Adding tests that show how to chain feature selectors and minor refac…

…toring.

TODO.txt

@@ -1,8 +1,6 @@
 CODE IMPROVEMENTS
 =================
 
-- Create a AnovaSelect feature selector and use it on the tests.
-
 - Consider dropping all the common.dataobjects and use their internalData directly instead.
 - Refactor the statistics package and replace all the static methods with proper inheritance.
 - Write generic optimizers instead of having optimization methods in the algorithms. Add the optimizers and regularization packages under mathematics.
@@ -35,7 +33,7 @@ NEW ALGORITHMS
 ==============
 
 - Create a PercentileScaler numerical scaler.
-- Create the following FeatureSelectors: KruskalWallisSelect, SpearmanSelect.
+- Create the following FeatureSelectors: AnovaSelect, KruskalWallisSelect, SpearmanSelect.
 - Speed up LDA: http://www.cs.ucsb.edu/~mingjia/cs240/doc/273811.pdf
 - Factorization Machines: http://www.csie.ntu.edu.tw/~b97053/paper/Rendle2010FM.pdf
 - Develop the FunkSVD and PLSI as probabilistic version of SVD.

datumbox-framework-applications/src/test/java/com/datumbox/framework/applications/datamodeling/ModelerTest.java

@@ -19,7 +19,9 @@
 import com.datumbox.framework.common.dataobjects.Dataframe;
 import com.datumbox.framework.common.dataobjects.Record;
 import com.datumbox.framework.core.machinelearning.MLBuilder;
-import com.datumbox.framework.core.machinelearning.classification.MultinomialNaiveBayes;
+import com.datumbox.framework.core.machinelearning.classification.SoftMaxRegression;
+import com.datumbox.framework.core.machinelearning.featureselection.ChisquareSelect;
+import com.datumbox.framework.core.machinelearning.featureselection.PCA;
 import com.datumbox.framework.core.machinelearning.modelselection.metrics.ClassificationMetrics;
 import com.datumbox.framework.core.machinelearning.preprocessing.OneHotEncoder;
 import com.datumbox.framework.core.machinelearning.preprocessing.MinMaxScaler;
@@ -53,8 +55,8 @@ public void testTrainAndValidate() {
         Dataframe[] data = Datasets.heartDiseaseClusters(configuration);
 
         Dataframe trainingData = data[0];
-        Dataframe validationData = data[1];
-
+        Dataframe validationData = data[0].copy();
+        Dataframe testData = data[1];
 
         String storageName = this.getClass().getSimpleName();
 
@@ -70,41 +72,47 @@ public void testTrainAndValidate() {
         trainingParameters.setCategoricalEncoderTrainingParameters(ceParams);
 
         //feature selection configuration
-        trainingParameters.setFeatureSelectorTrainingParametersList(Arrays.asList());
+
+        PCA.TrainingParameters pcaParams = new PCA.TrainingParameters();
+        pcaParams.setVariancePercentageThreshold(0.99999995);
+        trainingParameters.setFeatureSelectorTrainingParametersList(Arrays.asList(new ChisquareSelect.TrainingParameters(), pcaParams));
 
         //model Configuration
-        MultinomialNaiveBayes.TrainingParameters modelTrainingParameters = new MultinomialNaiveBayes.TrainingParameters();
-        modelTrainingParameters.setMultiProbabilityWeighted(true);
+        SoftMaxRegression.TrainingParameters modelTrainingParameters = new SoftMaxRegression.TrainingParameters();
+        modelTrainingParameters.setL1(0.0001);
+        modelTrainingParameters.setL2(0.0001);
+        modelTrainingParameters.setTotalIterations(100);
         trainingParameters.setModelerTrainingParameters(modelTrainingParameters);
 
         Modeler instance = MLBuilder.create(trainingParameters, configuration);
         instance.fit(trainingData);
         instance.save(storageName);
 
         instance.close();
+        trainingData.close();
 
         instance = MLBuilder.load(Modeler.class, storageName, configuration);
 
-        instance.predict(trainingData);
+        instance.predict(validationData);
 
