public void terminate(ClusModel model) throws IOException {
CompleteTreeIterator iter = new CompleteTreeIterator((ClusNode) model);
while (iter.hasMoreNodes()) {
ClusNode node = (ClusNode) iter.getNextNode();
node.getClusteringStat().calcMean();
}
}
public void initialize(ClusModel model, ClusSchema schema) {
CompleteTreeIterator iter = new CompleteTreeIterator((ClusNode) model);
while (iter.hasMoreNodes()) {
ClusNode node = (ClusNode) iter.getNextNode();
ClusStatistic stat = m_Clone.cloneStat();
node.setClusteringStat(stat);
stat.setSDataSize(1);
}
}
public final void printTree(PrintWriter writer, String prefix) {
int lvc = 0;
for (int i = 0; i < m_Folds.length; i++) {
ClusNode node = m_Nodes[i];
if (!node.hasBestTest()) {
if (lvc != 0) writer.print(", ");
writer.print(m_Folds[i] + ": ");
writer.print(ClusFormat.ONE_AFTER_DOT.format(node.getTotWeight()));
lvc++;
}
}
if (lvc > 0) {
writer.print(" ");
showPath(getPath(), writer);
}
int nb = getNbChildren();
if (nb > 0) {
if (lvc > 0) {
writer.println();
writer.print(prefix);
}
} else {
writer.println();
}
for (int i = 0; i < nb; i++) {
OptXValSplit split = (OptXValSplit) getChild(i);
if (i != 0) {
writer.println(prefix + "| ");
writer.print(prefix);
}
writer.print("G" + i + " ");
writer.print(MyIntArray.print(split.getFolds()));
writer.print(" - ");
writer.print(split.getTest().getString());
writer.println();
int mb = split.getNbChildren();
String gfix = (i != nb - 1) ? "| " : " ";
for (int j = 0; j < mb; j++) {
OptXValNode node = (OptXValNode) split.getChild(j);
String suffix = (j != mb - 1) ? "| " : " ";
if (j == 0) writer.print(prefix + gfix + "+-yes: ");
else {
writer.println(prefix + gfix + "|");
writer.print(prefix + gfix + "+-no: ");
}
node.printTree(writer, prefix + gfix + suffix);
}
}
writer.flush();
}
public void prune(ClusNode node) {
RegressionStat stat = (RegressionStat) node.getClusteringStat();
m_GlobalDeviation = Math.sqrt(stat.getSVarS(m_ClusteringWeights) / stat.getTotalWeight());
pruneRecursive(node);
// System.out.println("Performing test of M5 pruning");
// TestM5PruningRuleNode.performTest(orig, node, m_GlobalDeviation, m_TargetWeights,
// m_TrainingData);
}
public final ClusNode getTree(int fold) {
int idx = Arrays.binarySearch(m_Folds, fold);
ClusNode node = m_Nodes[idx];
if (node.hasBestTest() && node.atBottomLevel()) {
OptXValSplit split = null;
int nb = getNbChildren();
for (int i = 0; i < nb; i++) {
OptXValSplit msplit = (OptXValSplit) getChild(i);
if (msplit.contains(fold)) {
split = msplit;
break;
}
}
int arity = node.updateArity();
for (int i = 0; i < arity; i++) {
OptXValNode subnode = (OptXValNode) split.getChild(i);
node.setChild(subnode.getTree(fold), i);
}
}
return node;
}
public void pruneRecursive(ClusNode node) {
if (node.atBottomLevel()) {
return;
}
for (int i = 0; i < node.getNbChildren(); i++) {
ClusNode child = (ClusNode) node.getChild(i);
pruneRecursive(child);
}
RegressionStat stat = (RegressionStat) node.getClusteringStat();
double rmsLeaf = stat.getRMSE(m_ClusteringWeights);
double adjustedErrorLeaf = rmsLeaf * pruningFactor(stat.getTotalWeight(), 1);
double rmsSubTree =
Math.sqrt(node.estimateClusteringSS(m_ClusteringWeights) / stat.getTotalWeight());
double adjustedErrorTree =
rmsSubTree * pruningFactor(stat.getTotalWeight(), node.getModelSize());
// System.out.println("C leaf: "+rmsLeaf+" tree: "+rmsSubTree);
// System.out.println("C leafadj: "+adjustedErrorLeaf +" treeadj: "+rmsSubTree);
if ((adjustedErrorLeaf <= adjustedErrorTree)
|| (adjustedErrorLeaf < (m_GlobalDeviation * 0.00001))) {
node.makeLeaf();
}
}
public void modelUpdate(DataTuple tuple, ClusModel model) {
ClusNode node = (ClusNode) model;
node.getClusteringStat().updateWeighted(tuple, 0);
}
