edu.cmu.minorthird.classify.ranking
Class RankingBoosted
java.lang.Object
edu.cmu.minorthird.classify.BatchClassifierLearner
edu.cmu.minorthird.classify.BatchBinaryClassifierLearner
edu.cmu.minorthird.classify.ranking.BatchRankingLearner
edu.cmu.minorthird.classify.ranking.RankingBoosted
- All Implemented Interfaces:
- BinaryClassifierLearner, ClassifierLearner, java.lang.Cloneable
public class RankingBoosted
- extends BatchRankingLearner
A boosted version for ranking.
An implementation of - "Michael Collins and Terry Koo, Discriminative Reranking for Natural Language Parsing.
Computational Linguistics, March 2005", see page 45.
Requirements of this class:
- Considers only binary features
- Requires a particular cont. feature named "walkerScore", that contains the original log-probability for an example
To do:
- Handle multiple positive answers.
- Automatically discretize real-value features into binary.
- It is possible to incorporate example 'importance' weights, according to some 'goodness' evaluation measure. See Collins'.
- Author:
- Einat Minkov
Methods inherited from class java.lang.Object |
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait |
RankingBoosted
public RankingBoosted()
RankingBoosted
public RankingBoosted(int numEpochs,
int exampleSize)
batchTrain
public Classifier batchTrain(Dataset data)
- Description copied from class:
BatchClassifierLearner
- subclasses should use this method to implement a batch supervised learning algorithm.
- Specified by:
batchTrain
in class BatchClassifierLearner
initialExpLoss
public double initialExpLoss(double w0,
Example[][] rankedExamples)