-        ClassificationMetrics vm = new ClassificationMetrics(trainingData);
+        ClassificationMetrics vm = new ClassificationMetrics(validationData);
 
-        double expResult2 = 0.7867564534231202;
+        double expResult2 = 0.8428731762065095;
         assertEquals(expResult2, vm.getMacroF1(), Constants.DOUBLE_ACCURACY_HIGH);
 
-        trainingData.close();
+        validationData.close();
         instance.close();
 
 
         instance = MLBuilder.load(Modeler.class, storageName, configuration);
 
-        instance.predict(validationData);
+        instance.predict(testData);
 
 
 
         Map<Integer, Object> expResult = new HashMap<>();
         Map<Integer, Object> result = new HashMap<>();
-        for(Map.Entry<Integer, Record> e : validationData.entries()) {
+        for(Map.Entry<Integer, Record> e : testData.entries()) {
             Integer rId = e.getKey();
             Record r = e.getValue();
             expResult.put(rId, r.getY());
@@ -114,7 +122,7 @@ public void testTrainAndValidate() {
 
         instance.delete();
 
-        validationData.close();
+        testData.close();
     }
 
 }

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/common/abstracts/featureselectors/AbstractCountBasedFeatureSelector.java

@@ -134,13 +134,13 @@ protected void _fit(Dataframe trainingData) {
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedXDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL));
     }
 
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedYDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.CATEGORICAL, TypeInference.DataType.ORDINAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.CATEGORICAL, TypeInference.DataType.ORDINAL));
     }
 
     /**

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/common/abstracts/transformers/AbstractEncoder.java

@@ -20,7 +20,6 @@
 import com.datumbox.framework.core.machinelearning.common.abstracts.AbstractTrainer;
 
 import java.util.Arrays;
-import java.util.Collections;
 import java.util.HashSet;
 import java.util.Set;
 
@@ -54,6 +53,6 @@ protected AbstractEncoder(String storageName, Configuration configuration) {
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedXDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.CATEGORICAL, TypeInference.DataType.ORDINAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.CATEGORICAL, TypeInference.DataType.ORDINAL));
     }
 }

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/common/abstracts/transformers/AbstractScaler.java

@@ -21,7 +21,6 @@
 import com.datumbox.framework.core.machinelearning.common.abstracts.AbstractTrainer;
 
 import java.util.Arrays;
-import java.util.Collections;
 import java.util.HashSet;
 import java.util.Set;
 
@@ -92,7 +91,7 @@ protected AbstractScaler(String storageName, Configuration configuration) {
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedXDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.NUMERICAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.NUMERICAL));
     }
 
 }

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/featureselection/ChisquareSelect.java

@@ -31,7 +31,7 @@
 
 /**
  * Implementation of the Chisquare Feature Selection algorithm which can be used
- * for evaluating categorical and boolean variables.
+ * for evaluating boolean variables.
  * 
  * References: 
  * http://nlp.stanford.edu/IR-book/html/htmledition/feature-selectionchi2-feature-selection-1.html

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/featureselection/MutualInformation.java

@@ -27,7 +27,7 @@
 
 /**
  * Implementation of the Mutual Information Feature Selection algorithm which can be used
- * for evaluating categorical and boolean variables.
+ * for evaluating boolean variables.
  * 
  * References: 
  * http://nlp.stanford.edu/IR-book/html/htmledition/mutual-information-1.html

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/featureselection/PCA.java

@@ -365,7 +365,7 @@ protected void _transform(Dataframe newData) {
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedXDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL));
     }
 
     /** {@inheritDoc} */

datumbox-framework-core/src/main/java/com/datumbox/framework/core/machinelearning/featureselection/TFIDF.java

@@ -211,7 +211,7 @@ protected void _fit(Dataframe trainingData) {
     /** {@inheritDoc} */
     @Override
     protected Set<TypeInference.DataType> getSupportedXDataTypes() {
-        return Collections.unmodifiableSet(new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL)));
+        return new HashSet<>(Arrays.asList(TypeInference.DataType.BOOLEAN, TypeInference.DataType.NUMERICAL));
     }
 
     /** {@inheritDoc} */