// detect anomalies of given param data according to given model id.
// this function is in one task, and might take time.
// return null if failed.
public ANOMALY Detect(int idModel, Predict_Data data)
{
string fileName = "";
List <CorrelatedFeatures> correlation = null;
lock (L_NormalModels)
{
// get file name (json file). within our server program deleting files,
// is only via the beginning of Main or via AnomalyDetectorsManager_instance.Remove(), and only 1 instance exists,
// but we lock L_NormalModels so it won't be removed meanwhile
if (L_NormalModels.ContainsKey(idModel) && L_NormalModels[idModel].status == MODEL.Status_Ready)
{
fileName = L_NormalModels[idModel].FileName();
}
if (String.IsNullOrWhiteSpace(fileName))
{
return(null);
}
// load normal model
correlation = IO_Util.LoadNormalModel(fileName);
}
if (correlation == null)
{
return(null);
}
try
{
// get detection (default detection method = "hybrid" , which it's enough if the learning was with hybrid\regression method)
var detection = AnomalyDetection.GetDetection(data.predict_data, correlation);
if (detection == null)
{
return(null);
}
Dictionary <string, string> reason = AnomalyDetection.GetReportTypes(correlation, detection);
Dictionary <String, List <Span> > spanDictionary = AnomalyDetection.ToSpanDictionary(detection);
if (reason == null || spanDictionary == null)
{
return(null);
}
return(new ANOMALY()
{
anomalies = spanDictionary, reason = reason
});
}catch
{
return(null);
}
}
