DTBOblique: multi-class code revised

classifier/fpclassifier/randomforest/DTBOblique.cpp

@@ -93,6 +93,61 @@ bool DTBOblique::entropyLeftRight (
     return true;
 }
 
+void DTBOblique::adaptDataAndLabelForMultiClass (
+        const int posClass,
+        const int negClass,
+        NICE::Matrix & matX,
+        NICE::Vector & vecY,
+        bool & posHasExamples,
+        bool & negHasExamples )
+{
+    int posCount = 0;
+    int negCount = 0;
+    int outCount = 0;
+
+    // One-vs-one: Transforming into {-1,0,+1} problem
+    if ( useOneVsOne )
+        for ( int i = 0; i < vecY.size(); i++ )
+        {
+            if ( vecY[i] == posClass )
+            {
+                vecY[i] = 1.0;
+                posHasExamples = true;
+                posCount++;
+            }
+            else if ( vecY[i] == negClass )
+            {
+                vecY[i] = -1.0;
+                negHasExamples = true;
+                negCount++;
+            }
+            else
+            {
+                vecY[i] = 0.0;
+                matX.setRow( i, NICE::Vector( matX.cols(), 0.0 ) );
+                outCount++;
+            }
+        }
+    // One-vs-all: Transforming into {-1,+1} problem
+    else
+        for ( int i = 0; i < vecY.size(); i++ )
+        {
+            if ( vecY[i] == posClass )
+            {
+                vecY[i] = 1.0;
+                posHasExamples = true;
+                posCount++;
+            }
+            else
+            {
+                vecY[i] = -1.0;
+                negHasExamples = true;
+                negCount++;
+            }
+        }
+
+}
+
 /** refresh data matrix X and label vector y */
 void DTBOblique::getDataAndLabel(
         const FeaturePool &fp,
@@ -118,7 +173,7 @@ void DTBOblique::getDataAndLabel(
 
         NICE::Vector pixelRepr = f->getFeatureVector( &ce );
 
-        double label = p.first * ce.weight;
+        double label = p.first;// * ce.weight;
         pixelRepr *= ce.weight;
 
         y.set( matIndex, label );
@@ -315,98 +370,43 @@ DecisionNode *DTBOblique::buildRecursive(
     bestSplitInfo.entropyLeft = 0.0;
     bestSplitInfo.entropyRight = 0.0;
 
-//    double best_threshold = 0.0;
-//    double best_ig = -1.0;
-//    double *best_distribution_left = new double [maxClassNo+1];
-//    double *best_distribution_right = new double [maxClassNo+1];
-//    double best_entropy_left = 0.0;
-//    double best_entropy_right = 0.0;
-
     ConvolutionFeature *f = (ConvolutionFeature*)fp.begin()->second;
     bestSplitInfo.params = f->getParameterVector();
 
     // Creating data matrix X and label vector y
-    NICE::Matrix X, XTXr, G, temp;
+    NICE::Matrix X;
     NICE::Vector y, beta;
     getDataAndLabel( fp, examples, examples_selection, X, y );
 
