/** Computes average class values for each attribute and value */
private void computeAverageClassValues() {
double totalCounts, sum;
Instance instance;
double[] counts;
double[][] avgClassValues = new double[getInputFormat().numAttributes()][0];
m_Indices = new int[getInputFormat().numAttributes()][0];
for (int j = 0; j < getInputFormat().numAttributes(); j++) {
Attribute att = getInputFormat().attribute(j);
if (att.isNominal()) {
avgClassValues[j] = new double[att.numValues()];
counts = new double[att.numValues()];
for (int i = 0; i < getInputFormat().numInstances(); i++) {
instance = getInputFormat().instance(i);
if (!instance.classIsMissing() && (!instance.isMissing(j))) {
counts[(int) instance.value(j)] += instance.weight();
avgClassValues[j][(int) instance.value(j)] += instance.weight() * instance.classValue();
}
}
sum = Utils.sum(avgClassValues[j]);
totalCounts = Utils.sum(counts);
if (Utils.gr(totalCounts, 0)) {
for (int k = 0; k < att.numValues(); k++) {
if (Utils.gr(counts[k], 0)) {
avgClassValues[j][k] /= counts[k];
} else {
avgClassValues[j][k] = sum / totalCounts;
}
}
}
m_Indices[j] = Utils.sort(avgClassValues[j]);
}
}
}
/**
* Normalizes branch sizes so they contain frequencies (stored in "props") instead of counts
* (stored in "dist").
*
* <p>Overwrites the supplied "props"!
*
* <p>props.length must be == dist.length.
*/
protected static void countsToFreqs(double[][] dist, double[] props) {
for (int k = 0; k < props.length; k++) {
props[k] = Utils.sum(dist[k]);
}
if (Utils.eq(Utils.sum(props), 0)) {
for (int k = 0; k < props.length; k++) {
props[k] = 1.0 / (double) props.length;
}
} else {
FastRfUtils.normalize(props);
}
}
/**
* Normalizes branch sizes so they contain frequencies (stored in "props") instead of counts
* (stored in "dist"). Creates a new double[] which it returns.
*/
protected static double[] countsToFreqs(double[][] dist) {
double[] props = new double[dist.length];
for (int k = 0; k < props.length; k++) {
props[k] = Utils.sum(dist[k]);
}
if (Utils.eq(Utils.sum(props), 0)) {
for (int k = 0; k < props.length; k++) {
props[k] = 1.0 / (double) props.length;
}
} else {
FastRfUtils.normalize(props);
}
return props;
}
/**
* Normalize the instance
*
* @param inst instance to be normalized
* @return a new Instance with normalized values
*/
private Instance normalizeInstance(Instance inst) {
double[] vals = inst.toDoubleArray();
double sum = Utils.sum(vals);
for (int i = 0; i < vals.length; i++) {
vals[i] /= sum;
}
return new DenseInstance(inst.weight(), vals);
}
/**
* Calculates the class membership probabilities for the given test instance.
*
* @param instance the instance to be classified
* @return predicted class probability distribution
* @throws Exception if an error occurred during the prediction
*/
public double[] distributionForInstance(Instance instance) throws Exception {
String debug = "(KStar.distributionForInstance) ";
double transProb = 0.0, temp = 0.0;
double[] classProbability = new double[m_NumClasses];
double[] predictedValue = new double[1];
// initialization ...
for (int i = 0; i < classProbability.length; i++) {
classProbability[i] = 0.0;
}
predictedValue[0] = 0.0;
if (m_InitFlag == ON) {
// need to compute them only once and will be used for all instances.
// We are doing this because the evaluation module controls the calls.
if (m_BlendMethod == B_ENTROPY) {
generateRandomClassColomns();
}
m_Cache = new KStarCache[m_NumAttributes];
for (int i = 0; i < m_NumAttributes; i++) {
m_Cache[i] = new KStarCache();
}
m_InitFlag = OFF;
// System.out.println("Computing...");
}
// init done.
