Fix MostPopular, NMF. Modify configuration of HFT, WRMF, BPMF, MFALS, biasedMF, NMF.

core/src/main/java/net/librec/recommender/AbstractRecommender.java

@@ -27,7 +27,6 @@
 import net.librec.math.structure.MatrixEntry;
 import net.librec.math.structure.SparseMatrix;
 import net.librec.recommender.item.*;
-//import net.librec.util.ModelDataUtil;
 import net.librec.util.ReflectionUtil;
 import org.apache.commons.lang.StringUtils;
 import org.apache.commons.logging.Log;
@@ -186,7 +185,7 @@ protected void setup() throws LibrecException {
         Collections.sort(ratingScale);
         maxRate = Collections.max(trainMatrix.getValueSet());
         minRate = Collections.min(trainMatrix.getValueSet());
-        globalMean = trainMatrix.sum() / numRates;
+        globalMean = trainMatrix.mean();
 
         int[] numDroppedItemsArray = new int[numUsers]; // for AUCEvaluator
         int maxNumTestItemsByUser = 0; //for idcg
@@ -380,7 +379,7 @@ protected void cleanup() throws LibrecException {
      */
     @Override
     public void loadModel(String filePath) {
-//        ModelDataUtil.loadRecommenderModel(this, filePath);
+
     }
 
     /**
@@ -390,7 +389,7 @@ public void loadModel(String filePath) {
      */
     @Override
     public void saveModel(String filePath) {
-//        ModelDataUtil.saveRecommenderModel(this, filePath);
+
     }
 
     /**
@@ -460,6 +459,24 @@ public List<RecommendedItem> getRecommendedList() {
      * @throws LibrecException if error occurs
      */
     protected boolean isConverged(int iter) throws LibrecException{
-        return false;
+        float delta_loss = (float) (lastLoss - loss);
+
+        // print out debug info
+        if (verbose) {
+            String recName = getClass().getSimpleName().toString();
+            String info = recName + " iter " + iter + ": loss = " + loss + ", delta_loss = " + delta_loss;
+            LOG.info(info);
+        }
+
+        if (Double.isNaN(loss) || Double.isInfinite(loss)) {
+//            LOG.error("Loss = NaN or Infinity: current settings does not fit the recommender! Change the settings and try again!");
+            throw new LibrecException("Loss = NaN or Infinity: current settings does not fit the recommender! Change the settings and try again!");
+        }
+
+        // check if converged
+        boolean converged = Math.abs(loss) < 1e-5;
+        lastLoss = loss;
+
+        return converged;
     }
 }

core/src/main/java/net/librec/recommender/MatrixFactorizationRecommender.java

@@ -55,11 +55,6 @@ public abstract class MatrixFactorizationRecommender extends AbstractRecommender
      */
     protected float regItem;
 
-    /**
-     * global mean
-     */
-    protected double globalMean;
-
     /**
      * setup
      * init member method
@@ -68,7 +63,7 @@ public abstract class MatrixFactorizationRecommender extends AbstractRecommender
      */
     protected void setup() throws LibrecException {
         super.setup();
-        numIterations = conf.getInt("rec.iterator.maximum");
+        numIterations = conf.getInt("rec.iterator.maximum",100);
         learnRate = conf.getFloat("rec.iterator.learnrate", 0.01f);
         maxLearnRate = conf.getFloat("rec.iterator.learnrate.maximum", 1000.0f);
 
@@ -81,7 +76,6 @@ protected void setup() throws LibrecException {
 
         userFactors = new DenseMatrix(numUsers, numFactors);
         itemFactors = new DenseMatrix(numItems, numFactors);
-        globalMean = trainMatrix.mean();
 
         initMean = 0.0f;
         initStd = 0.1f;
@@ -103,38 +97,6 @@ protected double predict(int userIdx, int itemIdx) throws LibrecException {
         return DenseMatrix.rowMult(userFactors, userIdx, itemFactors, itemIdx);
     }
 
