org.grouplens.lenskit.eval.metrics.predict.NDCGPredictMetric Maven / Gradle / Ivy
Show all versions of lenskit-eval Show documentation
/*
* LensKit, an open source recommender systems toolkit.
* Copyright 2010-2014 LensKit Contributors. See CONTRIBUTORS.md.
* Work on LensKit has been funded by the National Science Foundation under
* grants IIS 05-34939, 08-08692, 08-12148, and 10-17697.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as
* published by the Free Software Foundation; either version 2.1 of the
* License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
* FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
* details.
*
* You should have received a copy of the GNU General Public License along with
* this program; if not, write to the Free Software Foundation, Inc., 51
* Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*/
package org.grouplens.lenskit.eval.metrics.predict;
import it.unimi.dsi.fastutil.longs.LongIterator;
import it.unimi.dsi.fastutil.longs.LongList;
import org.grouplens.lenskit.Recommender;
import org.grouplens.lenskit.eval.Attributed;
import org.grouplens.lenskit.eval.data.traintest.TTDataSet;
import org.grouplens.lenskit.eval.metrics.AbstractMetric;
import org.grouplens.lenskit.eval.metrics.ResultColumn;
import org.grouplens.lenskit.eval.traintest.TestUser;
import org.grouplens.lenskit.util.statistics.MeanAccumulator;
import org.grouplens.lenskit.vectors.SparseVector;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import static java.lang.Math.log;
/**
* Evaluate a recommender's predictions with normalized discounted cumulative gain.
*
* This is a prediction evaluator that uses base-2 nDCG to evaluate recommender
* accuracy. The items are ordered by predicted preference and the nDCG is
* computed using the user's real rating as the gain for each item. Doing this
* only over the queried items, rather than in the general recommend condition,
* avoids penalizing recommenders for recommending items that would be better
* if the user had known about them and provided ratings (e.g., for doing their
* job).
*
*
nDCG is computed per-user and then averaged over all users.
*
* @author GroupLens Research
*/
public class NDCGPredictMetric extends AbstractMetric {
private static final Logger logger = LoggerFactory.getLogger(NDCGPredictMetric.class);
public NDCGPredictMetric() {
super(Result.class, Result.class);
}
@Override
public MeanAccumulator createContext(Attributed algo, TTDataSet ds, Recommender rec) {
return new MeanAccumulator();
}
/**
* Compute the DCG of a list of items with respect to a value vector.
*/
static double computeDCG(LongList items, SparseVector values) {
final double lg2 = log(2);
double gain = 0;
int rank = 0;
LongIterator iit = items.iterator();
while (iit.hasNext()) {
final long item = iit.nextLong();
final double v = values.get(item);
rank++;
if (rank < 2) {
gain += v;
} else {
gain += v * lg2 / log(rank);
}
}
return gain;
}
@Override
public Result doMeasureUser(TestUser user, MeanAccumulator context) {
SparseVector predictions = user.getPredictions();
if (predictions == null) {
return null;
}
SparseVector ratings = user.getTestRatings();
LongList ideal = ratings.keysByValue(true);
LongList actual = predictions.keysByValue(true);
double idealGain = computeDCG(ideal, ratings);
double gain = computeDCG(actual, ratings);
double score = gain / idealGain;
context.add(score);
return new Result(score);
}
@Override
protected Result getTypedResults(MeanAccumulator context) {
return new Result(context.getMean());
}
public static class Result {
@ResultColumn("nDCG")
public final double utility;
public Result(double util) {
utility = util;
}
}
}