Add new oracles

instructions_API.md

@@ -66,8 +66,9 @@ L'usage de `-d`, `-m` et `--seed` est le même que précédemment (ces paramètr
 
 - `--tt` : correspond au chemin des fichiers d'entraînement de test du modèle (i.e. la valeur correspondante à `-r` lors de l'appel de l'API précédente)
 - `-o` : nom de l'oracle utilisé, pour l'instant les valeurs suivantes sont possibles :
-  - **linear** : fonction linéaire (somme pondérée)
+  - **choquetPearson** : Intégrale de Choquet calculée avec les corrélations de Pearson entre les mesures
   - **owa** : Ordered Weighted Average
+  - **chiSquared** : Chi-Squared test statistique
 - `-l` : nom de l'algorithme d'apprentissage utilisé pour apprendre le modèle, les valeurs suivantes sont possibles :
   - **kappalab**
   - **ahp**

src/main/java/io/gitlab/chaver/minimax/cli/LearnFunctionAndRankCli.java

@@ -3,9 +3,7 @@ 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.oracle.*;
 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;
@@ -16,6 +14,7 @@ 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 io.gitlab.chaver.mining.rules.io.IRule;
 import picocli.CommandLine;
 import picocli.CommandLine.Option;
 
@@ -96,13 +95,26 @@ public class LearnFunctionAndRankCli implements Callable<Integer> {
         return alternatives;
     }
 
-    private Comparator<IAlternative> getOracle(double[] weights) {
+    private Comparator<IAlternative> getOracle(double[] weights, List<IAlternative> trainingAlternatives,
+                                               List<IAlternative> testAlternatives, List<RuleWithMeasures> trainingRules,
+                                               List<RuleWithMeasures> testRules, int nbTransactions) {
         if (oracleName.equals(OWAOracle.TYPE)) {
             return new OWAOracle(weights);
         }
         if (oracleName.equals(LinearFunctionOracle.TYPE)) {
             return new LinearFunctionOracle(weights);
         }
+        if (oracleName.equals("choquetPearson")) {
+            return new ChoquetOracle(
+                    new CorrelationChoquetFuncBuilder(weights, trainingAlternatives.toArray(new IAlternative[0])).getCapacity()
+            );
+        }
+        if (oracleName.equals("chiSquared")) {
+            ChiSquaredOracle2 oracle = new ChiSquaredOracle2(nbTransactions);
+            oracle.addAlternativesRules(trainingAlternatives, trainingRules.stream().map(i -> (IRule) i).collect(Collectors.toList()));
+            oracle.addAlternativesRules(testAlternatives, testRules.stream().map(i -> (IRule) i).collect(Collectors.toList()));
+            return oracle;
+        }
         throw new RuntimeException("Wrong oracle type: " + oracleName);
     }
 
@@ -181,7 +193,7 @@ public class LearnFunctionAndRankCli implements Callable<Integer> {
         List<IAlternative> testAlternatives = getAlternatives(testRules);
         RandomUtil.getInstance().setSeed(seed);
         double[] randomWeights = RandomUtil.getInstance().generateRandomWeights(nbMeasures);
-        Comparator<IAlternative> oracle = getOracle(randomWeights);
+        Comparator<IAlternative> oracle = getOracle(randomWeights, trainingAlternatives, testAlternatives, trainingRules, testRules, nbTransactions);
         Ranking<IAlternative> expectedRanking = computeRankingWithOracle(oracle, trainingAlternatives);
         AbstractRankingLearning algo = getLearningAlgo(expectedRanking);
         FunctionParameters functionParameters = algo.learn();

src/main/java/io/gitlab/chaver/minimax/learn/oracle/ChiSquaredOracle2.java

@@ -22,6 +22,16 @@ public class ChiSquaredOracle2 extends ScoreOracle {
         func = new ChiSquaredScoreFunction(nbTransactions);
     }
 