Instance trainInstance;
Enumeration enu = m_Train.enumerateInstances();
while (enu.hasMoreElements()) {
trainInstance = (Instance) enu.nextElement();
transProb = instanceTransformationProbability(instance, trainInstance);
switch (m_ClassType) {
case Attribute.NOMINAL:
classProbability[(int) trainInstance.classValue()] += transProb;
break;
case Attribute.NUMERIC:
predictedValue[0] += transProb * trainInstance.classValue();
temp += transProb;
break;
}
}
if (m_ClassType == Attribute.NOMINAL) {
double sum = Utils.sum(classProbability);
if (sum <= 0.0)
for (int i = 0; i < classProbability.length; i++)
classProbability[i] = (double) 1 / (double) m_NumClasses;
else Utils.normalize(classProbability, sum);
return classProbability;
} else {
predictedValue[0] = (temp != 0) ? predictedValue[0] / temp : 0.0;
return predictedValue;
}
}
private Matrix getTransposedNormedMatrix(Instances data) {
Matrix matrix = new Matrix(data.numAttributes(), data.numInstances());
for (int i = 0; i < data.numInstances(); i++) {
double[] vals = data.instance(i).toDoubleArray();
double sum = Utils.sum(vals);
for (int v = 0; v < vals.length; v++) {
vals[v] /= sum;
matrix.set(v, i, vals[v]);
}
}
return matrix;
}
/**
* Classifies a given instance. Either this or distributionForInstance() needs to be implemented
* by subclasses.
*
* @param instance the instance to be assigned to a cluster
* @return the number of the assigned cluster as an integer
* @exception Exception if instance could not be clustered successfully
*/
@Override
public int clusterInstance(Instance instance) throws Exception {
double[] dist = distributionForInstance(instance);
if (dist == null) {
throw new Exception("Null distribution predicted");
}
if (Utils.sum(dist) <= 0) {
throw new Exception("Unable to cluster instance");
}
return Utils.maxIndex(dist);
}
/**
* Calculates the class membership probabilities for the given test instance
*
* @param instance the instance to be classified
* @return predicted class probability distribution
* @throws Exception if there is a problem generating the prediction
*/
public double[] distributionForInstance(Instance instance) throws Exception {
// default model?
if (m_ZeroR != null) {
return m_ZeroR.distributionForInstance(instance);
}
// Definition of local variables
double[] probs = new double[m_NumClasses];
double prob;
double mutualInfoSum;
// store instance's att values in an int array
int[] attIndex = new int[m_NumAttributes];
for (int att = 0; att < m_NumAttributes; att++) {
if (att == m_ClassIndex) attIndex[att] = -1;
else attIndex[att] = m_StartAttIndex[att] + (int) instance.value(att);
}
// calculate probabilities for each possible class value
for (int classVal = 0; classVal < m_NumClasses; classVal++) {
probs[classVal] = 0;
prob = 1;
mutualInfoSum = 0.0;
for (int parent = 0; parent < m_NumAttributes; parent++) {
if (attIndex[parent] == -1) continue;
prob =
(m_ClassAttAttCounts[classVal][attIndex[parent]][attIndex[parent]]
+ 1.0 / (m_NumClasses * m_NumAttValues[parent]))
/ (m_NumInstances + 1.0);
for (int son = 0; son < m_NumAttributes; son++) {
if (attIndex[son] == -1 || son == parent) continue;
prob *=
(m_ClassAttAttCounts[classVal][attIndex[parent]][attIndex[son]]
+ 1.0 / m_NumAttValues[son])
/ (m_ClassAttAttCounts[classVal][attIndex[parent]][attIndex[parent]] + 1.0);
}
mutualInfoSum += m_mutualInformation[parent];
probs[classVal] += m_mutualInformation[parent] * prob;
}
probs[classVal] /= mutualInfoSum;
}
if (!Double.isNaN(Utils.sum(probs))) Utils.normalize(probs);
return probs;
}
/**
* Calculates the class membership probabilities for the given test instance.
