public static DecisionTreeScore ScoreWithTreeWithTestSet(DecisionTreeLevel decisionTree, List <DataSetValue> testDataSetValues)
{
DecisionTreeScore score = new DecisionTreeScore(0, 0, 0, 0, decisionTree);
foreach (var testDataSetValue in testDataSetValues)
{
bool output = decisionTree.Evaluate(testDataSetValue.Values);
if (output && testDataSetValue.Output)
{
score.PositiveHit++;
}
else if (!output && !testDataSetValue.Output)
{
score.NegativeHits++;
}
else if (output && !testDataSetValue.Output)
{
score.FalsePositive++;
}
else if (!output && testDataSetValue.Output)
{
score.FalseNegative++;
}
}
score.NodeCount = decisionTree.GetNodeCount();
return(score);
}
public DecisionTreeScore(double positiveHit, double falsePositive, double negativeHits, double falseNegative, DecisionTreeLevel decisionTree)
{
_decisionTree = decisionTree;
PositiveHit = positiveHit;
FalsePositive = falsePositive;
NegativeHits = negativeHits;
FalseNegative = falseNegative;
}
public void D3(List <DataSetAttribute> attributes, List <DataSetValue> values)
{
// Check whether we even need to split or not
int totalTrueValues = values.Count(v => v.Output);
int totalFalseValues = values.Count(v => !v.Output);
if (totalFalseValues == 0 && totalTrueValues > 0)
{
_localValue = true;
return;
}
if (totalTrueValues == 0 && totalFalseValues > 0)
{
_localValue = false;
return;
}
// Can we split on attributes?
if (attributes.Count == 0)
{
// Can't split anymore. We'll decide on the most prevalent value
_localValue = totalTrueValues > totalFalseValues;
return;
}
// First, find the attribute with the highest "E"
List <DataSetAttributeWithCounts> e = CalculateEForAllAttributes(attributes, values);
DataSetAttributeWithCounts attributeWithMinEntropy = FindAttributeWithMinEntropy(e);
_attributeToSplitOn = attributeWithMinEntropy;
// Is it worth it to split on attributes
if (!ShouldSplitOnAttributeAccordingToChiSquared(attributeWithMinEntropy))
{
// Not worth it to split. We'll decide on the most prevalent value
_localValue = totalTrueValues > totalFalseValues;
return;
}
// Remove this attribute from the list of new attributes to create new subtrees
List <DataSetAttribute> newAttributes = attributes.Where(a => a.Name != attributeWithMinEntropy.Name).ToList();
// Split the values in many sets
_dictionaryOfSubTrees = new Dictionary <string, DecisionTreeLevel>(attributeWithMinEntropy.PossibleValues.Count);
var dictionaryOfValues = new Dictionary <string, List <DataSetValue> >();
foreach (var dataSetValue in values)
{
string value = dataSetValue.Values[attributeWithMinEntropy.ValueIndex];
DecisionTreeLevel localTreeLevel;
List <DataSetValue> localValues;
if (!_dictionaryOfSubTrees.TryGetValue(value, out localTreeLevel))
{
localTreeLevel = new DecisionTreeLevel(ChiTestLimit);
_dictionaryOfSubTrees[value] = localTreeLevel;
localValues = new List <DataSetValue>();
dictionaryOfValues[value] = localValues;
}
else
{
localValues = dictionaryOfValues[value];
}
localValues.Add(dataSetValue);
}
// Recursively run D3 on them
foreach (var decisionTree in _dictionaryOfSubTrees)
{
var localValues = dictionaryOfValues[decisionTree.Key];
decisionTree.Value.D3(newAttributes, localValues);
}
}
