org.knime.base.node.mine.decisiontree2.learner
Class SplitContinuous

java.lang.Object
  org.knime.base.node.mine.decisiontree2.learner.Split
      org.knime.base.node.mine.decisiontree2.learner.SplitContinuous

public class SplitContinuous
extends Split

This class determines the best split for a numeric attribute.

Author:

Christoph Sieb, University of Konstanz

Field Summary

Fields inherited from class org.knime.base.node.mine.decisiontree2.learner.Split

m_splitQualityMeasure

Constructor Summary

SplitContinuous(InMemoryTable table, int attributeIndex, SplitQualityMeasure splitQualityMeasure, boolean averageSplitpoint, double minObjectsCount)
Constructs the best split for the given numeric attribute list and the class distribution.

Method Summary

boolean	canBeFurtherUsed() For numeric splits it makes sense to use the corresponding atribute in deeper levels.
double	getBestSplitValue() Returns the split value which was evaluated as the best according to the induced partition purity.
int	getNumberPartitions() The number of partitions of a numeric split is always 2.
int	getPartitionForRow(DataRowWeighted row) Returns the partition the given row belongs to according to this split.
double[]	getPartitionWeights() Returns the partition weights.
String	toString()

Methods inherited from class org.knime.base.node.mine.decisiontree2.learner.Split

getAttributeIndex, getBestQualityMeasure, getQualityMeasureName, getSplitAttributeName, getTable, isValidSplit, setBestQualityMeasure

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

SplitContinuous

public SplitContinuous(InMemoryTable table,
                       int attributeIndex,
                       SplitQualityMeasure splitQualityMeasure,
                       boolean averageSplitpoint,
                       double minObjectsCount)

Constructs the best split for the given numeric attribute list and the class distribution. The results can be retrieved from getter methods.

Parameters:

table - the table with the data for which to create the split

attributeIndex - the index of the attribute for which to create the split

splitQualityMeasure - the quality measure (e.g. gini or gain ratio)

averageSplitpoint - if true, the split point is set as the average of the partition borders, else the upper value of the lower partition is used

minObjectsCount - the minimum number of objects in at least two partitions

Method Detail

getBestSplitValue

public double getBestSplitValue()

Returns the split value which was evaluated as the best according to the induced partition purity.

Returns:

the best split value for the underlying attribute

getNumberPartitions

public int getNumberPartitions()

The number of partitions of a numeric split is always 2. Return the number of partitions resulting from this split.

Specified by:

getNumberPartitions in class Split

Returns:

the number of partitions resulting from this split

canBeFurtherUsed

public boolean canBeFurtherUsed()

For numeric splits it makes sense to use the corresponding atribute in deeper levels. Returns true if it makes sense to use this split's attribute further in deeper levels, false if not.

Specified by:

canBeFurtherUsed in class Split

Returns:

true if it makes sense to use this split's attribute further in deeper levels, false if not

getPartitionForRow

public int getPartitionForRow(DataRowWeighted row)

Returns the partition the given row belongs to according to this split. If the value of the split attribute is missing (i.e. NaN) -1 is returned.

Specified by:

getPartitionForRow in class Split

Parameters:

row - the row for which to get the partition index

Returns:

the partition the given row belongs to according to this split; if the value of the split attribute is missing (i.e. NaN) -1 is returned

getPartitionWeights

public double[] getPartitionWeights()

Returns the partition weights. The weights represent the relative frequency of valid rows per partition. The weights are normally used to adapt the weight of rows whose split value is missing. Such a row is then assigned to each parition with the adapted weight.

Specified by:

getPartitionWeights in class Split

Returns:

the partition weights

toString

public String toString()

Overrides:

toString in class Split