private Leaf build(DecisionSystem ds, DecisionClassAttribute parentLink)
{
var leaf = new Leaf { ParentLink = parentLink };
// an end of recursion
// if entropy equals 0 decision is deterministic
if (parentLink != null && parentLink.Entropy == 0)
return leaf;
// divide decision system into decision classes, assuming that
// first row contains labels,
// first column contains ordinal,
// last column contains decision
var records = ds.Records;
var decisionClasses = new List<DecisionClass>();
var decisionsVector = ds.
GetDecisionClassVector(ds.Records[0].Length - 1);
for (int i = 1; i < records[0].Length - 1; i++)
{
var decisionClassVector = ds.GetDecisionClassVector(i);
var dc = new DecisionClass(decisionClassVector,
decisionsVector);
decisionClasses.Add(dc);
}
decisionClasses.ForEach(x => x.BuildAttributesList());
// sort decision classes by their gain
decisionClasses.Sort();
// get class with the highest gain as a leaf
leaf.DecisionClass = decisionClasses.Last();
// we continue to build tree,
// searching for the next decision class for leaf
foreach (var attribute in leaf.DecisionClass.Attributes)
{
var subset = ds.GetSubset(leaf.DecisionClass.Name,
attribute.AttributeName);
leaf.Children.Add(build(subset, attribute));
}
return leaf;
}
private void buildGraph(Leaf parentVertex)
{
foreach (var childVertex in parentVertex.Children)
{
var edge = new TaggedEdge<Leaf, string>(
parentVertex,
childVertex,
childVertex.ParentLink.AttributeName);
graph.AddVerticesAndEdge(edge);
buildGraph(childVertex);
}
}
public void Build(DecisionSystem ds)
{
Root = build(ds, null);
}