*
* @param instance the instance to be classified
* @return predicted class probability distribution
* @exception Exception if distribution can't be computed successfully
*/
public double[] distributionForInstance(Instance instance) throws Exception {
if (instance.classAttribute().isNumeric()) {
throw new UnsupportedClassTypeException("Decorate can't handle a numeric class!");
}
double[] sums = new double[instance.numClasses()], newProbs;
Classifier curr;
for (int i = 0; i < m_Committee.size(); i++) {
curr = (Classifier) m_Committee.get(i);
newProbs = curr.distributionForInstance(instance);
for (int j = 0; j < newProbs.length; j++) sums[j] += newProbs[j];
}
if (Utils.eq(Utils.sum(sums), 0)) {
return sums;
} else {
Utils.normalize(sums);
return sums;
}
}
/**
* Compute the JS divergence between an instance and a cluster, used for test data
*
* @param inst instance to be clustered
* @param t index of the cluster
* @param pi1
* @param pi2
* @return the JS divergence
*/
private double JS(Instance inst, int t, double pi1, double pi2) {
if (Math.min(pi1, pi2) <= 0) {
System.out.format(
"Warning: zero or negative weights in JS calculation! (pi1 %s, pi2 %s)\n", pi1, pi2);
return 0;
}
double sum = Utils.sum(inst.toDoubleArray());
double kl1 = 0.0, kl2 = 0.0, tmp = 0.0;
for (int i = 0; i < inst.numValues(); i++) {
tmp = inst.valueSparse(i) / sum;
if (tmp != 0) {
kl1 += tmp * Math.log(tmp / (tmp * pi1 + pi2 * bestT.Py_t.get(inst.index(i), t)));
}
}
for (int i = 0; i < m_numAttributes; i++) {
if ((tmp = bestT.Py_t.get(i, t)) != 0) {
kl2 += tmp * Math.log(tmp / (inst.value(i) * pi1 / sum + pi2 * tmp));
}
}
return pi1 * kl1 + pi2 * kl2;
}
/**
* Test using Kononenko's MDL criterion.
*
* @param priorCounts
* @param bestCounts
* @param numInstances
* @param numCutPoints
* @return true if the split is acceptable
*/
private boolean KononenkosMDL(
double[] priorCounts, double[][] bestCounts, double numInstances, int numCutPoints) {
double distPrior, instPrior, distAfter = 0, sum, instAfter = 0;
double before, after;
int numClassesTotal;
// Number of classes occuring in the set
numClassesTotal = 0;
for (double priorCount : priorCounts) {
if (priorCount > 0) {
numClassesTotal++;
}
}
// Encode distribution prior to split
distPrior =
SpecialFunctions.log2Binomial(numInstances + numClassesTotal - 1, numClassesTotal - 1);
// Encode instances prior to split.
instPrior = SpecialFunctions.log2Multinomial(numInstances, priorCounts);
before = instPrior + distPrior;
// Encode distributions and instances after split.
for (double[] bestCount : bestCounts) {
sum = Utils.sum(bestCount);
distAfter += SpecialFunctions.log2Binomial(sum + numClassesTotal - 1, numClassesTotal - 1);
instAfter += SpecialFunctions.log2Multinomial(sum, bestCount);
}
// Coding cost after split
after = Utils.log2(numCutPoints) + distAfter + instAfter;
// Check if split is to be accepted
return (before > after);
}
/**
* Compute the entropy score based on an array of probabilities
*
* @param probs array of non-negative and normalized probabilities
* @return the entropy value
*/
private double Entropy(double[] probs) {
for (double prob : probs) {
if (prob <= 0) {
if (m_verbose) {
System.out.println("Warning: Negative probability.");
}
return Double.NaN;
}
}
// could be unormalized, when normalization is not specified
if (Math.abs(Utils.sum(probs) - 1) >= 1e-6) {
if (m_verbose) {
System.out.println("Warning: Not normalized.");
}
return Double.NaN;
}
double mi = 0.0;
for (double prob : probs) {
mi += prob * Math.log(prob);
}
mi = -mi;
return mi;
}
/**
* Generates the classifier.
*
* @param instances set of instances serving as training data
* @exception Exception if the classifier has not been generated successfully
*/
@Override
public void buildClassifier(Instances instances) throws Exception {
int attIndex = 0;
double sum;
// can classifier handle the data?
