[EXPERTS] Allow exact expert weighting (#9)

ldss-hse · May 9, 2022 · cfd07e8 · cfd07e8
1 parent 6954478
commit cfd07e8
Show file tree

Hide file tree

Showing 6 changed files with 50 additions and 21 deletions.
diff --git a/src/main/java/dss/lingvo/samples/TT2HFLTSCoordinator.java b/src/main/java/dss/lingvo/samples/TT2HFLTSCoordinator.java
@@ -76,7 +76,7 @@ private void processNumericSample() {
 
     private void processWeightsGenerationSample() {
         float[] expDistr = {0.8f, 0.2f};
-        float[] resVector = TTUtils.calculateWeightsVector(expDistr, 4);
+        float[] resVector = TTUtils.calculateWeightsVectorFromDistribution(expDistr, 4);
         System.out.println(Arrays.toString(resVector));
 
         float sum = 0f;
@@ -126,21 +126,11 @@ private void processMultiLevelAdvancedSample(File inputFile, File outputDirector
                 tt2HFLTSMHTWOWAMultiLevelOperator.transposeByAbstractionLevel(model.getAbstractionLevels().size(),
                         model.getAlternatives().size(), model.getExperts().size(), allByLevel);
 
-        int numExperts = model.getExpertWeightsRule().values().size();
-        float[] a = new float[numExperts];
-        float curMax = 0f;
-        for (Map.Entry<String, Float> e : model.getExpertWeightsRule().entrySet()) {
-            if (e.getKey().equals("1")) {
-                curMax = e.getValue();
-                break;
-            }
-        }
-        a[0] = curMax;
-        if (numExperts > 1) {
-            a[1] = 1 - curMax;
-        }
+        float[] weights = TTUtils.calculateWeights(model);
+        System.out.println("\n\n\n[MULTILEVEL] Weights: " + Arrays.toString(weights));
         List<ArrayList<TT2HFLTS>> altToLevel = tt2HFLTSMHTWOWAMultiLevelOperator.aggregateByExpert(
-                model.getAbstractionLevels().size(), model.getAlternatives().size(), targetScaleSize, allByExpert, a);
+                model.getAbstractionLevels().size(), model.getAlternatives().size(), targetScaleSize, allByExpert,
+                weights);
 
         List<TT2HFLTS> altVec = tt2HFLTSMHTWOWAMultiLevelOperator.aggregateFinalAltEst(targetScaleSize, altToLevel);
 
@@ -151,7 +141,6 @@ private void processMultiLevelAdvancedSample(File inputFile, File outputDirector
         Path outputJSONFilePath = Paths.get(outputDirectory.toString(), "result.json");
         TTJSONUtils.getInstance().writeResultToJSON(outputJSONFilePath, res);
 
-
         // printing to console
         List<Pair<String, TT2HFLTS>> resZippedVec = IntStream.range(0, altVec.size()).mapToObj(i -> new Pair<>(model.getAlternatives().get(i).getAlternativeID(), altVec.get(i))).collect(Collectors.toList());
 

diff --git a/src/main/java/dss/lingvo/t2hflts/multilevel/TT2HFLTSMHTWOWAMultiLevelOperator.java b/src/main/java/dss/lingvo/t2hflts/multilevel/TT2HFLTSMHTWOWAMultiLevelOperator.java
@@ -68,7 +68,7 @@ public List<ArrayList<TT2HFLTS>> aggregateByExpert(int levelsSize,
                                                        int altSize,
                                                        int targetScaleSize,
                                                        List<ArrayList<ArrayList<TT2HFLTS>>> all,
-                                                       float[] distribution) {
+                                                       float[] weights) {
         TT2HFLTSMHTWAOperator tt2HFLTSMHTWAOperator = new TT2HFLTSMHTWAOperator();
         List<ArrayList<TT2HFLTS>> levelEstimates = new ArrayList<>();
 
@@ -82,7 +82,6 @@ public List<ArrayList<TT2HFLTS>> aggregateByExpert(int levelsSize,
         // fill matrix
         for (int levelIndex = 0; levelIndex < levelsSize; levelIndex++) {
             for (int altIndex = 0; altIndex < altSize; altIndex++) {
-                float[] weights = TTUtils.calculateWeightsVector(distribution, all.get(levelIndex).get(altIndex).size());
                 List<TT2HFLTS> newSet = TTUtils.sortTT2HFLTS(all.get(levelIndex).get(altIndex), true);
                 TT2HFLTS aggRes = tt2HFLTSMHTWAOperator.calculate(newSet, weights, targetScaleSize);
                 levelEstimates.get(altIndex).set(levelIndex, aggRes);

diff --git a/src/main/java/dss/lingvo/utils/TTUtils.java b/src/main/java/dss/lingvo/utils/TTUtils.java
@@ -207,6 +207,37 @@ public static TTJSONOutputModel prepareAllResultsForJSON(List<TT2HFLTS> altOvera
         return ttjsonOutputModel;
     }
 
