LearnFunctionAndRankCli.java

package io.gitlab.chaver.minimax.cli;

import com.google.gson.Gson;
import io.gitlab.chaver.minimax.io.Alternative;
import io.gitlab.chaver.minimax.io.IAlternative;
import io.gitlab.chaver.minimax.learn.oracle.LinearFunctionOracle;
import io.gitlab.chaver.minimax.learn.oracle.OWAOracle;
import io.gitlab.chaver.minimax.learn.oracle.ScoreFunctionOracle;
import io.gitlab.chaver.minimax.learn.train.AbstractRankingLearning;
import io.gitlab.chaver.minimax.learn.train.passive.AHPRankLearn;
import io.gitlab.chaver.minimax.learn.train.passive.KappalabRankLearn;
import io.gitlab.chaver.minimax.learn.train.passive.SVMRankLearn;
import io.gitlab.chaver.minimax.ranking.*;
import io.gitlab.chaver.minimax.rules.io.RuleWithMeasures;
import io.gitlab.chaver.minimax.score.FunctionParameters;
import io.gitlab.chaver.minimax.score.IScoreFunction;
import io.gitlab.chaver.minimax.score.ScoreFunctionFactory;
import io.gitlab.chaver.minimax.util.RandomUtil;
import picocli.CommandLine;
import picocli.CommandLine.Option;

import java.io.*;
import java.util.*;
import java.util.concurrent.Callable;
import java.util.stream.Collectors;

import static io.gitlab.chaver.minimax.learn.train.LearnUtil.*;

public class LearnFunctionAndRankCli implements Callable<Integer> {

    @Option(names = "-d", description = "Path of the rules data", required = true)
    private String dataPath;
    @Option(names = "--tt", description = "Path of the training/test data", required = true)
    private String trainingTestDataPath;
    @Option(names = "-m", description = "Measures used in the function", required = true, split = ":")
    private String[] measures;
    @Option(names = "-o", description = "Name of the oracle", required = true)
    private String oracleName;
    @Option(names = "-l", description = "Learning algorithm", required = true)
    private String learnAlgorithm;
    @Option(names = "--seed", description = "Seed for random number generation")
    private long seed = 2994274L;
    @Option(names = "-r", description = "Path of the result files", required = true)
    private String resPath;

    // Kappalab parameters
    @Option(names = "--delta", description = "Delta parameter (kappalab, default value : 1e-5)")
    private double delta = 1e-5;
    @Option(names = "--kadd", description = "k-additivity of the model (kappalab, default value : 2)")
    private int kAdd = 2;
    @Option(names = "--sigf", description = "Number of significant figures (kappalab, default value : 3)")
    private int sigf = 3;

    // SVM parameters
    @Option(names = "-c", description = "Regularisation parameter (svm, default value : 0.01)")
    private double regularisationParameter = 0.01;


    private List<RuleWithMeasures> readRules(String path) throws IOException {
        try (BufferedReader reader = new BufferedReader(new FileReader(path))) {
            Gson gson = new Gson();
            List<RuleWithMeasures> rules = new ArrayList<>();
            String line;
            while ((line = reader.readLine()) != null) {
                rules.add(gson.fromJson(line, RuleWithMeasures.class));
            }
            return rules;
        }
    }

    private void writeRules(String path, List<RuleWithMeasures> rules) throws IOException {
        try (BufferedWriter writer = new BufferedWriter(new FileWriter(path))) {
            Gson gson = new Gson();
            for (RuleWithMeasures r : rules) {
                writer.write(gson.toJson(r) + "\n");
            }
        }
    }

    private void writeObject(String path, Object o) throws IOException {
        try (BufferedWriter writer = new BufferedWriter(new FileWriter(path))) {
            Gson gson = new Gson();
            writer.write(gson.toJson(o));
        }
    }

    private List<IAlternative> getAlternatives(List<RuleWithMeasures> rules) {
        List<IAlternative> alternatives = new ArrayList<>();
        for (RuleWithMeasures r : rules) {
            double[] vector = Arrays
                    .stream(measures)
                    .mapToDouble(m -> r.getMeasureValues().get(m))
                    .toArray();
            alternatives.add(new Alternative(vector));
        }
        return alternatives;
    }

    private Comparator<IAlternative> getOracle(double[] weights) {
        if (oracleName.equals(OWAOracle.TYPE)) {
            return new OWAOracle(weights);
        }
        if (oracleName.equals(LinearFunctionOracle.TYPE)) {
            return new LinearFunctionOracle(weights);
        }
        throw new RuntimeException("Wrong oracle type: " + oracleName);
    }

    private AbstractRankingLearning getLearningAlgo(Ranking<IAlternative> expectedRanking) {
        if (learnAlgorithm.equals("kappalab")) {
            KappalabRankLearn kappalab = new KappalabRankLearn(expectedRanking);
            kappalab.setDelta(delta);
            kappalab.setKAdditivity(kAdd);
            kappalab.setSigf(sigf);
            return kappalab;
        }
        if (learnAlgorithm.equals("ahp")) {
            AHPRankLearn ahp = new AHPRankLearn(expectedRanking);
            return ahp;
        }
        if (learnAlgorithm.equals("svm")) {
            SVMRankLearn svm = new SVMRankLearn(expectedRanking);
            svm.setRegularisationParameter(regularisationParameter);
            return svm;
        }
        throw new RuntimeException("Wrong learning algorithm: " + learnAlgorithm);
    }