getCapabilities().testWithFail(instances);
// remove instances with missing class
instances = new Instances(instances);
instances.deleteWithMissingClass();
m_Instances = new Instances(instances, 0);
// Reserve space
m_Counts = new double[instances.numClasses()][instances.numAttributes() - 1][0];
m_Means = new double[instances.numClasses()][instances.numAttributes() - 1];
m_Devs = new double[instances.numClasses()][instances.numAttributes() - 1];
m_Priors = new double[instances.numClasses()];
Enumeration<Attribute> enu = instances.enumerateAttributes();
while (enu.hasMoreElements()) {
Attribute attribute = enu.nextElement();
if (attribute.isNominal()) {
for (int j = 0; j < instances.numClasses(); j++) {
m_Counts[j][attIndex] = new double[attribute.numValues()];
}
} else {
for (int j = 0; j < instances.numClasses(); j++) {
m_Counts[j][attIndex] = new double[1];
}
}
attIndex++;
}
// Compute counts and sums
Enumeration<Instance> enumInsts = instances.enumerateInstances();
while (enumInsts.hasMoreElements()) {
Instance instance = enumInsts.nextElement();
if (!instance.classIsMissing()) {
Enumeration<Attribute> enumAtts = instances.enumerateAttributes();
attIndex = 0;
while (enumAtts.hasMoreElements()) {
Attribute attribute = enumAtts.nextElement();
if (!instance.isMissing(attribute)) {
if (attribute.isNominal()) {
m_Counts[(int) instance.classValue()][attIndex][(int) instance.value(attribute)]++;
} else {
m_Means[(int) instance.classValue()][attIndex] += instance.value(attribute);
m_Counts[(int) instance.classValue()][attIndex][0]++;
}
}
attIndex++;
}
m_Priors[(int) instance.classValue()]++;
}
}
// Compute means
Enumeration<Attribute> enumAtts = instances.enumerateAttributes();
attIndex = 0;
while (enumAtts.hasMoreElements()) {
Attribute attribute = enumAtts.nextElement();
if (attribute.isNumeric()) {
for (int j = 0; j < instances.numClasses(); j++) {
if (m_Counts[j][attIndex][0] < 2) {
throw new Exception(
"attribute "
+ attribute.name()
+ ": less than two values for class "
+ instances.classAttribute().value(j));
}
m_Means[j][attIndex] /= m_Counts[j][attIndex][0];
}
}
attIndex++;
}
// Compute standard deviations
enumInsts = instances.enumerateInstances();
while (enumInsts.hasMoreElements()) {
Instance instance = enumInsts.nextElement();
if (!instance.classIsMissing()) {
enumAtts = instances.enumerateAttributes();
attIndex = 0;
while (enumAtts.hasMoreElements()) {
Attribute attribute = enumAtts.nextElement();
if (!instance.isMissing(attribute)) {
if (attribute.isNumeric()) {
m_Devs[(int) instance.classValue()][attIndex] +=
(m_Means[(int) instance.classValue()][attIndex] - instance.value(attribute))
* (m_Means[(int) instance.classValue()][attIndex]
- instance.value(attribute));
}
}
attIndex++;
}
}
}
enumAtts = instances.enumerateAttributes();
attIndex = 0;
while (enumAtts.hasMoreElements()) {
Attribute attribute = enumAtts.nextElement();
if (attribute.isNumeric()) {
for (int j = 0; j < instances.numClasses(); j++) {
if (m_Devs[j][attIndex] <= 0) {
throw new Exception(
"attribute "
+ attribute.name()
+ ": standard deviation is 0 for class "
+ instances.classAttribute().value(j));
} else {
m_Devs[j][attIndex] /= m_Counts[j][attIndex][0] - 1;
m_Devs[j][attIndex] = Math.sqrt(m_Devs[j][attIndex]);
}
}
}
attIndex++;
}
// Normalize counts
enumAtts = instances.enumerateAttributes();
attIndex = 0;
while (enumAtts.hasMoreElements()) {
Attribute attribute = enumAtts.nextElement();
if (attribute.isNominal()) {
for (int j = 0; j < instances.numClasses(); j++) {
sum = Utils.sum(m_Counts[j][attIndex]);
for (int i = 0; i < attribute.numValues(); i++) {
m_Counts[j][attIndex][i] =
(m_Counts[j][attIndex][i] + 1) / (sum + attribute.numValues());
}
}
}
attIndex++;
}
// Normalize priors
sum = Utils.sum(m_Priors);
for (int j = 0; j < instances.numClasses(); j++) {
m_Priors[j] = (m_Priors[j] + 1) / (sum + instances.numClasses());
}
}
/**
* Classifies the given test instance.