-    /**
-     * Post each iteration, we do things:
-     * <ol>
-     * <li>print debug information</li>
-     * <li>check if converged</li>
-     * <li>if not, adjust learning rate</li>
-     * </ol>
-     *
-     * @param iter current iteration
-     * @return boolean: true if it is converged; false otherwise
-     */
-    protected boolean isConverged(int iter) throws LibrecException{
-        float delta_loss = (float) (lastLoss - loss);
-
-        // print out debug info
-        if (verbose) {
-            String recName = getClass().getSimpleName().toString();
-            String info = recName + " iter " + iter + ": loss = " + loss + ", delta_loss = " + delta_loss;
-            LOG.info(info);
-        }
-
-        if (Double.isNaN(loss) || Double.isInfinite(loss)) {
-//            LOG.error("Loss = NaN or Infinity: current settings does not fit the recommender! Change the settings and try again!");
-            throw new LibrecException("Loss = NaN or Infinity: current settings does not fit the recommender! Change the settings and try again!");
-        }
-
-        // check if converged
-        boolean converged = Math.abs(loss) < 1e-5;
-        lastLoss = loss;
-
-        return converged;
-    }
 
     /**
      * Update current learning rate after each epoch <br>

core/src/main/java/net/librec/recommender/cf/ranking/LDARecommender.java

@@ -53,6 +53,7 @@ public class LDARecommender extends ProbabilisticGraphicalRecommender {
      * Dirichlet hyper-parameters of topic-item distribution, typical value is 0.01
      */
     protected float initBeta;
+
     /**
      * entry[k, i]: number of tokens assigned to topic k, given item i.
      */
@@ -140,9 +141,9 @@ protected void setup() throws LibrecException {
         for (MatrixEntry matrixEntry : trainMatrix) {
             int userIdx = matrixEntry.row();
             int itemIdx = matrixEntry.column();
-            int num = (int) (matrixEntry.get());
+            int num = (int) (matrixEntry.get());    // problem 1 : the for cycle is not necessary
             for(int numIdx = 0; numIdx < num; numIdx++) {
-                int topicIdx = (int) (Math.random() * numTopics); // 0 ~ k-1
+                int topicIdx = (int) (Math.random() * numTopics); // 0 ~ k-1     // problem 2 : the random
 
                 // assign a topic t to pair (u, i)
                 topicAssignments.add(topicIdx);
@@ -171,7 +172,7 @@ protected void eStep() {
             int itemIdx = matrixEntry.column();
 
             int num = (int) (matrixEntry.get());
-            for (int numIdx = 0; numIdx < num; numIdx++) {
+            for (int numIdx = 0; numIdx < num; numIdx++) {      // problem 1 again
                 int topicIdx = topicAssignments.get(topicAssignmentsIdx); // topic
 
                 userTopicNumbers.add(userIdx, topicIdx, -1);
@@ -220,7 +221,7 @@ protected void mStep() {
 
             topicAlpha = alpha.get(topicIdx);
             double numerator = 0, denominator = 0;
-            for (int itemIdx = 0; itemIdx < numUsers; itemIdx++) {
+            for (int itemIdx = 0; itemIdx < numUsers; itemIdx++) {     // problem 2 : numUsers should be numItems ? or item index should be user index?
                 numerator += digamma(userTopicNumbers.get(itemIdx, topicIdx) + topicAlpha) - digamma(topicAlpha);
                 denominator += digamma(userTokenNumbers.get(itemIdx) + sumAlpha) - digamma(sumAlpha);
             }

core/src/main/java/net/librec/recommender/cf/rating/NMFRecommender.java

@@ -23,6 +23,7 @@
 import net.librec.math.structure.DenseVector;
 import net.librec.math.structure.MatrixEntry;
 import net.librec.math.structure.SparseVector;
+import net.librec.recommender.AbstractRecommender;
 import net.librec.recommender.MatrixFactorizationRecommender;
 
 /**
@@ -31,26 +32,42 @@
  * @author guoguibing and Keqiang Wang
  */
 @ModelData({"isRating", "nmf", "transUserFactors", "transItemFactors"})
-public class NMFRecommender extends MatrixFactorizationRecommender {
+public class NMFRecommender extends AbstractRecommender {
     /**
      * userFactors and itemFactors matrix transpose
      */
     DenseMatrix transUserFactors;
     DenseMatrix transItemFactors;
 
+    /**
+     * the number of latent factors;
+     */
+    protected int numFactors;
+
+    /**
+     * the number of iterations
+     */
+    protected int numIterations;
+
     @Override
     protected void setup() throws LibrecException {
         super.setup();
 
-        transUserFactors = userFactors.transpose();
-        transItemFactors = itemFactors.transpose();
+        numFactors = conf.getInt("rec.factor.number", 10);
+        numIterations = conf.getInt("rec.iterator.maximum",100);
+
+
+        transUserFactors = new DenseMatrix(numFactors, numUsers);
+        transItemFactors = new DenseMatrix(numFactors, numItems);
+        transUserFactors.init(0.01);
+        transItemFactors.init(0.01);
     }
 
