burlap.behavior.singleagent.planning.vfa.fittedvi

Class WekaVFATrainer

java.lang.Object

burlap.behavior.singleagent.planning.vfa.fittedvi.WekaVFATrainer

All Implemented Interfaces:

SupervisedVFA

public class WekaVFATrainer
extends java.lang.Object
implements SupervisedVFA

A class for using supervised learning algorithms provided by the Weka library for fitted value iteration. This class takes a generator for Weka Classifier objects (specified with the WekaVFATrainer.WekaClassifierGenerator interface and a StateToFeatureVectorGenerator to turn BURLAP State objects into feature vectors usable by Weka.

This class also provides the static method getKNNTrainer(burlap.behavior.singleagent.vfa.StateToFeatureVectorGenerator, int) for construction of the Weka instance-based regression algorithm IBk, since instance-based methods have convergence guarantees for fitted value iteration.

Author:

James MacGlashan.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	WekaVFATrainer.WekaClassifierGenerator An interface for generating Weka Classifier objects to use for training.
static class	WekaVFATrainer.WekaVFA A class for predicting state values by using a trained Weka Classifier.

Nested classes/interfaces inherited from interface burlap.behavior.singleagent.planning.vfa.fittedvi.SupervisedVFA

SupervisedVFA.SupervisedVFAInstance

Field Summary

Fields

Modifier and Type	Field and Description
protected WekaVFATrainer.WekaClassifierGenerator	baseClassifier The generator of the Weka Classifier to use for training.
protected StateToFeatureVectorGenerator	fvGen The StateToFeatureVectorGenerator used to convert BURLAP State objects to feature vectors.

Constructor Summary

Constructors

Constructor and Description
WekaVFATrainer(WekaVFATrainer.WekaClassifierGenerator baseClassifier, StateToFeatureVectorGenerator fvGen) Initializes.

Method Summary

Methods

Modifier and Type	Method and Description
static WekaVFATrainer	getKNNTrainer(StateToFeatureVectorGenerator fvGen, int k) Creates a standard supervised VFA trainer that uses Weka's IBk instance-based algorithm with a KD-tree and 1-distance similarity measure.
ValueFunction	train(java.util.List<SupervisedVFA.SupervisedVFAInstance> trainingData) Uses supervised learning (regression) to learn a value function approximation of the input training data.

Methods inherited from class java.lang.Object

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

Field Detail

baseClassifier

protected WekaVFATrainer.WekaClassifierGenerator baseClassifier

The generator of the Weka Classifier to use for training.

fvGen

protected StateToFeatureVectorGenerator fvGen

The StateToFeatureVectorGenerator used to convert BURLAP State objects to feature vectors.

Constructor Detail

WekaVFATrainer

public WekaVFATrainer(WekaVFATrainer.WekaClassifierGenerator baseClassifier,
              StateToFeatureVectorGenerator fvGen)

Initializes.

Parameters:

baseClassifier - The generator of the Weka Classifier to use for training.

fvGen - The StateToFeatureVectorGenerator used to convert BURLAP State objects to feature vectors.

Method Detail

train

public ValueFunction train(java.util.List<SupervisedVFA.SupervisedVFAInstance> trainingData)

Description copied from interface: SupervisedVFA

Uses supervised learning (regression) to learn a value function approximation of the input training data.

Specified by:

train in interface SupervisedVFA

Parameters:

trainingData - the training data to fit.

Returns:

a ValueFunction that fits the training data.

getKNNTrainer

public static WekaVFATrainer getKNNTrainer(StateToFeatureVectorGenerator fvGen,
                           int k)

Creates a standard supervised VFA trainer that uses Weka's IBk instance-based algorithm with a KD-tree and 1-distance similarity measure.

Parameters:

fvGen - the StateToFeatureVectorGenerator for converting the BURLAP state into a feature vector usable by Weka.

k - the number of nearest neighbors uses in the regression algorithm.

Returns:

the WekaVFATrainer for Weka's IBk algorithm.