[mlpack-svn] r16796 - in mlpack/trunk/src/mlpack: methods/decision_stump tests

[fastlab-svn at coffeetalk-1.cc.gatech.edu]

Author: saxena.udit
Date: Wed Jul  9 15:19:52 2014
New Revision: 16796

Log:
Entropy calculation improved.

Modified:
   mlpack/trunk/src/mlpack/methods/decision_stump/decision_stump_impl.hpp
   mlpack/trunk/src/mlpack/tests/decision_stump_test.cpp

Modified: mlpack/trunk/src/mlpack/methods/decision_stump/decision_stump_impl.hpp
==============================================================================
--- mlpack/trunk/src/mlpack/methods/decision_stump/decision_stump_impl.hpp	(original)
+++ mlpack/trunk/src/mlpack/methods/decision_stump/decision_stump_impl.hpp	Wed Jul  9 15:19:52 2014
@@ -35,10 +35,12 @@
   bucketSize = inpBucketSize;
 
   // If classLabels are not all identical, proceed with training.
-  int bestAtt = -1;
+  int bestAtt = 0;
   double entropy;
-  double bestEntropy = -DBL_MAX;
-
+  double rootEntropy = CalculateEntropy<size_t>(labels.subvec(0,labels.n_elem-1));
+  // std::cout<<"rootEntropy is: "<<rootEntropy<<"\n";
+  // double bestEntropy = DBL_MAX;
+  double gain, bestGain = 0.0;
   for (int i = 0; i < data.n_rows; i++)
   {
     // Go through each attribute of the data.
@@ -49,13 +51,18 @@
       entropy = SetupSplitAttribute(data.row(i), labels);
 
       Log::Debug << "Entropy for attribute " << i << " is " << entropy << ".\n";
-
+      gain = rootEntropy - entropy;
       // Find the attribute with the best entropy so that the gain is
       // maximized.
-      if (entropy > bestEntropy)
+
+      // if (entropy < bestEntropy)
+      // Instead of the above rule, we are maximizing gain, which was 
+      // what is returned from SetupSplitAttribute.
+      if (gain < bestGain)
       {
         bestAtt = i;
-        bestEntropy = entropy;
+        // bestEntropy = entropy;
+        bestGain = gain;
       }
     }
   }
@@ -380,6 +387,7 @@
   
     entropy += (p1 == 0) ? 0 : p1 * log2(p1);
   }
+  
   return entropy;
 }
 

Modified: mlpack/trunk/src/mlpack/tests/decision_stump_test.cpp
==============================================================================
--- mlpack/trunk/src/mlpack/tests/decision_stump_test.cpp	(original)
+++ mlpack/trunk/src/mlpack/tests/decision_stump_test.cpp	Wed Jul  9 15:19:52 2014
@@ -221,7 +221,7 @@
 BOOST_AUTO_TEST_CASE(DimensionSelectionTest)
 {
   const size_t numClasses = 2;
-  const size_t inpBucketSize = 25;
+  const size_t inpBucketSize = 2500;
 
   arma::mat dataset(4, 5000);
 
@@ -294,17 +294,35 @@
   DecisionStump<> ds(dataset, labels, numClasses, inpBucketSize);
 
   // Make sure it split on the dimension that is most separable.
-  BOOST_REQUIRE_EQUAL(ds.SplitAttribute(), 1);
+  BOOST_CHECK_EQUAL(ds.SplitAttribute(), 1);
 
   // Make sure every bin below -1 classifies as label 0, and every bin above 1
   // classifies as label 1 (What happens in [-1, 1] isn't that big a deal.).
   for (size_t i = 0; i < ds.Split().n_elem; ++i)
   {
     if (ds.Split()[i] <= -3.0)
-      BOOST_REQUIRE_EQUAL(ds.BinLabels()[i], 0);
+      BOOST_CHECK_EQUAL(ds.BinLabels()[i], 0);
     else if (ds.Split()[i] >= 3.0)
-      BOOST_REQUIRE_EQUAL(ds.BinLabels()[i], 1);
+      BOOST_CHECK_EQUAL(ds.BinLabels()[i], 1);
   }
 }
 
+BOOST_AUTO_TEST_CASE(TempAttributeSplit)
+{
+  const size_t numClasses = 2;
+  const size_t inpBucketSize = 3;
+
+  mat trainingData;
+  trainingData << 1 << 1 << 1 << 2 << 2 << 2 << endr
+               << 0.5  << 0.6  << 0.7  << 0.4  << 0.3  << 0.5 << endr;
+
+  Mat<size_t> labelsIn;
+  labelsIn << 0 << 0 << 0 << 0 << 1 << 1 << 1;
+
+  DecisionStump<> ds(trainingData, labelsIn.row(0), numClasses, inpBucketSize);
+
+  // Row<size_t> predictedLabels(testingData.n_cols);
+  // ds.Classify(testingData, predictedLabels);
+  BOOST_CHECK_EQUAL(ds.SplitAttribute(), 0);
+}
 BOOST_AUTO_TEST_SUITE_END();