@Test
public void testRoundtrip() {
Estimator estimator = getEstimator();
for (int i = 0; i < estimator.getNumberOfBuckets(); i++) {
estimator.setIfGreater(i, i % 16);
}
DenseEstimator other = new DenseEstimator(estimator.buckets());
assertEquals(estimator.buckets(), other.buckets());
}
/**
* Recreate a tree node content with the specified data based on the tree node's existing content.
*
* @param first The start(inclusive) index of the train data used for creating the tree node
* content.
* @param last The end(exclusive) index of the train data used for creating the tree node content.
* @param node The tree node whose content need to be recreated.
* @return The recreated tree node content.
*/
private TreeNodeContent createContent(int first, int last, TreeNode node) {
// Compute the total weight and its class distribution of [first
// last) prune cases
float totalWeight = 0;
AttributeDelegate classAttributeDelegate =
this.attributeDelegates[dataSet.getClassAttributeIndex()];
float[] totalClassDistri = new float[this.dataSet.getClassCount()];
Arrays.fill(totalClassDistri, 0);
for (int i = first; i < last; i++) {
int classLabel = classAttributeDelegate.getClassBranch(this.cases[i]);
totalClassDistri[classLabel] += this.weight[cases[i]];
}
// Find the original classification of the tree node
String nodeClassification = node.getContent().getClassification();
String[] classValues = this.dataSet.getClassValues();
int maxClassIndex = indexOf(nodeClassification, classValues);
// Find the most probable classification of the prune data on
// the current tree node
for (int i = 0; i < totalClassDistri.length; i++) {
totalWeight += totalClassDistri[i];
if (totalClassDistri[i] > totalClassDistri[maxClassIndex]) maxClassIndex = i;
}
String classification = classValues[maxClassIndex];
// Estimate the leafError of the tree node with the [first last)
// prune data
float basicLeafError = totalWeight - totalClassDistri[maxClassIndex];
float extraLeafError = Estimator.getExtraError(totalWeight, basicLeafError);
float estimatedLeafError = basicLeafError + extraLeafError;
return new TreeNodeContent(totalWeight, totalClassDistri, classification, estimatedLeafError);
}
/** @return true if the estimation was affected by this addition */
public boolean add(long value) {
BucketAndHash bucketAndHash = fromHash(computeHash(value), estimator.getNumberOfBuckets());
int lowestBitPosition = Long.numberOfTrailingZeros(bucketAndHash.getHash()) + 1;
if (estimator.getClass() == SparseEstimator.class
&& (estimator.estimateSizeInBytes()
>= DenseEstimator.estimateSizeInBytes(estimator.getNumberOfBuckets())
|| lowestBitPosition >= SparseEstimator.MAX_BUCKET_VALUE)) {
estimator = new DenseEstimator(estimator.buckets());
}
return estimator.setIfGreater(bucketAndHash.getBucket(), lowestBitPosition);
}
public int[] buckets() {
return estimator.buckets();
}
public int getNumberOfBuckets() {
return estimator.getNumberOfBuckets();
}
public int getSizeInBytes() {
return estimator.estimateSizeInBytes() + INSTANCE_SIZE;
}
public long estimate() {
return estimator.estimate();
}