*
* @param instance the instance to be classified
* @return the predicted class for the instance
* @throws Exception if the instance can't be classified
*/
public double[] distributionForInstance(Instance instance) throws Exception {
double[] dist = new double[m_NumClasses];
double[] temp = new double[m_NumClasses];
double weight = 1.0;
for (int i = 0; i < instance.numAttributes(); i++) {
if (i != m_ClassIndex && !instance.isMissing(i)) {
double val = instance.value(i);
boolean ok = false;
if (instance.attribute(i).isNumeric()) {
int k;
for (k = m_intervalBounds[i].length - 1; k >= 0; k--) {
if (val > m_intervalBounds[i][k]) {
for (int j = 0; j < m_NumClasses; j++) {
if (m_globalCounts[j] > 0) {
temp[j] = ((m_counts[i][k][j] + TINY) / (m_globalCounts[j] + TINY));
}
}
ok = true;
break;
} else if (val == m_intervalBounds[i][k]) {
for (int j = 0; j < m_NumClasses; j++) {
if (m_globalCounts[j] > 0) {
temp[j] = ((m_counts[i][k][j] + m_counts[i][k - 1][j]) / 2.0) + TINY;
temp[j] /= (m_globalCounts[j] + TINY);
}
}
ok = true;
break;
}
}
if (!ok) {
throw new Exception("This shouldn't happen");
}
} else { // nominal attribute
ok = true;
for (int j = 0; j < m_NumClasses; j++) {
if (m_globalCounts[j] > 0) {
temp[j] = ((m_counts[i][(int) val][j] + TINY) / (m_globalCounts[j] + TINY));
}
}
}
double sum = Utils.sum(temp);
if (sum <= 0) {
for (int j = 0; j < temp.length; j++) {
temp[j] = 1.0 / (double) temp.length;
}
} else {
Utils.normalize(temp, sum);
}
if (m_weightByConfidence) {
weight = weka.core.ContingencyTables.entropy(temp);
weight = Math.pow(weight, m_bias);
if (weight < 1.0) {
weight = 1.0;
}
}
for (int j = 0; j < m_NumClasses; j++) {
dist[j] += (temp[j] * weight);
}
}
}
double sum = Utils.sum(dist);
if (sum <= 0) {
for (int j = 0; j < dist.length; j++) {
dist[j] = 1.0 / (double) dist.length;
}
return dist;
} else {
Utils.normalize(dist, sum);
return dist;
}
}
private double[] calculateRegionProbs(int j, int i) throws Exception {
double[] sumOfProbsForRegion = new double[m_trainingData.classAttribute().numValues()];
for (int u = 0; u < m_numOfSamplesPerRegion; u++) {
double[] sumOfProbsForLocation = new double[m_trainingData.classAttribute().numValues()];
m_weightingAttsValues[m_xAttribute] = getRandomX(j);
m_weightingAttsValues[m_yAttribute] = getRandomY(m_panelHeight - i - 1);
m_dataGenerator.setWeightingValues(m_weightingAttsValues);
double[] weights = m_dataGenerator.getWeights();
double sumOfWeights = Utils.sum(weights);
int[] indices = Utils.sort(weights);
// Prune 1% of weight mass
int[] newIndices = new int[indices.length];
double sumSoFar = 0;
double criticalMass = 0.99 * sumOfWeights;
int index = weights.length - 1;
int counter = 0;
for (int z = weights.length - 1; z >= 0; z--) {
newIndices[index--] = indices[z];
sumSoFar += weights[indices[z]];
counter++;
if (sumSoFar > criticalMass) {
break;
}
}
indices = new int[counter];
System.arraycopy(newIndices, index + 1, indices, 0, counter);
for (int z = 0; z < m_numOfSamplesPerGenerator; z++) {
m_dataGenerator.setWeightingValues(m_weightingAttsValues);
double[][] values = m_dataGenerator.generateInstances(indices);
for (int q = 0; q < values.length; q++) {
if (values[q] != null) {
System.arraycopy(values[q], 0, m_vals, 0, m_vals.length);
m_vals[m_xAttribute] = m_weightingAttsValues[m_xAttribute];
m_vals[m_yAttribute] = m_weightingAttsValues[m_yAttribute];
// classify the instance
