org.encog.ml.importance

Class CorrelationFeatureImportanceCalc

java.lang.Object

org.encog.ml.importance.AbstractFeatureImportance

org.encog.ml.importance.CorrelationFeatureImportanceCalc

All Implemented Interfaces:

FeatureImportance

public class CorrelationFeatureImportanceCalc
extends AbstractFeatureImportance

Compute feature importance using correlation between each of the features and the target. This ranker should only be used with datasets that are either regression or binary classifiers. If there are multiple outputs, this ranker will throw an error. Also, if there are are more than 2 classes encoded into a single output neuron (e.g. a SVM designed for more than two classes), this ranker should not be used. This ranker does not require a trained model, it simply looks at the training data and determines how closely each of the inputs correlates to the target. https://en.wikipedia.org/wiki/Correlation_and_dependence

Constructor Summary

Constructors

Constructor and Description
CorrelationFeatureImportanceCalc()

Method Summary

Modifier and Type	Method and Description
void	performRanking() Perform the ranking, without using a specific training set.
void	performRanking(MLDataSet theDataset) Perform the ranking, using a specific training set.

Methods inherited from class org.encog.ml.importance.AbstractFeatureImportance

getFeatures, getFeaturesSorted, getModel, init, toString

Methods inherited from class java.lang.Object

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

Constructor Detail

CorrelationFeatureImportanceCalc

public CorrelationFeatureImportanceCalc()

Method Detail

performRanking

public void performRanking()

Perform the ranking, without using a specific training set. Not all ranking algorithms support this.

performRanking

public void performRanking(MLDataSet theDataset)

Perform the ranking, using a specific training set. Not all ranking algorithms can make use of a dataset.

Parameters:

theDataset - The dataset.