     @Override
     protected void trainModel() throws LibrecException {
         for (int iter = 0; iter <= numIterations; ++iter) {
             // update userFactors by fixing itemFactors
-            for (int userIdx = 0; userIdx < numRates; userIdx++) {
+            for (int userIdx = 0; userIdx < numUsers; userIdx++) {
                 SparseVector itemRatingsVector = trainMatrix.row(userIdx);
 
                 if (itemRatingsVector.getCount() > 0) {
@@ -65,7 +82,8 @@ protected void trainModel() throws LibrecException {
                         double realValue = factorItemsVector.inner(itemRatingsVector);
                         double estmValue = factorItemsVector.inner(itemPredictsVector) + 1e-9;
 
-                        transUserFactors.set(factorIdx, userIdx, transUserFactors.get(factorIdx, userIdx) * (realValue / estmValue));
+                        transUserFactors.set(factorIdx, userIdx, transUserFactors.get(factorIdx, userIdx)
+                                * (realValue / estmValue));
                     }
                 }
             }
@@ -78,15 +96,16 @@ protected void trainModel() throws LibrecException {
                     SparseVector userPredictsVector = new SparseVector(numUsers, userRatingsVector.size());
 
                     for (int userIdx : userRatingsVector.getIndex()) {
-                        userPredictsVector.set(userIdx, predict(userIdx, itemIdx));
+                        userPredictsVector.append(userIdx, predict(userIdx, itemIdx));
                     }
 
                     for (int factorIdx = 0; factorIdx < numFactors; factorIdx++) {
                         DenseVector factorUsersVector = transUserFactors.row(factorIdx, false);
                         double realValue = factorUsersVector.inner(userRatingsVector);
                         double estmValue = factorUsersVector.inner(userPredictsVector) + 1e-9;
 
-                        transItemFactors.set(factorIdx, itemIdx, transItemFactors.get(factorIdx, itemIdx) * (realValue / estmValue));
+                        transItemFactors.set(factorIdx, itemIdx, transItemFactors.get(factorIdx, itemIdx)
+                                * (realValue / estmValue));
                     }
                 }
             }
@@ -109,7 +128,6 @@ protected void trainModel() throws LibrecException {
             if (isConverged(iter) && earlyStop) {
                 break;
             }
-            updateLRate(iter);
         }
     }
 

core/src/main/java/net/librec/recommender/content/EFMRecommender.java

@@ -35,7 +35,7 @@
 /**
  * EFM Recommender
  * Zhang Y, Lai G, Zhang M, et al. Explicit factor models for explainable recommendation based on phrase-level sentiment analysis[C]
- * //Proceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval. ACM, 2014: 83-92.
+ * {@code Proceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval.  ACM, 2014: 83-92}.
  *
  * @author ChenXu
  */
@@ -53,7 +53,6 @@ public class EFMRecommender extends BiasedMFRecommender {
     protected DenseMatrix userFeatureAttention;
     protected DenseMatrix itemFeatureQuality;
     protected DenseMatrix rating;
-    protected SparseMatrix trainData,testData;
     protected double lambdaX;
     protected double lambdaY;
     protected double lambdaU;
@@ -90,7 +89,6 @@ protected void setup() throws LibrecException {
         numberOfItems = 0;
         String user = "";
         String item = "";
-        int r = 0;
         String line = null;
 
         try {
@@ -134,14 +132,14 @@ protected void setup() throws LibrecException {
         } catch (IOException e) {
             e.printStackTrace();
         }
-        
-        
+
+
         // Create V,U1,H1,U2,H2
         featureMatrix = new DenseMatrix(numberOfFeatures, featureFactor);
         userFactors = new DenseMatrix(numberOfUsers, numFactors);
         itemFactors = new DenseMatrix(numberOfItems, numFactors);
-        
-        
+
+
         featureMatrix.init(initMean, initStd);
         userFeatureMatrix = userFactors.getSubMatrix(0, userFactors.numRows() - 1, 0, featureFactor - 1);
         userHiddenMatrix = userFactors.getSubMatrix(0, userFactors.numRows() - 1, featureFactor, userFactors.numColumns() - 1);
@@ -152,7 +150,7 @@ protected void setup() throws LibrecException {
         userFeatureAttention.init(0);
         itemFeatureQuality = new DenseMatrix(itemFactors.numRows(), numberOfFeatures);
         itemFeatureQuality.init(0);
-        
+
 
