private void RecordStep(IInference infer, string classifierField)
{
double newPrediction;
if (null != infer.GetClassification(classifierField).GetMostProbableValue(1))
{
newPrediction = (double)infer.GetClassification(classifierField).GetMostProbableValue(1);
}
else
{
newPrediction = _predictedValue;
}
if (infer.GetRecordNum() > 0)
{
double actual = (double)infer.GetClassifierInput()[classifierField].Get("inputValue");
double error = Math.Abs(_predictedValue - actual);
PredictionValue value = new PredictionValue();
value.RecordNum = infer.GetRecordNum();
value.ActualValue = actual;
value.PredictionError = error;
value.PredictedValue = newPrediction;
value.AnomalyFactor = infer.GetAnomalyScore();
_predictions.Add(value);
}
_predictedValue = newPrediction;
}
/**
* Used to manually input data into a {@link Region}, the other way
* being the call to {@link Region#start()} for a Region that contains
* a {@link Layer} which in turn contains a {@link Sensor} <em>-OR-</em>
* subscribing a receiving Region to this Region's output Observable.
*
* @param input One of (int[], String[], {@link ManualInput}, or Map<String, Object>)
*/
public void Compute(IInference input)
{
if (!assemblyClosed)
{
close();
}
this.input = input;
((Layer <IInference>)tail).Compute(input);
}
public override object PostDeSerialize(object i)
{
IInference retVal = (IInference)base.PostDeSerialize(i);
Assert.IsNotNull(retVal);
Assert.IsTrue(retVal != i); // Ensure Objects not same
Assert.IsTrue(retVal.Equals(i)); // However they are still equal!
//callVerify.add(retVal);
//Assert.IsTrue(callVerify.size() == 1);
callVerify.Add(retVal);
Assert.IsTrue(callVerify.Count == 1);
return((ManualInput)retVal);
}
public void TestSerializeDeSerialize()
{
SerializerCore serializer = Persistence.Get().Serializer();
IInference inf = new ManualInputWithPostDeserialize();
byte[] bytes = serializer.Serialize(inf);
Assert.IsNotNull(bytes);
IInference serializedInf = serializer.Deserialize <ManualInput>(bytes);
Assert.IsNotNull(serializedInf);
Assert.IsTrue(callVerify.Count == 1);
}
/**
* Primitive file appender for collecting output. This just demonstrates how to use
* {@link Subscriber#onNext(Object)} to accomplish some work.
*
* @param infer The {@link Inference} object produced by the Network
* @param classifierField The field we use in this demo for anomaly computing.
*/
private void WriteToFile(IInference infer, string classifierField)
{
try
{
double newPrediction;
if (null != infer.GetClassification(classifierField).GetMostProbableValue(1))
{
newPrediction = (double)infer.GetClassification(classifierField).GetMostProbableValue(1);
}
else
{
newPrediction = predictedValue;
}
if (infer.GetRecordNum() > 0)
{
double actual = (double)infer.GetClassifierInput()[classifierField].Get("inputValue");
double error = Math.Abs(predictedValue - actual);
StringBuilder sb = new StringBuilder()
.Append(infer.GetRecordNum()).Append(", ")
//.Append("classifier input=")
.Append(string.Format("{0}", actual.ToString(NumberFormatInfo.InvariantInfo))).Append(",")
//.Append("prediction= ")
.Append(string.Format("{0}", predictedValue.ToString(NumberFormatInfo.InvariantInfo))).Append(",")
.Append(string.Format("{0}", error.ToString(NumberFormatInfo.InvariantInfo))).Append(",")
//.Append("anomaly score=")
.Append(infer.GetAnomalyScore().ToString(NumberFormatInfo.InvariantInfo));
pw.WriteLine(sb.ToString());
pw.Flush();
if (infer.GetRecordNum() % 100 == 0)
{
Console.WriteLine(sb.ToString());
}
}
predictedValue = newPrediction;
}
catch (Exception e)
{
Console.WriteLine(e);
pw.Flush();
}
}
