private void parseLeaf(
LeafAggregationTreeNode node,
double[] dataOfUpperLevels,
List<double[]> allGroupCombinations,
List<List<Aggregator>> allAggregators,
DataRowFactory factory,
Attribute[] newAttributes,
List<AggregationFunction> aggregationFunctions) {
// first copying data from groups
double[] newGroupCombination = new double[dataOfUpperLevels.length];
System.arraycopy(dataOfUpperLevels, 0, newGroupCombination, 0, dataOfUpperLevels.length);
allGroupCombinations.add(newGroupCombination);
// DoubleArrayDataRow row = new DoubleArrayDataRow(newData);
// check whether leaf exists
if (node != null) {
// int i = dataOfUpperLevels.length; // number of group attributes
int i = 0;
for (Aggregator aggregator : node.getAggregators()) {
allAggregators.get(i).add(aggregator);
// aggregator.set(newAttributes[i], row);
i++;
}
} else {
// fill in defaults for all aggregation functions
// int i = dataOfUpperLevels.length; // number of group attributes
// for (AggregationFunction function : aggregationFunctions) {
// function.setDefault(newAttributes[i], row);
// i++;
for (List<Aggregator> current : allAggregators) {
current.add(null);
}
}
// table.addDataRow(row);
}
@Override
public ExampleSet apply(ExampleSet exampleSet) throws OperatorException {
// creating data structures for building aggregates
List<AggregationFunction> aggregationFunctions = createAggreationFunctions(exampleSet);
// getting attributes that define groups and weights
Attribute[] groupAttributes =
getMatchingAttributes(
exampleSet.getAttributes(), getParameterAsString(PARAMETER_GROUP_BY_ATTRIBUTES));
Attribute weightAttribute = exampleSet.getAttributes().getWeight();
boolean useWeights = weightAttribute != null;
// running over exampleSet and aggregate data of each example
AggregationTreeNode rootNode = new AggregationTreeNode();
LeafAggregationTreeNode leafNode = null;
if (groupAttributes.length == 0) {
// if no grouping, we will directly insert into leaf node
leafNode = new LeafAggregationTreeNode(aggregationFunctions);
}
for (Example example : exampleSet) {
if (groupAttributes.length > 0) {
AggregationTreeNode currentNode = rootNode;
// now traversing aggregation tree for m-1 group attributes
for (int i = 0; i < groupAttributes.length - 1; i++) {
Attribute currentAttribute = groupAttributes[i];
if (currentAttribute.isNominal()) {
currentNode = currentNode.getOrCreateChild(example.getValueAsString(currentAttribute));
} else {
currentNode = currentNode.getOrCreateChild(example.getValue(currentAttribute));
}
}
// now we have to get the leaf node containing the aggregators
Attribute currentAttribute = groupAttributes[groupAttributes.length - 1];
if (currentAttribute.isNominal()) {
leafNode =
currentNode.getOrCreateLeaf(
example.getValueAsString(currentAttribute), aggregationFunctions);
} else {
leafNode =
currentNode.getOrCreateLeaf(example.getValue(currentAttribute), aggregationFunctions);
}
}
// now count current example
if (!useWeights) leafNode.count(example);
else leafNode.count(example, example.getValue(weightAttribute));
}
// now derive new example set from aggregated values
boolean isCountingAllCombinations = getParameterAsBoolean(PARAMETER_ALL_COMBINATIONS);
// building new attributes from grouping attributes and aggregation functions
Attribute[] newAttributes = new Attribute[groupAttributes.length + aggregationFunctions.size()];
for (int i = 0; i < groupAttributes.length; i++) {
newAttributes[i] = AttributeFactory.createAttribute(groupAttributes[i]);
}
int i = groupAttributes.length;
for (AggregationFunction function : aggregationFunctions) {
newAttributes[i] = function.getTargetAttribute();
i++;
}
// creating example table
MemoryExampleTable table = new MemoryExampleTable(newAttributes);
;
