// 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);
}
}
// try to learn new normal model and save it in correspond json file.
// the learning is in new Task and the return MODEL status field is therfore "pending".
// does afterFinishingLearning Action after the learning finished.
public MODEL LearnAndAddNewModel(string detectoionType, Train_Data data, Action afterFinishingLearning)
{
MODEL model;
// random new model_id
Random rnd = new Random(DateTime.Now.Millisecond);
int id = rnd.Next();
// lock L_NormalModels in order to add the new mode_id
lock (L_NormalModels)
{
// check it doesn't appear already, otherwise random new id.
while (L_NormalModels.ContainsKey(id) || System.IO.File.Exists(new MODEL()
{
model_id = id
}.FileName()))
{
id = rnd.Next();
}
model = new MODEL()
{
model_id = id, status = MODEL.Status_Pending,
upload_time = DateTime.Now
};
L_NormalModels.Add(id, model);
}
// start new task of learning correlative features
Task.Run(() => {
try
{
// learn noraml mode [that might take while], only if id not deleted yet
var correlation = AnomalyDetection.GetNormal(data.train_data, detectoionType, () => this.IsExist(id));
if (correlation == null)
{
throw new Exception();
}
// save it to json file
bool isSuccess = IO_Util.SaveNormalModel(model.FileName(), correlation,
new MODEL()
{
model_id = model.model_id, status = MODEL.Status_Ready, upload_time = model.upload_time
});
if (!isSuccess)
{
throw new Exception();
}
lock (L_NormalModels)
{
if (L_NormalModels.ContainsKey(id))
{
L_NormalModels[id].status = MODEL.Status_Ready;
}
else
{
// probably we won't get here since correlaion wil be null if id was deleted from L_NormalModels,
// because we sent the lambda ()=>this.IsExist(id) to AnomalyDetection.GetNormal
try { System.IO.File.Delete(new MODEL()
{
model_id = id
}.FileName()); }
catch { }
}
}
} catch {
lock (L_NormalModels)
{
if (L_NormalModels.ContainsKey(id))
{
L_NormalModels[id].status = MODEL.Status_Corrupted;
}
}
}
afterFinishingLearning();
});
// return MODEL even before the learning finished
// { model_id = id, status = MODEL.Status_Pending, upload_time = DateTime.Now };
return(model);
}
