org.grouplens.lenskit.eval.metrics.topn.PrecisionRecallTopNMetric Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of lenskit-eval Show documentation
Facilities for evaluating recommender algorithms.
There is a newer version: 3.0-T5
/*
 * 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.topn;

import it.unimi.dsi.fastutil.longs.LongSet;
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.scored.ScoredId;
import org.hamcrest.Matchers;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.util.List;

/**
 * A metric to compute the precision and recall of a recommender given a 
 * set of candidate items to recommend from and a set of desired items.  The aggregate results are
 * means of the user results.
 * 
 * This can be used to compute metrics like fallout (probability that a 
 * recommendation is bad) by configuring bad items as the test item set.
 *
 * @author GroupLens Research
 */
public class PrecisionRecallTopNMetric extends AbstractMetric {
    private static final Logger logger = LoggerFactory.getLogger(PrecisionRecallTopNMetric.class);
    private final String prefix;
    private final String suffix;
    private final int listSize;
    private final ItemSelector candidates;
    private final ItemSelector exclude;
    private final ItemSelector goodItems;

    /**
     * Construct a new recall and precision top n metric
     * @param pre the prefix label for this evaluation, or {@code null} for no prefix.
     * @param sfx the suffix label for this evaluation, or {@code null} for no suffix.
     * @param listSize The number of recommendations to fetch.
     * @param candidates The candidate selector, provides a list of items which can be recommended
     * @param exclude The exclude selector, provides a list of items which must not be recommended 
     *                (These items are removed from the candidate items to form the final candidate set)
     * @param goodItems The list of items to consider "true positives", all other items will be treated
     *                  as "false positives".
     */
    public PrecisionRecallTopNMetric(String pre, String sfx, int listSize, ItemSelector candidates, ItemSelector exclude, ItemSelector goodItems) {
        super(Result.class, Result.class);
        prefix = pre;
        suffix = sfx;
        this.listSize = listSize;
        this.candidates = candidates;
        this.exclude = exclude;
        this.goodItems = goodItems;
    }

    @Override
    protected String getPrefix() {
        return prefix;
    }

    @Override
    protected String getSuffix() {
        return suffix;
    }

    @Override
    public Context createContext(Attributed algo, TTDataSet ds, Recommender rec) {
        return new Context();
    }

    @Override
    public Result doMeasureUser(TestUser user, Context context) {
        int tp = 0;

        LongSet items = goodItems.select(user);
        if (items.isEmpty()) {
            logger.warn("no good items for user {}", user.getUserId());
        }

        List recs = user.getRecommendations(listSize, candidates, exclude);
        if (recs == null) {
            logger.warn("no recommendations for user {}", user.getUserId());
            return null;
        }

        logger.debug("searching for {} good items among {} recommendations for {}",
                     items.size(), recs.size(), user.getUserId());
        for(ScoredId s : recs) {
            if(items.contains(s.getId())) {
                tp += 1;
            }
        }

        if (items.size() > 0 && recs.size() > 0) {
            // if both the items set and recommendations are non-empty (no division by 0).
            double precision = (double) tp / recs.size();
            double recall = (double) tp / items.size();
            context.addUser(precision, recall);
            return new Result(precision, recall);
        } else {
            return null;
        }
    }

    @Override
    protected Result getTypedResults(Context context) {
        return context.finish();
    }

    public static class Result {
        @ResultColumn("Precision")
        public final double precision;
        @ResultColumn("Recall")
        public final double recall;

        public Result(double prec, double rec) {
            precision = prec;
            recall = rec;
        }
    }

    public class Context {
        double totalPrecision = 0;
        double totalRecall = 0;
        int nusers = 0;

        private void addUser(double prec, double rec) {
            totalPrecision += prec;
            totalRecall += rec;
            nusers += 1;
        }

        public Result finish() {
            if (nusers > 0) {
                return new Result(totalPrecision / nusers, totalRecall / nusers);
            } else {
                return null;
            }
        }
    }

    /**
     * @author GroupLens Research
     */
    public static class Builder extends TopNMetricBuilder{
        private ItemSelector goodItems = ItemSelectors.testRatingMatches(Matchers.greaterThanOrEqualTo(4.0d));

        public Builder() {
            // override the default candidate items with a more reasonable set.
            setCandidates(ItemSelectors.allItems());
        }

        public ItemSelector getGoodItems() {
            return goodItems;
        }

        /**
         * Set the set of items that will be considered ‘good’ by the evaluation.
         *
         * @param goodItems A selector for good items.
         */
        public void setGoodItems(ItemSelector goodItems) {
            this.goodItems = goodItems;
        }

        @Override
        public PrecisionRecallTopNMetric build() {
            return new PrecisionRecallTopNMetric(prefix, suffix, listSize, candidates, exclude, goodItems);
        }
    }

}