private void LearnHelper(IDictionary <string, List <float> > table, float maxCore, string f1, string f2)
{
CorrelatedFeatures cfs = new CorrelatedFeatures();
cfs.Feature1 = f1;
cfs.Feature2 = f2;
cfs.Corr = maxCore;
cfs.Lin_Reg = AnomalyUtils.LinearReg(table[f1], table[f2]);
cfs.Threshold = FindThreshold(table[f1], table[f2], cfs.Lin_Reg) * 1.1f;
this.cf.Add(cfs);
}
//sends each x y pair to dev and returns the highest
public float HighestDev(List <float> x, List <float> y, Line line)
{
float maxDev = 0;
for (int i = 0; i < x.Count; i++)
{
Point p = new Point(x[i], y[i]);
float newDev = AnomalyUtils.Dev(p, line);
if (newDev > maxDev)
{
maxDev = newDev;
}
}
return(maxDev);
}
private float FindThreshold(List <float> l1, List <float> l2, Line lin_reg)
{
float max = 0;
int size = l1.Count;
for (int i = 0; i < size; i++)
{
Point point = new Point(l1[i], l2[i]);
float distance = AnomalyUtils.Dev(point, lin_reg);
if (distance > max)
{
max = distance;
}
}
return(max);
}
public List <AnomalyReport> detect(IDictionary <string, List <float> > table)
{
List <AnomalyReport> anomalyReport = new List <AnomalyReport>();
foreach (CorrelatedFeatures correlatedFeature in cf)
{
for (int i = 0; i < table[correlatedFeature.Feature1].Count; i++)
{
Point p = new Point(table[correlatedFeature.Feature1].ElementAt(i), table[correlatedFeature.Feature1].ElementAt(i));
if (AnomalyUtils.Dev(p, correlatedFeature.Lin_Reg) > correlatedFeature.Threshold)
{
anomalyReport.Add(new AnomalyReport(correlatedFeature.Feature1 + "-" + correlatedFeature.Feature2, i));
}
}
}
return(anomalyReport);
}
public void LearnNormal(IDictionary <string, List <float> > table)
{
int columnSize = table.ElementAt(0).Value.Count; //table.Keys.ElementAt(0)
int rowSize = table.Count;
for (int i = 0; i < table.Count; i++)
{
CorrelatedFeatures cfs = new CorrelatedFeatures();
cfs.Corr = -1;
cfs.Feature1 = table.ElementAt(i).Key;
for (int j = i + 1; j < table.Count; j++)
{
float correlation = AnomalyUtils.Pearson(table.ElementAt(i).Value, table.ElementAt(j).Value);
if (correlation > correlationThreshold)
{
if (correlation > cfs.Corr)
{
cfs.Feature2 = table.ElementAt(j).Key;
cfs.Corr = correlation;
cfs.Lin_Reg = AnomalyUtils.LinearReg(table.ElementAt(i).Value,
table.ElementAt(j).Value);
cfs.Threshold = HighestDev(table.ElementAt(i).Value,
table.ElementAt(j).Value, cfs.Lin_Reg);
List <Point> pointsList = new List <Point>();
for (int k = 1; k < columnSize; k++)
{
pointsList.Add(new Point(table[cfs.Feature1].ElementAt(k), table[cfs.Feature2].ElementAt(k)));
}
cfs.CircleThreshold = MinCircle.FindMinCircle(pointsList);
cfs.CircleThreshold.radius *= 1.1;
}
}
}
if (cfs.Corr > -1)
{
cf.Add(cfs);
}
}
}
public void LearnNormal(IDictionary <string, List <float> > table)
{
int size = table.Count;
for (int i = 0; i < size; i++)
{
string f1 = table.ElementAt(i).Key;
float maxCore = 0;
int maxIndex = 0;
for (int j = i + 1; j < size; j++)
{
float cor = Math.Abs(AnomalyUtils.Pearson(table.ElementAt(i).Value, table.ElementAt(j).Value));
if (cor > maxCore)
{
maxCore = cor;
maxIndex = j;
}
}
string fMostCore = table.ElementAt(maxIndex).Key;
LearnHelper(table, maxCore, f1, fMostCore);
}
}
public Line CorrLinReg(string fName, string corrName)
{
this.linReg = AnomalyUtils.LinearReg(table[fName], table[corrName]);
return(this.linReg);
}
