public virtual bool isAnomaly(Timeseries ts, correlatedFeatures cf, int timeStep)
{
List <float> Column1 = ts.GetColumn(cf.feature1);
List <float> Column2 = ts.GetColumn(cf.feature2);
Point currPoint = new Point(Column1[timeStep], Column2[timeStep]);
float currDev = Anomaly_Detection_Util.dev(currPoint, cf.lin_reg);
return(currDev > cf.threshold * 1.1);
}
public float getMaxDev(List <Point> points, Line lin_reg)
{
float maxDev = 0;
for (int i = 0; i < points.Count; i++)
{
float currDev = Anomaly_Detection_Util.dev(points[i], lin_reg);
if (currDev > maxDev)
{
maxDev = currDev;
}
}
return(maxDev);
}
public virtual void addCorrelation(Timeseries ts, string feat1, string feat2, float pearson)
{
if (pearson > linThreshold)
{
correlatedFeatures correlation = new correlatedFeatures();
correlation.feature1 = feat1;
correlation.feature2 = feat2;
correlation.corrlation = pearson;
var Column1 = ts.GetColumn(feat1);
var Column2 = ts.GetColumn(feat2);
correlation.lin_reg = Anomaly_Detection_Util.LinReg(Column1, Column2);
correlation.threshold = getMaxDev(Timeseries.CombineColumns(Column1, Column2), correlation.lin_reg);
cf.Add(correlation);
}
}
public void LearnNormal(Timeseries ts)
{
for (int i = 0; i < ts.NumOfColumns - 1; i++)
{
List <float> Column1 = ts.GetColumn(i);
float max = 0;
int maxIndex = -1;
for (int j = i + 1; j < ts.NumOfColumns; j++)
{
List <float> Column2 = ts.GetColumn(j);
float correlativity = (float)Math.Abs(Anomaly_Detection_Util.Pearson(Column1, Column2));
if (correlativity > max)
{
max = correlativity;
maxIndex = j;
}
}
if (maxIndex >= 0)
{
addCorrelation(ts, ts.GetColumnName(i), ts.GetColumnName(maxIndex), max);
}
}
}