/// <summary>
///
/// </summary>
/// <param name="dataSet"></param>
/// <param name="labels">label name for each feature column</param>
/// <param name="node">the parent node</param>
private static void CreateTree(List <DataSetItem> dataSet, List <string> labels, DecisionTreeNode node)
{
var classList = (from r in dataSet select r.ClassName).ToList();
if (classList.Distinct().Count() <= 1)
{
//stop when all class are equal
DecisionTreeNode child = new DecisionTreeNode()
{
ClassName = classList[0]
};
node.ChildrenList.Add(child);
return;
}
if (dataSet[0].Features == null || dataSet[0].Features.Length <= 0)
{
//no more feature to split,choose the most frequent class as classname
DecisionTreeNode child = new DecisionTreeNode()
{
ClassName = MajorityClass(classList), IsMajority = true
};
node.ChildrenList.Add(child);
return;
}
int bestIndex = ChooseBestFeatureToSplit(dataSet);
string bestFeatLabel = labels[bestIndex];
labels.RemoveAt(bestIndex);
var uniqueVals = (from r in dataSet select r.Features[bestIndex]).Distinct().ToList();
foreach (var featVal in uniqueVals)
{
DecisionTreeNode child = new DecisionTreeNode()
{
label = bestFeatLabel, FeatValue = featVal
};
node.ChildrenList.Add(child);
CreateTree(SplitDataSet(dataSet, bestIndex, featVal), labels.ToList(), child);
}
}