+
+    public static float[] calculateWeights(TTCommonInputModel model) {
+        float[] weights = new float[model.getExperts().size()];
+        if (model.getExpertWeights() != null) {
+            // if particular weights are defined in the task description there is no need to
+            // automatically assign weights
+            int index = 0;
+            for (TTExpertModel expert: model.getExperts()) {
+                weights[index++] = model.getExpertWeights().get(expert.getExpertID());
+            }
+
+            return weights;
+        }
+
+        float curMax = 0f;
+        for (Map.Entry<String, Float> e : model.getExpertWeightsRule().entrySet()) {
+            if (e.getKey().equals("1")) {
+                curMax = e.getValue();
+                break;
+            }
+        }
+
+        int numExperts = model.getExperts().size();
+        float[] distribution = new float[numExperts];
+        distribution[0] = curMax;
+        if (numExperts > 1) {
+            distribution[1] = 1 - curMax;
+        }
+        return calculateWeightsVectorFromDistribution(distribution, numExperts);
+    }
+
     /**
      * When calculating weights for the experts, we introduce a new formula
      * E.g. distribution is [0.8, 0.2] and we have 3 experts.
@@ -218,7 +249,7 @@ public static TTJSONOutputModel prepareAllResultsForJSON(List<TT2HFLTS> altOvera
      * @param numExperts   - number of experts
      * @return vector of weights
      */
-    public static float[] calculateWeightsVector(float[] distribution, int numExperts) {
+    public static float[] calculateWeightsVectorFromDistribution(float[] distribution, int numExperts) {
         float[] res = new float[numExperts];
         for (int i = 0; i < numExperts; i++) {
             res[i] = calculateWeight(distribution[0], i, numExperts);

diff --git a/src/main/java/dss/lingvo/utils/models/input/common/TTCommonInputModel.java b/src/main/java/dss/lingvo/utils/models/input/common/TTCommonInputModel.java
@@ -15,6 +15,8 @@ public class TTCommonInputModel {
     private Map<String, List<TTExpertEstimationsModel>> estimations;
     @JsonProperty("abstractionLevelWeights")
     private Map<String, Float> abstractionLevelWeights;
+    @JsonProperty("expertWeights")
+    private Map<String, Float> expertWeights;
     @JsonProperty("expertWeightsRule")
     private Map<String, Float> expertWeightsRule;
     @JsonProperty("criteriaWeightsPerGroup")
@@ -76,6 +78,14 @@ public void setExpertWeightsRule(Map<String, Float> expertWeightsRule) {
         this.expertWeightsRule = expertWeightsRule;
     }
 
+    public Map<String, Float> getExpertWeights() {
+        return expertWeights;
+    }
+
+    public void setExpertWeights(Map<String, Float> expertWeights) {
+        this.expertWeights = expertWeights;
+    }
+
     public Map<String, List<Float>> getCriteriaWeightsPerGroup() {
         return criteriaWeightsPerGroup;
     }

diff --git a/src/test/java/dss/lingvo/t2hflts/multilevel/TT2HFLTSMHTWOWAMultiLevelOperatorTest.java b/src/test/java/dss/lingvo/t2hflts/multilevel/TT2HFLTSMHTWOWAMultiLevelOperatorTest.java
@@ -49,7 +49,7 @@ public static void runOnceBeforeClass() throws IOException {
                         model.getExperts().size(),
                         allByLevel);
 
-        float []a = {0.8f, 0.2f};
+        float[] a = TTUtils.calculateWeights(model);
         altToLevel = tt2HFLTSMHTWOWAMultiLevelOperator
                 .aggregateByExpert(model.getAbstractionLevels().size(),
                         model.getAlternatives().size(),

diff --git a/src/test/java/dss/lingvo/utils/TTUtilsTest.java b/src/test/java/dss/lingvo/utils/TTUtilsTest.java
@@ -215,7 +215,7 @@ public void testSortTT2HFLTS3() throws Exception {
     @Test
     public void testWeightsDistribution() throws Exception {
         float[] expDistr = {0.8f, 0.2f};
-        float[] resVector = TTUtils.calculateWeightsVector(expDistr, 4);
+        float[] resVector = TTUtils.calculateWeightsVectorFromDistribution(expDistr, 4);
         float[] expV = {0.8f, 0.16f, 0.03199997f, 0.008000016f};
         float sum = 0f;
         for (int i = 0; i < resVector.length; i++) {