internal static Decision GetBestDecisionFromDataSet(IReadOnlyCollection <DataSetObject> dataSet, out NodeSplit splitResult)
{
var currentEntropy = dataSet.GetEntropy();
double minEntropy = currentEntropy;
Decision bestDecision = k => true;
splitResult = new NodeSplit(dataSet, Array.Empty <DataSetObject>());
foreach (var obj in dataSet)
{
for (int i = 0; i < obj.Features.Length; i++)
{
int j = i;
Decision lessDecision = k => obj.Features[j] < k.Features[j];
var split = dataSet.SplitByDecision(lessDecision);
var lessDecisionEntropy = split.GetWeightedEntropy();
if (lessDecisionEntropy < minEntropy)
{
(minEntropy, bestDecision, splitResult) = (lessDecisionEntropy, lessDecision, split);
}
}
}
return(bestDecision);
}
private static DecisionTreeItem InitializeTreeItem(Decision decision, IReadOnlyCollection <DataSetObject> dataSet, NodeSplit curSplit, int levelAvailableCount)
{
var label = dataSet.First().Label;
if (dataSet.All(k => k.Label == label))
{
return(new DecisionTreeLeaf(dataSet.First().Label));
}
if (levelAvailableCount == 1)
{
return(new DecisionTreeLeaf(dataSet.GetClassByMaxCount()));
}
levelAvailableCount--;
var leftDecision = DecisionTreeItem.GetBestDecisionFromDataSet(curSplit.LeftPart, out var leftSplit);
var leftChild = InitializeTreeItem(leftDecision, curSplit.LeftPart, leftSplit, levelAvailableCount);
var rightDecision = DecisionTreeItem.GetBestDecisionFromDataSet(curSplit.RightPart, out var rightSplit);
var rightChild = InitializeTreeItem(rightDecision, curSplit.RightPart, rightSplit, levelAvailableCount);
return(new DecisionTreeNode(leftChild, rightChild, decision, dataSet.GetClassByMaxCount()));
}