    private List<RankingMetric> getRankingMetrics(int nbRules) {
        int top1 = (int) (0.01 * nbRules);
        int top10 = (int) (0.1 * nbRules);
        return Arrays.asList(new KendallConcordanceCoeff(), new SpearmanRankCorrelationCoefficient(),
                new RecallMetric(top1), new RecallMetric(top10), new AveragePrecision(top1), new AveragePrecision(top10));
    }

    private Map<String, String> getRankingMetricLabels(int nbRules) {
        int top1 = (int) (0.01 * nbRules);
        int top10 = (int) (0.1 * nbRules);
        Map<String, String> labels = new HashMap<>();
        labels.put(KendallConcordanceCoeff.TYPE, KendallConcordanceCoeff.TYPE);
        labels.put(SpearmanRankCorrelationCoefficient.TYPE, SpearmanRankCorrelationCoefficient.TYPE);
        labels.put(RecallMetric.TYPE + "@" + top1, RecallMetric.TYPE + "@1%");
        labels.put(RecallMetric.TYPE + "@" + top10, RecallMetric.TYPE + "@10%");
        labels.put(AveragePrecision.TYPE + "@" + top1, AveragePrecision.TYPE + "@1%");
        labels.put(AveragePrecision.TYPE + "@" + top10, AveragePrecision.TYPE + "@10%");
        return labels;
    }

    private Map<String, Double> computeRankingMetricValues(List<RankingMetric> rankingMetrics,
                                                           Map<String, String> rankingMetricLabels,
                                                           Ranking<IAlternative> actualRanking,
                                                           Ranking<IAlternative> expectedRanking) {
        Map<String, Double> rankingMetricValues = new HashMap<>();
        for (RankingMetric metric : rankingMetrics) {
            double value = metric.compute(expectedRanking, actualRanking);
            String label = rankingMetricLabels.get(metric.getName());
            rankingMetricValues.put(label, value);
        }
        return rankingMetricValues;
    }

    @Override
    public Integer call() throws Exception {
        int nbTransactions = getNbTransactions(dataPath + "_prop.jsonl");
        int nbMeasures = measures.length;
        List<RuleWithMeasures> trainingRules = readRules(trainingTestDataPath + "_train.jsonl");
        List<RuleWithMeasures> testRules = readRules(trainingTestDataPath + "_test.jsonl");
        List<IAlternative> trainingAlternatives = getAlternatives(trainingRules);
        List<IAlternative> testAlternatives = getAlternatives(testRules);
        RandomUtil.getInstance().setSeed(seed);
        double[] randomWeights = RandomUtil.getInstance().generateRandomWeights(nbMeasures);
        Comparator<IAlternative> oracle = getOracle(randomWeights);
        Ranking<IAlternative> expectedRanking = computeRankingWithOracle(oracle, trainingAlternatives);
        AbstractRankingLearning algo = getLearningAlgo(expectedRanking);
        FunctionParameters functionParameters = algo.learn();
        functionParameters.addWeightLabels(measures);
        functionParameters.setMeasureNames(measures);
        IScoreFunction<IAlternative> func = ScoreFunctionFactory.getScoreFunction(functionParameters);
        Ranking<IAlternative> actualTestAlternativesRanking = computeRankingWithOracle(new ScoreFunctionOracle(func), testAlternatives);
        Ranking<IAlternative> expectedTestAlternativesRanking = computeRankingWithOracle(oracle, testAlternatives);
        List<RankingMetric> rankingMetrics = getRankingMetrics(testRules.size());
        Map<String, String> rankingMetricLabels = getRankingMetricLabels(testRules.size());
        Map<String, Double> rankingMetricValues = computeRankingMetricValues(rankingMetrics, rankingMetricLabels,
                actualTestAlternativesRanking, expectedTestAlternativesRanking);
        List<RuleWithMeasures> testRulesRankedWithLearnedFunc = Arrays
                .stream(actualTestAlternativesRanking.getRanking())
                .mapToObj(i -> testRules.get(i))
                .collect(Collectors.toCollection(ArrayList::new));
        for (int i = 0; i < testRulesRankedWithLearnedFunc.size(); i++) {
            int pos = actualTestAlternativesRanking.getRanking()[i];
            IAlternative a = testAlternatives.get(pos);
            testRulesRankedWithLearnedFunc.get(i).setScore(func.computeScore(a));
        }
        writeRules(resPath + "_ordered_test_rules.jsonl", testRulesRankedWithLearnedFunc);
        writeObject(resPath + "_func.jsonl", functionParameters);
        writeObject(resPath + "_metrics.jsonl", rankingMetricValues);
        return 0;
    }

    public static void main(String[] args) {
        int exitCode = new CommandLine(new LearnFunctionAndRankCli()).execute(args);
        System.exit(exitCode);
    }
}