public override void buildClassifier(Instances instances)
{
m_mustValue = null;
var weights = MincostLiblinearClassifier.GetCount(instances);
if (weights[0] == 0)
{
m_mustValue = 2;
m_delta = 0;
return;
}
else if (weights[2] == 0)
{
m_mustValue = 0;
m_delta = 0;
return;
}
m_sampleInstances = new Instances(instances, 0);
// can classifier handle the data?
getCapabilities().testWithFail(instances);
Instances trainInstances = new Instances(instances, 0, instances.numInstances());
AddInstancesAccordWeight(trainInstances);
if (System.IO.File.Exists(m_trainingFile))
{
System.IO.File.Delete(m_trainingFile);
}
libsvmSaver.setInstances(trainInstances);
libsvmSaver.setFile(new java.io.File(m_trainingFile));
libsvmSaver.writeBatch();
//ConvertNorminalToString(m_trainingFile);
if (System.IO.File.Exists(m_modelFile))
{
System.IO.File.Delete(m_modelFile);
}
string[] options = Utils.splitOptions(m_trainArgs);
int idx = Utils.getOptionPos('c', options);
if (idx != -1)
{
double c = Convert.ToDouble(options[idx + 1]);
c = c * trainInstances.numInstances() / 100.0;
options[idx + 1] = c.ToString();
m_trainArgs = Utils.joinOptions(options);
}
learner.ExecuteLearner(s_learnerPath, m_trainingFile, m_modelFile, m_trainArgs);
if (!System.IO.File.Exists(m_modelFile))
{
throw new InvalidOperationException(learner.Output);
}
m_modelData = System.IO.File.ReadAllBytes(m_modelFile);
GetBestDelta(instances);
if (System.IO.File.Exists(m_trainingFile))
{
System.IO.File.Delete(m_trainingFile);
}
}
//private void ConvertNorminalToString(string fileName)
//{
// List<string> list = new List<string>();
// using (System.IO.StreamReader sr = new System.IO.StreamReader(fileName))
// {
// while (true)
// {
// if (sr.EndOfStream)
// break;
// string s = sr.ReadLine();
// if (string.IsNullOrEmpty(s))
// continue;
// int idx = s.IndexOf(' ');
// string c = idx == -1 ? s : s.Substring(0, idx);
// if (Convert.ToDouble(c) == 0)
// {
// list.Add("-1.0 " + (idx == -1 ? string.Empty : s.Substring(idx + 1)));
// }
// else if (Convert.ToDouble(c) == 1)
// {
// list.Add("0.0 " + (idx == -1 ? string.Empty : s.Substring(idx + 1)));
// }
// else if (Convert.ToDouble(c) == 2)
// {
// list.Add("+1.0 " + (idx == -1 ? string.Empty : s.Substring(idx + 1)));
// }
// else
// {
// list.Add(s);
// }
// }
// }
// using (System.IO.StreamWriter sw = new System.IO.StreamWriter(fileName))
// {
// foreach (string s in list)
// {
// sw.WriteLine(s);
// }
// }
//}
//private Random m_randomGenerator;
private void AddInstancesAccordWeight(Instances instances)
{
// 0, 2
double[] weights = MincostLiblinearClassifier.GetCount(instances);
if (weights == null)
{
return;
}
double c = m_tp / m_sl;
if (c == 1 && weights[0] == weights[1])
{
return;
}
int n = 0;
int toCopyClass = 0;
if (c >= 1)
{
int shouldWeight1 = (int)(c * weights[1]);
n = (int)(shouldWeight1 - weights[1]);
toCopyClass = 2;
}
else
{
int shouldShouldWeight0 = (int)(1 / c * weights[0]);
n = (int)(weights[1] - weights[0]);
toCopyClass = 0;
}
//m_randomGenerator = new Random((int)System.DateTime.Now.Ticks);
List <Instance> copyInstances = new List <Instance>();
for (int i = 0; i < instances.numInstances(); ++i)
{
if (instances.instance(i).classValue() == toCopyClass)
{
copyInstances.Add(instances.instance(i));
}
}
int nAll = n / copyInstances.Count;
for (int j = 0; j < nAll; ++j)
{
for (int i = 0; i < copyInstances.Count; ++i)
{
Instance newInstance = new weka.core.DenseInstance(copyInstances[i]);
instances.add(newInstance);
newInstance.setDataset(instances);
}
}
//for (int j = 0; j < n - nAll * copyInstances.Count; ++j)
//{
// int idx = (int)(m_randomGenerator.NextDouble() * copyInstances.Count);
// idx = Math.Min(idx, copyInstances.Count - 1);
// Instance newInstance = new weka.core.DenseInstance(copyInstances[idx]);
// instances.add(newInstance);
// newInstance.setDataset(instances);
//}
if (n - nAll * copyInstances.Count > 0)
{
Instance avgInstance = new weka.core.DenseInstance(instances.numAttributes());
for (int i = 0; i < avgInstance.numAttributes(); ++i)
{
double sum = 0;
for (int j = 0; j < copyInstances.Count; ++j)
{
sum += copyInstances[j].value(i);
}
avgInstance.setValue(i, sum / copyInstances.Count);
}
for (int j = 0; j < n - nAll * copyInstances.Count; ++j)
{
Instance newInstance = new weka.core.DenseInstance(avgInstance);
instances.add(newInstance);
}
}
}