         // compute UserFeatureAttention
         double[] featureValues = new double[numberOfFeatures];
@@ -203,7 +201,7 @@ protected void trainModel() throws LibrecException {
                 for (int j = 0; j < featureFactor; j++) {
                     double updateValue = ((userFeatureAttention.transpose().mult(userFeatureMatrix).scale(lambdaX)).add(itemFeatureQuality.transpose().mult(itemFeatureMatrix).scale(lambdaX))).get(i, j);
                     updateValue /= featureMatrix.mult((userFeatureMatrix.transpose().mult(userFeatureMatrix).scale(lambdaX)).add(itemFeatureMatrix.transpose().mult(itemFeatureMatrix).scale(lambdaY))
-                    		.add(DenseMatrix.eye(featureFactor).scale(lambdaV))).get(i, j);
+                            .add(DenseMatrix.eye(featureFactor).scale(lambdaV))).get(i, j);
                     updateValue = Math.sqrt(updateValue);
                     featureMatrix.set(i, j, featureMatrix.get(i, j) * updateValue);
                 }

core/src/main/java/net/librec/recommender/content/HFTRecommender.java

@@ -98,7 +98,6 @@ protected void setup() throws LibrecException {
         numberOfItems = 0;
         String user = "";
         String item = "";
-        int r = 0;
         String line = null;
         Table<Integer, Integer, String> res = HashBasedTable.create();
         Table<Integer, Integer, Double> ratings = HashBasedTable.create();

core/src/main/resources/rec/baseline/mostpopular-test.properties

@@ -1 +1,2 @@
 rec.recommender.class=mostpopular
+rec.recommender.isranking=true

core/src/main/resources/rec/cf/ranking/wrmf-test.properties

@@ -1,12 +1,8 @@
 rec.recommender.class=wrmf
-rec.iterator.learnrate=0.01
-rec.iterator.learnrate.maximum=0.01
 rec.iterator.maximum=20
 rec.user.regularization=0.01
 rec.item.regularization=0.01
 rec.factor.number=10
-rec.learnrate.bolddriver=false
-rec.learnrate.decay=1.0
 rec.recommender.isranking=true
 rec.recommender.ranking.topn=10
 

core/src/main/resources/rec/cf/rating/biasedmf-test.properties

@@ -1,7 +1,7 @@
 rec.recommender.class=biasedmf
 rec.iterator.learnrate=0.01
 rec.iterator.learnrate.maximum=0.01
-rec.iterator.maximum=1
+rec.iterator.maximum=10
 rec.user.regularization=0.01
 rec.item.regularization=0.01
 rec.bias.regularization=0.01

core/src/main/resources/rec/cf/rating/bpmf-test.properties

@@ -1,9 +1,9 @@
 rec.recommender.class=bpmf
-rec.iterator.learnrate=0.01
-rec.iterator.learnrate.maximum=0.01
+# rec.iterator.learnrate=0.01
+# rec.iterator.learnrate.maximum=0.01
 rec.iterator.maximum=100
 rec.user.regularization=0.01
 rec.item.regularization=0.01
 rec.factor.number=10
-rec.learnrate.bolddriver=false
-rec.learnrate.decay=1.0
+# rec.learnrate.bolddriver=false
+# rec.learnrate.decay=1.0

core/src/main/resources/rec/cf/rating/mfals-test.properties

@@ -1,9 +1,5 @@
 rec.recommender.class=mfals
-rec.iterator.learnrate=0.01
-rec.iterator.learnrate.maximum=0.01
 rec.iterator.maximum=100
 rec.user.regularization=0.01
 rec.item.regularization=0.01
 rec.factor.number=10
-rec.learnrate.bolddriver=false
-rec.learnrate.decay=1.0

core/src/main/resources/rec/cf/rating/nmf-test.properties

@@ -1,9 +1,5 @@
 rec.recommender.class=nmf
-rec.iterator.learnrate=0.01
-rec.iterator.learnrate.maximum=0.01
-rec.iterator.maximum=100
-rec.user.regularization=0.01
-rec.item.regularization=0.01
-rec.factor.number=10
+rec.iterator.maximum=10
+rec.factor.number=100
 rec.learnrate.bolddriver=false
 rec.learnrate.decay=1.0

core/src/main/resources/rec/content/hft-test.properties

@@ -8,7 +8,6 @@ rec.item.regularization=0.01
 rec.factor.number=10
 rec.learnrate.bolddriver=false
 rec.learnrate.decay=1.0
-rec.eval.enable = 1
 rec.recommender.lambda.user=0.05
 rec.recommender.lambda.item=0.05
 rec.bias.regularization = 0.01