public ClassificationResult Classify(FeatureVector features)
{
if (decisionTreeRoot == null)
{
throw new Exception("You must learn classifier before work");
}
var currentNode = decisionTreeRoot;
while (currentNode is NonTerminalNode)
{
var node = currentNode as NonTerminalNode;
var featureValue = features[node.SplittingFeature];
if (!node.Childs.ContainsKey(featureValue))
{
return new ClassificationResult
{
PredictedClass = node.TrainingExamples.Select(s => s.ExpectedResult).GetMostPopular(),
ConfidenceMeasure = -1.0
};
}
currentNode = node.Childs[featureValue];
}
return new ClassificationResult
{
PredictedClass = (currentNode as TerminalNode).PredictedClass,
ConfidenceMeasure = -1.0
};
}
public ClassificationResult Classify(FeatureVector features)
{
return Enumerable
.Range(0, outputClassesCount)
.Select(c => Estimate(features, c))
.OrderByDescending(x => x.ConfidenceMeasure)
.First();
}
public ClassificationResult Classify(FeatureVector features)
{
// not sure if this classifier returns confidence, maybe distance to decision hyperplane
double score;
var result = classifier.Compute(features.ToDoubleArray(), out score);
return new ClassificationResult
{
ConfidenceMeasure = score,
PredictedClass = result,
};
}
private ClassificationResult Estimate(FeatureVector features, int classNumber)
{
var confidence = classProb[classNumber];
for (var featureNumber = 0; featureNumber < featuresCount; featureNumber++)
{
confidence *= featureWithValueWithinClassProb[featureNumber, features[featureNumber], classNumber];
confidence /= featureWithValueProb[featureNumber, features[featureNumber]];
}
return new ClassificationResult
{
PredictedClass = classNumber,
ConfidenceMeasure = confidence
};
}
