensity /= (double)nodes;
WAaverageDegree /= (double)nodes;
WAexpansion /= (double)nodes;
WAcutRatio /= (double)nodes;
WAconductance /= (double)nodes;
WAseparability /= (double)nodes;
for (int cluster = 0; cluster < clusterFile.Clusters.Count; cluster++)
{
report += clusterFileName.Substring(clusterFileName.LastIndexOf('\\') + 1) + "," + cluster + "," +
(isAllNoise[cluster] ? 1 : 0) + "," + ns[cluster] + "," + ms[cluster] + "," + cs[cluster] +
"," + internalDensity[cluster] + "," + internalDensity.Min() + "," + WAinternalDensity + "," + internalDensity.Max() +
//"," + averageDegree[cluster] + "," + averageDegree.Min() + "," + averageDegree.Average() + "," + averageDegree.Max() +
"," + averageDegree[cluster] + "," + averageDegree.Min() + "," + WAaverageDegree + "," + averageDegree.Max() +
"," + expansion[cluster] + "," + expansion.Min() + "," + WAexpansion + "," + expansion.Max() +
"," + cutRatio[cluster] + "," + cutRatio.Min() + "," + WAcutRatio + "," + cutRatio.Max() +
"," + conductance[cluster] + "," + conductance.Min() + "," + WAconductance + "," + conductance.Max() +
"," + separability[cluster] + "," + separability.Min() + "," + WAseparability + "," + separability.Max() + "\n";
}
return(report);
}
/// <summary>
/// Creates a LabelList
/// </summary>
/// <param name="f">File holding the label column</param>
/// <param name="col">0-Based index of the label column</param>
public LabelList(DelimitedFile f, int col) : this(f.GetColumn(col))
{
}
