Package es.uam.eps.ir.relison.links.linkprediction.metrics

Class AbstractClassificationMetric<U>

java.lang.Object
es.uam.eps.ir.relison.links.linkprediction.metrics.AbstractClassificationMetric<U>
Type Parameters:
U - type of the users.
All Implemented Interfaces:
LinkPredictionMetric<U>
Direct Known Subclasses:
Accuracy, F1Score, Precision, Recall
public abstract class AbstractClassificationMetric<U>
extends java.lang.Object
implements LinkPredictionMetric<U>
Implementation of the area under the receiver-operating characteristic curve for link prediction.

  • Field Summary

    Fields 
    Modifier and Type Field Description
    private int cutoff
    The number of items in the ranking to consider.
    private double threshold
    The classification threshold (values below this will be classified as negative).
    private boolean useCutoff
    True if we use the cutoff of the prediction to differentiate between positive and negative examples, false if we use a threshold value.

  • Constructor Summary

    Constructors 
    Constructor Description
    AbstractClassificationMetric​(double threshold)
    Constructor.
    AbstractClassificationMetric​(int cutoff)
    Constructor.

  • Method Summary

    Modifier and Type Method Description
    protected abstract double compute​(long size, long truePos, long trueNeg, long falsePos, long falseNeg)
    Computes the actual classification metric.
    double evaluate​(Graph<U> graph, Prediction<U> pred)
    Evaluates a link prediction algorithm.
    double evaluate​(Graph<U> graph, Prediction<U> pred, java.util.function.Predicate<Pair<U>> filter)
    Evaluates a link prediction algorithm.
    double evaluate​(Graph<U> graph, Prediction<U> pred, java.util.stream.Stream<Pair<U>> pairs)
    Evaluates a link prediction algorithm.

    Methods inherited from class java.lang.Object

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

  • Field Details

    • cutoff

      private final int cutoff
      The number of items in the ranking to consider.

    • threshold

      private final double threshold
      The classification threshold (values below this will be classified as negative).

    • useCutoff

      private final boolean useCutoff
      True if we use the cutoff of the prediction to differentiate between positive and negative examples, false if we use a threshold value.

  • Constructor Details

    • AbstractClassificationMetric

      public AbstractClassificationMetric​(int cutoff)
      Constructor.
      Parameters:
      cutoff - the number of items in the ranking to consider. If cutoff is smaller or equal than 0, we consider that the links in the ranking are all the seen objects.

    • AbstractClassificationMetric

      public AbstractClassificationMetric​(double threshold)
      Constructor.
      Parameters:
      threshold - the minimum threshold. Pairs given a value greater or equal than the threshold will be considered as if they were positively classified. The rest, as negative.

  • Method Details

    • evaluate

      public double evaluate​(Graph<U> graph, Prediction<U> pred)
      Description copied from interface: LinkPredictionMetric
      Evaluates a link prediction algorithm. It considers that the prediction is applied over all pairs of users in the network.
      Specified by:
      evaluate in interface LinkPredictionMetric<U>
      Parameters:
      graph - a graph containing the set of links to predict.
      pred - the prediction.
      Returns:
      the value of the metric.

    • evaluate

      public double evaluate​(Graph<U> graph, Prediction<U> pred, java.util.function.Predicate<Pair<U>> filter)
      Description copied from interface: LinkPredictionMetric
      Evaluates a link prediction algorithm. It considers that the prediction is only applied for a subset of pairs of users in the network: those who pass a given filter.
      Specified by:
      evaluate in interface LinkPredictionMetric<U>
      Parameters:
      graph - a graph containing the set of links to predict.
      pred - the prediction.
      filter - the filter for the different pairs of users.
      Returns:
      the value of the metric.

    • evaluate

      public double evaluate​(Graph<U> graph, Prediction<U> pred, java.util.stream.Stream<Pair<U>> pairs)
      Description copied from interface: LinkPredictionMetric
      Evaluates a link prediction algorithm. It considers that we only apply this over a given set of links.
      Specified by:
      evaluate in interface LinkPredictionMetric<U>
      Parameters:
      graph - a graph containing the set of links to predict.
      pred - the prediction.
      pairs - the set of users to apply the recommendation to.
      Returns:
      the value of the metric.

    • compute

      protected abstract double compute​(long size, long truePos, long trueNeg, long falsePos, long falseNeg)
      Computes the actual classification metric.
      Parameters:
      size - the total number of links to consider.
      truePos - the number of correctly classified positive links.
      trueNeg - the number of correctly classified negative links.
      falsePos - the number of wrongly classified positive links.
      falseNeg - the number of wrongly classified negative links.
      Returns:
      the value of the metric.