DataRowFactory factory = new DataRowFactory(DataRowFactory.TYPE_DOUBLE_ARRAY, '.');
double[] dataOfUpperLevels = new double[groupAttributes.length];
// prepare empty lists
ArrayList<List<Aggregator>> allAggregators = new ArrayList<List<Aggregator>>();
for (int aggregatorIdx = 0; aggregatorIdx < aggregationFunctions.size(); ++aggregatorIdx) {
allAggregators.add(new ArrayList<Aggregator>());
}
ArrayList<double[]> allGroupCombinations = new ArrayList<double[]>();
if (groupAttributes.length > 0) {
// going through all possible groups recursively
parseTree(
rootNode,
groupAttributes,
dataOfUpperLevels,
0,
allGroupCombinations,
allAggregators,
factory,
newAttributes,
isCountingAllCombinations,
aggregationFunctions);
} else {
// just enter values from single leaf node
parseLeaf(
leafNode,
dataOfUpperLevels,
allGroupCombinations,
allAggregators,
factory,
newAttributes,
aggregationFunctions);
}
// apply post-processing
int currentFunctionIdx = 0;
for (AggregationFunction aggregationFunction : aggregationFunctions) {
aggregationFunction.postProcessing(allAggregators.get(currentFunctionIdx));
++currentFunctionIdx;
}
// write data into table
int currentRow = 0;
for (double[] groupValues : allGroupCombinations) {
double[] rowData = new double[newAttributes.length];
// copy group values into row
System.arraycopy(groupValues, 0, rowData, 0, groupValues.length);
DoubleArrayDataRow dataRow = new DoubleArrayDataRow(rowData);
// copy aggregated values into row
int currentColumn = groupValues.length;
for (List<Aggregator> aggregatorsForColumn : allAggregators) {
Aggregator aggregatorForCurrentCell = aggregatorsForColumn.get(currentRow);
Attribute currentAttribute = newAttributes[currentColumn];
if (aggregatorForCurrentCell != null) {
aggregatorForCurrentCell.set(currentAttribute, dataRow);
} else {
aggregationFunctions
.get(currentColumn - groupAttributes.length)
.setDefault(currentAttribute, dataRow);
}
++currentColumn;
}
table.addDataRow(dataRow);
++currentRow;
}
// postprocessing for remaining compatibility: Old versions automatically added group "all".
// Must remain this way for old operator
// version
if (getCompatibilityLevel().isAtMost(VERSION_5_1_6)) {
if (groupAttributes.length == 0) {
Attribute resultGroupAttribute =
AttributeFactory.createAttribute(GENERIC_GROUP_NAME, Ontology.NOMINAL);
table.addAttribute(resultGroupAttribute);
table
.getDataRow(0)
.set(
resultGroupAttribute,
resultGroupAttribute.getMapping().mapString(GENERIC_ALL_NAME));
ExampleSet resultSet = table.createExampleSet();
resultSet.getAnnotations().addAll(exampleSet.getAnnotations());
for (Attribute attribute : newAttributes) {
resultSet.getAttributes().remove(attribute);
resultSet.getAttributes().addRegular(attribute);
}
return resultSet;
} else {
// make attributes nominal
ExampleSet resultSet = table.createExampleSet();
resultSet.getAnnotations().addAll(exampleSet.getAnnotations());
try {
NumericToNominal toNominalOperator =
OperatorService.createOperator(NumericToPolynominal.class);
toNominalOperator.setParameter(
AttributeSubsetSelector.PARAMETER_FILTER_TYPE,
AttributeSubsetSelector.CONDITION_REGULAR_EXPRESSION + "");
toNominalOperator.setParameter(
RegexpAttributeFilter.PARAMETER_REGULAR_EXPRESSION,
getParameterAsString(PARAMETER_GROUP_BY_ATTRIBUTES));
toNominalOperator.setParameter(
AttributeSubsetSelector.PARAMETER_INCLUDE_SPECIAL_ATTRIBUTES, "true");
return toNominalOperator.apply(resultSet);
} catch (OperatorCreationException e) {
// otherwise compatibility could not be ensured
return resultSet;
}
}
}
// for recent version table is correct: Deliver example set
ExampleSet resultSet = table.createExampleSet();
resultSet.getAnnotations().addAll(exampleSet.getAnnotations());
return resultSet;
}