m_dist = m_classifier.distributionForInstance(m_predInst);
for (int k = 0; k < sumOfProbsForLocation.length; k++) {
sumOfProbsForLocation[k] += (m_dist[k] * weights[q]);
}
}
}
}
for (int k = 0; k < sumOfProbsForRegion.length; k++) {
sumOfProbsForRegion[k] += (sumOfProbsForLocation[k] * sumOfWeights);
}
}
// average
Utils.normalize(sumOfProbsForRegion);
// cache
double[] tempDist = new double[sumOfProbsForRegion.length];
System.arraycopy(sumOfProbsForRegion, 0, tempDist, 0, sumOfProbsForRegion.length);
return tempDist;
}
/**
* Recursively generates a tree. A derivative of the buildTree function from the
* "weka.classifiers.trees.RandomTree" class, with the following changes made:
*
* <ul>
* <li>m_ClassProbs are now remembered only in leaves, not in every node of the tree
* <li>m_Distribution has been removed
* <li>members of dists, splits, props and vals arrays which are not used are dereferenced prior
* to recursion to reduce memory requirements
* <li>a check for "branch with no training instances" is now (FastRF 0.98) made before
* recursion; with the current implementation of splitData(), empty branches can appear only
* with nominal attributes with more than two categories
* <li>each new 'tree' (i.e. node or leaf) is passed a reference to its 'mother forest',
* necessary to look up parameters such as maxDepth and K
* <li>pre-split entropy is not recalculated unnecessarily
* <li>uses DataCache instead of weka.core.Instances, the reference to the DataCache is stored
* as a field in FastRandomTree class and not passed recursively down new buildTree() calls
* <li>similarly, a reference to the random number generator is stored in a field of the
* DataCache
* <li>m_ClassProbs are now normalized by dividing with number of instances in leaf, instead of
* forcing the sum of class probabilities to 1.0; this has a large effect when
* class/instance weights are set by user
* <li>a little imprecision is allowed in checking whether there was a decrease in entropy after
* splitting
* <li>0.99: the temporary arrays splits, props, vals now are not wide as the full number of
* attributes in the dataset (of which only "k" columns of randomly chosen attributes get
* filled). Now, it's just a single array which gets replaced as the k features are
* evaluated sequentially, but it gets replaced only if a next feature is better than a
* previous one.
* <li>0.99: the SortedIndices are now not cut up into smaller arrays on every split, but rather
* re-sorted within the same array in the splitDataNew(), and passed down to buildTree() as
* the original large matrix, but with start and end points explicitly specified
* </ul>
*
* @param sortedIndices the indices of the instances of the whole bootstrap replicate
* @param startAt First index of the instance to consider in this split; inclusive.
* @param endAt Last index of the instance to consider; inclusive.
* @param classProbs the class distribution
* @param debug whether debugging is on
* @param attIndicesWindow the attribute window to choose attributes from
* @param depth the current depth
*/
protected void buildTree(
int[][] sortedIndices,
int startAt,
int endAt,
double[] classProbs,
boolean debug,
int[] attIndicesWindow,
int depth) {
m_Debug = debug;
int sortedIndicesLength = endAt - startAt + 1;
// Check if node doesn't contain enough instances or is pure
// or maximum depth reached, make leaf.
