public void induceRandomForestRecursive2(ClusNode node, RowData data, Object[] attrs) {
// System.out.println("INDUCE SPARSE with " + attrs.length + " attributes and " +
// data.getNbRows() + " examples");
// Initialize selector and perform various stopping criteria
if (initSelectorAndStopCrit(node, data)) {
makeLeaf(node);
return;
}
// Find best test
for (int i = 0; i < attrs.length; i++) {
ClusAttrType at = (ClusAttrType) attrs[i];
if (at instanceof NominalAttrType) m_FindBestTest.findNominal((NominalAttrType) at, data);
else m_FindBestTest.findNumeric((NumericAttrType) at, data);
}
// Partition data + recursive calls
CurrentBestTestAndHeuristic best = m_FindBestTest.getBestTest();
if (best.hasBestTest()) {
node.testToNode(best);
// Output best test
if (Settings.VERBOSE > 0)
System.out.println("Test: " + node.getTestString() + " -> " + best.getHeuristicValue());
// Create children
int arity = node.updateArity();
NodeTest test = node.getTest();
RowData[] subsets = new RowData[arity];
for (int j = 0; j < arity; j++) {
subsets[j] = data.applyWeighted(test, j);
}
if (getSettings().showAlternativeSplits()) {
filterAlternativeSplits(node, data, subsets);
}
if (node != m_Root && getSettings().hasTreeOptimize(Settings.TREE_OPTIMIZE_NO_INODE_STATS)) {
// Don't remove statistics of root node; code below depends on them
node.setClusteringStat(null);
node.setTargetStat(null);
}
for (int j = 0; j < arity; j++) {
ClusNode child = new ClusNode();
node.setChild(child, j);
child.initClusteringStat(m_StatManager, m_Root.getClusteringStat(), subsets[j]);
child.initTargetStat(m_StatManager, m_Root.getTargetStat(), subsets[j]);
induceRandomForestRecursive(child, subsets[j]);
}
} else {
makeLeaf(node);
}
}
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 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();
}
}
