/*
* File: AbstractDecisionTreeLearner.java
* Authors: Justin Basilico
* Company: Sandia National Laboratories
* Project: Cognitive Foundry
*
* Copyright November 29, 2007, Sandia Corporation. Under the terms of Contract
* DE-AC04-94AL85000, there is a non-exclusive license for use of this work by
* or on behalf of the U.S. Government. Export of this program may require a
* license from the United States Government. See CopyrightHistory.txt for
* complete details.
*
*
*/
package gov.sandia.cognition.learning.algorithm.tree;
import gov.sandia.cognition.algorithm.AbstractIterativeAlgorithm;
import gov.sandia.cognition.learning.data.InputOutputPair;
import gov.sandia.cognition.learning.function.categorization.Categorizer;
import gov.sandia.cognition.util.ObjectUtil;
import java.io.Serializable;
import java.util.Collection;
import java.util.HashMap;
import java.util.LinkedList;
import java.util.Map;
/**
* The {@code AbstractDecisionTreeLearner} implements common functionality for
* learning algorithms that learn a decision tree.
*
* @param <InputType> The input type for the decision tree.
* @param <OutputType> The output type for the decision tree.
* @author Justin Basilico
* @since 2.0
*/
public abstract class AbstractDecisionTreeLearner<InputType, OutputType>
extends AbstractIterativeAlgorithm
implements Serializable
{
/** The learning algorithm for the decision function. */
protected DeciderLearner<? super InputType, OutputType, ?, ?> deciderLearner;
/**
* Creates a new instance of AbstractDecisionTreeLearner
*/
public AbstractDecisionTreeLearner()
{
this(null);
}
/**
* Creates a new instance of AbstractDecisionTreeLearner.
*
* @param deciderLearner The learner for the decision function
*/
public AbstractDecisionTreeLearner(
final DeciderLearner<? super InputType, OutputType, ?, ?> deciderLearner)
{
super();
this.setDeciderLearner(deciderLearner);
}
@Override
public AbstractDecisionTreeLearner<InputType, OutputType> clone()
{
@SuppressWarnings("unchecked")
final AbstractDecisionTreeLearner<InputType, OutputType> result = (AbstractDecisionTreeLearner<InputType, OutputType>)
super.clone();
result.deciderLearner = ObjectUtil.cloneSafe(this.deciderLearner);
return result;
}
/**
* Recursively learns the decision tree using the given collection
* of data, returning the created node.
*
* @param data The set of data to learn a node from.
* @param parent The parent node.
* @return The decision tree node learned from the given data.
*/
protected abstract AbstractDecisionTreeNode<InputType, OutputType, ?>
learnNode(
final Collection
<? extends InputOutputPair<? extends InputType, OutputType>>
data,
final AbstractDecisionTreeNode<InputType, OutputType, ?> parent);
/**
* Learns the child nodes for a node using the given data at the node
* plus the decision function for the node. It recursively calls the
* learnNode method on each child and then adds the child to the given
* node.
*
* @param <DecisionType> The type of decision function.
* @param node
* The node to learn the children for. The child nodes are added
* by this method.
* @param data The data at the node to learn the children for.
* @param decider The decision function to use.
*/
protected <DecisionType> void learnChildNodes(
final AbstractDecisionTreeNode<InputType, OutputType, DecisionType>
node,
final Collection
<? extends InputOutputPair<? extends InputType, OutputType>>
data,
final Categorizer<? super InputType, ? extends DecisionType> decider)
{
// We split the data up by the decider that we just created.
final Map<DecisionType, LinkedList<InputOutputPair<? extends InputType, OutputType>>>
splitsMap = this.splitData(data, decider);
if (splitsMap.size() < 2)
{
// Don't make child nodes if there are less than 2 children.
return;
}
// Loop through the split up data and learn a child node for each value.
for (Map.Entry<DecisionType, LinkedList<InputOutputPair<? extends InputType, OutputType>>> entry
: splitsMap.entrySet())
{
// Get the decision value the child node is for.
final DecisionType value = entry.getKey();
// Learn the child node.
final AbstractDecisionTreeNode<InputType, OutputType, ?> child =
this.learnNode(entry.getValue(), node);
if (child != null)
{
// Add the child.
child.setIncomingValue(value);
node.addChild(value, child);
}
// else - This should not happen unless something went wrong in
// the creation of the child.
}
}
/**
* Splits the data into new lists based on the given decision function.
*
* @param <DecisionType> The type of decision function.
* @param data The data to split.
* @param decider The decision function.
* @return
* A mapping of category decided by the decision function to a list
* of examples that have that value as indicated by the decision
* function.
*/
public <DecisionType> Map
<DecisionType, LinkedList
<InputOutputPair<? extends InputType, OutputType>>>
splitData(
final Collection
<? extends InputOutputPair<? extends InputType, OutputType>>
data,
final Categorizer<? super InputType, ? extends DecisionType> decider)
{
// Create the map to store the output.
final Map<DecisionType, LinkedList
<InputOutputPair<? extends InputType, OutputType>>>
splitsMap = new HashMap
<DecisionType,
LinkedList
<InputOutputPair<? extends InputType, OutputType>>>(2);
// Go through all the examples and apply the decider to them.
for ( InputOutputPair<? extends InputType, OutputType> example : data )
{
// Make a decision on the value.
final DecisionType value = decider.evaluate(example.getInput());
// See if a split exists for the value.
LinkedList<InputOutputPair<? extends InputType, OutputType>>
split = splitsMap.get(value);
if ( split == null )
{
// No split exists, so add it to the map.
split = new LinkedList
<InputOutputPair<? extends InputType, OutputType>>();
splitsMap.put(value, split);
}
// Add the example to the split.
split.add(example);
}
// Return the map of splits.
return splitsMap;
}
/**
* Determines if all of the output values in the collection are equal. It
* does this in a fast way by getting the first value and then checking
* it against the subsequent values and failing if one does not match.
*
* @param data The data to check for equality on the outputs.
* @return True if all the output values are equal; otherwise, false.
*/
public boolean areAllOutputsEqual(
final Collection
<? extends InputOutputPair<? extends InputType, OutputType>>
data)
{
// Store the output value to see if they are all equal.
OutputType allOutput = null;
for ( InputOutputPair<? extends InputType, OutputType> example : data )
{
// Get the output.
final OutputType output = example.getOutput();
if ( allOutput == null )
{
// First output.
allOutput = output;
}
else if ( output != null && !allOutput.equals(output) )
{
// These output values are not equal.
return false;
}
}
// All the output values are equal.
return true;
}
/**
* Gets the learner for the decision function.
*
* @return deciderLearner The learner for the decision function
*/
public DeciderLearner<? super InputType, OutputType, ?, ?> getDeciderLearner()
{
return this.deciderLearner;
}
/**
* Sets the learner for the decision function.
*
* @param deciderLearner The learner for the decision function
*/
public void setDeciderLearner(
final DeciderLearner<? super InputType, OutputType, ?, ?> deciderLearner)
{
this.deciderLearner = deciderLearner;
}
}