+    public ChiSquaredOracle2(int nbTransactions) {
+        func = new ChiSquaredScoreFunction(nbTransactions);
+    }
+
+    public void addAlternativesRules(List<IAlternative> alternatives, List<IRule> rules) {
+        for (int i = 0; i < alternatives.size(); i++) {
+            mapAlternativeToRule.put(alternatives.get(i), rules.get(i));
+        }
+    }
+
     @Override
     public double computeScore(IAlternative a) {
         return func.computeScore(mapAlternativeToRule.get(a));

src/main/java/io/gitlab/chaver/minimax/rules/io/RuleWithMeasures.java

@@ -9,16 +9,22 @@ import java.util.HashMap;
 import java.util.Map;
 
 @Getter
-public class RuleWithMeasures {
+public class RuleWithMeasures implements IRule {
 
     private int[] x;
     private int[] y;
     private Map<String, Double> measureValues;
     private @Setter Double score;
+    private int freqX;
+    private int freqZ;
+    private int freqY;
 
     public RuleWithMeasures(IRule r, IAlternative a, String[] measures) {
         x = r.getX();
         y = r.getY();
+        freqX = r.getFreqX();
+        freqY = r.getFreqY();
+        freqZ = r.getFreqZ();
         measureValues = new HashMap<>();
         for (int i = 0; i < measures.length; i++) {
             measureValues.put(measures[i], a.getVector()[i]);
@@ -28,6 +34,9 @@ public class RuleWithMeasures {
     public RuleWithMeasures(RuleWithMeasures copy, boolean withScore) {
         x = copy.getX();
         y = copy.getY();
+        freqX = copy.getFreqX();
+        freqY = copy.getFreqY();
+        freqZ = copy.getFreqZ();
         measureValues = copy.getMeasureValues();
         score = withScore ? copy.score : null;
     }

src/test/java/io/gitlab/chaver/minimax/cli/LearnFunctionAndRankCliTest.java

@@ -19,7 +19,45 @@ class LearnFunctionAndRankCliTest {
         String[] args = {"-d", rulesPath, "--train", trainingPercentage, "-r", resPath, "-m", measures,
                 "--seed", seed};
         new CommandLine(new SplitTrainingTestCli()).execute(args);
-        String oracleName = "linear";
+        String oracleName = "owa";
+        String learningAlgorithm = "kappalab";
+        String learnResPath = File.createTempFile("learn", "").getAbsolutePath();
+        args = new String[]{"-d", rulesPath, "-r", learnResPath, "-m", measures,
+                "--seed", seed, "-o", oracleName, "-l", learningAlgorithm, "--tt", resPath};
+        int exitCode = new CommandLine(new LearnFunctionAndRankCli()).execute(args);
+        assertEquals(0, exitCode);
+    }
+
+    @Test
+    void test2() throws Exception {
+        String rulesPath = "results/rules/iris";
+        String trainingPercentage = "0.26";
+        String resPath = File.createTempFile("rules", "").getAbsolutePath();
+        String measures = "phi:kruskal:yuleQ";
+        String seed = "1234";
+        String[] args = {"-d", rulesPath, "--train", trainingPercentage, "-r", resPath, "-m", measures,
+                "--seed", seed};
+        new CommandLine(new SplitTrainingTestCli()).execute(args);
+        String oracleName = "chiSquared";
+        String learningAlgorithm = "kappalab";
+        String learnResPath = File.createTempFile("learn", "").getAbsolutePath();
+        args = new String[]{"-d", rulesPath, "-r", learnResPath, "-m", measures,
+                "--seed", seed, "-o", oracleName, "-l", learningAlgorithm, "--tt", resPath};
+        int exitCode = new CommandLine(new LearnFunctionAndRankCli()).execute(args);
+        assertEquals(0, exitCode);
+    }
+
+    @Test
+    void test3() throws Exception {
+        String rulesPath = "results/rules/iris";
+        String trainingPercentage = "0.26";
+        String resPath = File.createTempFile("rules", "").getAbsolutePath();
+        String measures = "phi:kruskal:yuleQ";
+        String seed = "1234";
+        String[] args = {"-d", rulesPath, "--train", trainingPercentage, "-r", resPath, "-m", measures,
+                "--seed", seed};
+        new CommandLine(new SplitTrainingTestCli()).execute(args);
+        String oracleName = "choquetPearson";
         String learningAlgorithm = "kappalab";
         String learnResPath = File.createTempFile("learn", "").getAbsolutePath();
         args = new String[]{"-d", rulesPath, "-r", learnResPath, "-m", measures,