public void Build(weka.core.Instances instances)
{
WekaUtils.DebugAssert(instances.numClasses() == 3, "instance's numClasses should be 3.");
for (int i = 0; i < m_counts.Length; i++)
{
m_counts[i] = 0;
}
double c = m_tp / m_sl;
foreach (weka.core.Instance instance in instances)
{
int v = (int)instance.classValue();
if (v == 2)
{
m_counts[2] += c;
}
else if (v == 0)
{
m_counts[0]++;
}
else
{
m_counts[1]++;
}
}
}
///
/// <summary> * Generates the classifier.
/// * </summary>
/// * <param name="instances"> set of instances serving as training data </param>
/// * <exception cref="Exception"> if the classifier has not been generated successfully </exception>
///
public override void buildClassifier(Instances instances)
{
// can classifier handle the data?
getCapabilities().testWithFail(instances);
// remove instances with missing class
instances = new Instances(instances);
instances.deleteWithMissingClass();
double sumOfWeights = 0;
WekaUtils.DebugAssert(instances.numClasses() == 3, "instance's numClasses should be 3.");
m_counts = new double[instances.numClasses()];
m_normalCounts = new double[instances.numClasses()];
for (int i = 0; i < m_counts.Length; i++)
{
m_counts[i] = 0;
m_normalCounts[i] = 0;
}
double c = m_tp / m_sl;
foreach (Instance instance in instances)
{
int v = (int)instance.classValue();
if (v == 2)
{
m_counts[v] += instance.weight() * c;
sumOfWeights += instance.weight() * c;
}
else
{
m_counts[v] += instance.weight();
sumOfWeights += instance.weight();
}
}
double start = 0;
for (int i = 0; i < m_counts.Length; ++i)
{
m_normalCounts[i] = (double)m_counts[i] / sumOfWeights + start;
start = m_normalCounts[i];
}
}
///
/// <summary> * Generates the classifier.
/// * </summary>
/// * <param name="instances"> set of instances serving as training data </param>
/// * <exception cref="Exception"> if the classifier has not been generated successfully </exception>
///
public override void buildClassifier(Instances instances)
{
// can classifier handle the data?
getCapabilities().testWithFail(instances);
// remove instances with missing class
var trainInstances = new Instances(instances);
trainInstances.deleteWithMissingClass();
WekaUtils.DebugAssert(instances.numClasses() == 3, "instance's numClasses should be 3.");
m_counts = new double[instances.numClasses()];
for (int i = 0; i < m_counts.Length; i++)
{
m_counts[i] = 0;
}
//double c = m_tp / m_sl;
foreach (Instance instance in instances)
{
int v = (int)instance.classValue();
m_counts[v] += 1;
sumOfWeights += 1;
}
}
/// <summary> Applies the cost matrix to a set of instances. If a random number generator is
/// supplied the instances will be resampled, otherwise they will be rewighted.
/// Adapted from code once sitting in Instances.java
///
/// </summary>
/// <param name="data">the instances to reweight.
/// </param>
/// <param name="random">a random number generator for resampling, if null then instances are
/// rewighted.
/// </param>
/// <returns> a new dataset reflecting the cost of misclassification.
/// </returns>
/// <exception cref="Exception">if the data has no class or the matrix in inappropriate.
/// </exception>
public virtual Instances applyCostMatrix(Instances data, System.Random random)
{
double sumOfWeightFactors = 0, sumOfMissClassWeights, sumOfWeights;
double[] weightOfInstancesInClass, weightFactor, weightOfInstances;
Instances newData;
if (data.classIndex() < 0)
{
throw new System.Exception("Class index is not set!");
}
if (size() != data.numClasses())
{
throw new System.Exception("Misclassification cost matrix has " + "wrong format!");
}
weightFactor = new double[data.numClasses()];
weightOfInstancesInClass = new double[data.numClasses()];
for (int j = 0; j < data.numInstances(); j++)
{
//UPGRADE_WARNING: Data types in Visual C# might be different. Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
weightOfInstancesInClass[(int) data.instance(j).classValue()] += data.instance(j).weight();
}
sumOfWeights = Utils.sum(weightOfInstancesInClass);
// normalize the matrix if not already
for (int i = 0; i < size(); i++)
if (!Utils.eq(getXmlElement(i, i), 0))
{
CostMatrix normMatrix = new CostMatrix(this);
normMatrix.normalize();
return normMatrix.applyCostMatrix(data, random);
}
for (int i = 0; i < data.numClasses(); i++)
{
// Using Kai Ming Ting's formula for deriving weights for
// the classes and Breiman's heuristic for multiclass
// problems.
sumOfMissClassWeights = 0;
for (int j = 0; j < data.numClasses(); j++)
{
if (Utils.sm(getXmlElement(i, j), 0))
{
throw new System.Exception("Neg. weights in misclassification " + "cost matrix!");
}
sumOfMissClassWeights += getXmlElement(i, j);
}
weightFactor[i] = sumOfMissClassWeights * sumOfWeights;
sumOfWeightFactors += sumOfMissClassWeights * weightOfInstancesInClass[i];
}
for (int i = 0; i < data.numClasses(); i++)
{
weightFactor[i] /= sumOfWeightFactors;
}
// Store new weights
weightOfInstances = new double[data.numInstances()];
for (int i = 0; i < data.numInstances(); i++)
{
//UPGRADE_WARNING: Data types in Visual C# might be different. Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
weightOfInstances[i] = data.instance(i).weight() * weightFactor[(int) data.instance(i).classValue()];
}
// Change instances weight or do resampling
if (random != null)
{
return data.resampleWithWeights(random, weightOfInstances);
}
else
{
Instances instances = new Instances(data);
for (int i = 0; i < data.numInstances(); i++)
{
instances.instance(i).Weight = weightOfInstances[i];
}
return instances;
}
}