if ((sortedIndicesLength < Math.max(2, getMinNum())) // small
|| Utils.eq(classProbs[Utils.maxIndex(classProbs)], Utils.sum(classProbs)) // pure
|| ((getMaxDepth() > 0) && (depth >= getMaxDepth())) // deep
) {
m_Attribute = -1; // indicates leaf (no useful attribute to split on)
// normalize by dividing with the number of instances (as of ver. 0.97)
// unless leaf is empty - this can happen with splits on nominal
// attributes with more than two categories
if (sortedIndicesLength != 0)
for (int c = 0; c < classProbs.length; c++) {
classProbs[c] /= sortedIndicesLength;
}
m_ClassProbs = classProbs;
this.data = null;
return;
} // (leaf making)
// new 0.99: all the following are for the best attribute only! they're updated while
// sequentially through the attributes
double val = Double.NaN; // value of splitting criterion
double[][] dist =
new double[2]
[data.numClasses]; // class distributions (contingency table), indexed first by branch,
// then by class
double[] prop = new double[2]; // the branch sizes (as fraction)
double split = Double.NaN; // split point
// Investigate K random attributes
int attIndex = 0;
int windowSize = attIndicesWindow.length;
int k = getKValue();
boolean sensibleSplitFound = false;
double prior = Double.NaN;
double bestNegPosterior = -Double.MAX_VALUE;
int bestAttIdx = -1;
while ((windowSize > 0) && (k-- > 0 || !sensibleSplitFound)) {
int chosenIndex = data.reusableRandomGenerator.nextInt(windowSize);
attIndex = attIndicesWindow[chosenIndex];
// shift chosen attIndex out of window
attIndicesWindow[chosenIndex] = attIndicesWindow[windowSize - 1];
attIndicesWindow[windowSize - 1] = attIndex;
windowSize--;
// new: 0.99
double candidateSplit =
distributionSequentialAtt(
prop, dist, bestNegPosterior, attIndex, sortedIndices[attIndex], startAt, endAt);
if (Double.isNaN(candidateSplit)) {
continue; // we did not improve over a previous attribute! "dist" is unchanged from before
}
// by this point we know we have an improvement, so we keep the new split point
split = candidateSplit;
bestAttIdx = attIndex;
if (Double.isNaN(
prior)) { // needs to be computed only once per branch - is same for all attributes (even
// regardless of missing values)
prior = SplitCriteria.entropyOverColumns(dist);
}
double negPosterior =
-SplitCriteria.entropyConditionedOnRows(dist); // this is an updated dist
if (negPosterior > bestNegPosterior) {
bestNegPosterior = negPosterior;
} else {
throw new IllegalArgumentException("Very strange!");
}
val = prior - (-negPosterior); // we want the greatest reduction in entropy
if (val > 1e-2) { // we allow some leeway here to compensate
sensibleSplitFound = true; // for imprecision in entropy computation
}
} // feature by feature in window
if (sensibleSplitFound) {
m_Attribute = bestAttIdx; // find best attribute
m_SplitPoint = split;
m_Prop = prop;
prop = null; // can be GC'ed
// int[][][] subsetIndices =
// new int[dist.length][data.numAttributes][];
// splitData( subsetIndices, m_Attribute,
// m_SplitPoint, sortedIndices );
// int numInstancesBeforeSplit = sortedIndices[0].length;
int belowTheSplitStartsAt =
splitDataNew(m_Attribute, m_SplitPoint, sortedIndices, startAt, endAt);
m_Successors = new FastRandomTree[dist.length]; // dist.length now always == 2
for (int i = 0; i < dist.length; i++) {
m_Successors[i] = new FastRandomTree();
m_Successors[i].m_MotherForest = this.m_MotherForest;
m_Successors[i].data = this.data;
// new in 0.99 - used in distributionSequentialAtt()
m_Successors[i].tempDists = this.tempDists;
m_Successors[i].tempDistsOther = this.tempDistsOther;
m_Successors[i].tempProps = this.tempProps;
// check if we're about to make an empty branch - this can happen with
// nominal attributes with more than two categories (as of ver. 0.98)
if (belowTheSplitStartsAt - startAt == 0) {
// in this case, modify the chosenAttDists[i] so that it contains
// the current, before-split class probabilities, properly normalized
// by the number of instances (as we won't be able to normalize
// after the split)
for (int j = 0; j < dist[i].length; j++) dist[i][j] = classProbs[j] / sortedIndicesLength;
}
if (i == 0) { // before split
m_Successors[i].buildTree(
sortedIndices,
startAt,
belowTheSplitStartsAt - 1,
dist[i],
m_Debug,
attIndicesWindow,
depth + 1);
} else { // after split
m_Successors[i].buildTree(
sortedIndices,
belowTheSplitStartsAt,
endAt,
dist[i],
m_Debug,
attIndicesWindow,
depth + 1);
}
dist[i] = null;
}
sortedIndices = null;
} else { // ------ make leaf --------
m_Attribute = -1;
// normalize by dividing with the number of instances (as of ver. 0.97)
// unless leaf is empty - this can happen with splits on nominal attributes
if (sortedIndicesLength != 0)
for (int c = 0; c < classProbs.length; c++) {
classProbs[c] /= sortedIndicesLength;
}
m_ClassProbs = classProbs;
}
this.data = null; // dereference all pointers so data can be GC'd after tree is built
}