-    // Preparing system of linear equations
-    regularizeDataMatrix( X, XTXr, regularizationType, lambdaCurrent );
-    choleskyDecomp(XTXr, G);
-    choleskyInvert(G, XTXr);
-    temp = XTXr * X.transpose();
-
-
-    if ( useOneVsOne )
+    // Transforming into multi-class problem
+    for ( int posClass = 0; posClass <= maxClassNo; posClass++ )
     {
-        // One-vs-one: Transforming into {-1,0,+1} problem
-        for ( int curClass = 0; curClass <= maxClassNo; curClass++ )
-            for ( int opClass = 0; opClass <= maxClassNo; opClass++ )
-            {
-                if ( curClass == opClass ) continue;
-
-                NICE::Vector yCur ( y.size(), 0.0 );
-                int idx = 0;
-                bool curHasExamples = false;
-                bool opHasExamples = false;
-
-                for ( vector<int>::const_iterator si = examples_selection.begin();
-                      si != examples_selection.end();
-                      si++, idx++ )
-                {
-                    const pair<int, Example> & p = examples[*si];
-                    if ( p.first == curClass )
-                    {
-                        yCur.set( idx, 1.0 );
-                        curHasExamples = true;
-                    }
-                    else if ( p.first == opClass )
-                    {
-                        yCur.set( idx, -1.0 );
-                        opHasExamples = true;
-                    }
-                }
-
-                // are there positive examples for current and opposition class in current set?
-                if ( !curHasExamples || !opHasExamples ) continue;
-
-                // Solve system of linear equations in a least squares manner
-                beta.multiply(temp,yCur,false);
-
-                // Updating parameter vector in convolutional feature
-                f->setParameterVector( beta );
-
-                // Feature Values
-                values.clear();
-                f->calcFeatureValues( examples, examples_selection, values);
-
-                // complete search for threshold
-                findBestSplitThreshold ( values, bestSplitInfo, beta, e,
-                                         maxClassNo );
-            }
-    }
-    else
-    {
-        // One-vs-all: Transforming into {-1,+1} problem
-        for ( int curClass = 0; curClass <= maxClassNo; curClass++ )
+        bool gotInnerIteration = false;
+        for ( int negClass = 0; negClass <= maxClassNo; negClass++ )
         {
-            NICE::Vector yCur ( y.size(), -1.0 );
-            int idx = 0;
-            bool hasExamples = false;
-            for ( vector<int>::const_iterator si = examples_selection.begin();
-                  si != examples_selection.end();
-                  si++, idx++ )
-            {
-                const pair<int, Example> & p = examples[*si];
-                if ( p.first == curClass )
-                {
-                    yCur.set( idx, 1.0 );
-                    hasExamples = true;
-                }
-            }
+            if ( posClass == negClass ) continue;
+
+            NICE::Vector yCur = y;
+            NICE::Matrix XCur = X;
+
+            bool posHasExamples = false;
+            bool negHasExamples = false;
 
-            // is there a positive example for current class in current set?
-            if (!hasExamples) continue;
+            adaptDataAndLabelForMultiClass(
+                posClass, negClass, XCur, yCur, posHasExamples, negHasExamples );
+
+            // are there examples for positive and negative class?
+            if ( !posHasExamples || !negHasExamples ) continue;
+
+            // one-vs-all setting: only one iteration for inner loop
+            if ( !useOneVsOne && gotInnerIteration ) continue;
+
+            // Preparing system of linear equations
+            NICE::Matrix XTXr, G, temp;
+            regularizeDataMatrix( XCur, XTXr, regularizationType, lambdaCurrent );
+            choleskyDecomp(XTXr, G);
+            choleskyInvert(G, XTXr);
+            temp = XTXr * XCur.transpose();
 
             // Solve system of linear equations in a least squares manner
             beta.multiply(temp,yCur,false);
@@ -421,10 +421,10 @@ DecisionNode *DTBOblique::buildRecursive(
             // complete search for threshold
             findBestSplitThreshold ( values, bestSplitInfo, beta, e, maxClassNo );
 
+            gotInnerIteration = true;
         }
     }
 
-
     // supress strange behaviour for values near zero (8.88178e-16)
     if (bestSplitInfo.entropyLeft < 1.0e-10 ) bestSplitInfo.entropyLeft = 0.0;
     if (bestSplitInfo.entropyRight < 1.0e-10 ) bestSplitInfo.entropyRight = 0.0;

classifier/fpclassifier/randomforest/DTBOblique.h

@@ -75,6 +75,23 @@ class DTBOblique : public DecisionTreeBuilder
     /////////////////////////
     /////////////////////////
 
+    /**
+     * @brief adaptDataAndLabelForMultiClass
+     * @param posClass positive class number
+     * @param negClass negative class number
+     * @param matX adapted data matrix
+     * @param vecY adapted label vector
+     * @param posHasExamples whether positive class has examples or not
+     * @param negHasExamples whether negative class has examples or not
+     */
+    void adaptDataAndLabelForMultiClass (
+            const int posClass,
+            const int negClass,
+            NICE::Matrix & matX,
+            NICE::Vector & vecY,
+            bool & posHasExamples,
+            bool & negHasExamples );
+
     /**
     * @brief get data matrix X and label vector y
     * @param fp feature pool