Merge pull request #56 from imbs-hl/fix_regression

Check for pure nodes in regression trees

NEWS.md

@@ -1,3 +1,6 @@
+##### Version 0.4.1
+* Runtime improvement for regression forests on classification data
+
 ##### Version 0.4.0
 * New CRAN version. New CRAN versions will be 0.x.0, development versions 0.x.y
 

ranger-r-package/ranger/DESCRIPTION

@@ -1,8 +1,8 @@
 Package: ranger
 Type: Package
 Title: A Fast Implementation of Random Forests
-Version: 0.4.0
-Date: 2016-04-07
+Version: 0.4.1
+Date: 2016-04-28
 Author: Marvin N. Wright
 Maintainer: Marvin N. Wright <[email protected]>
 Description: A fast implementation of Random Forests, particularly suited for high dimensional data. Ensembles

ranger-r-package/ranger/NEWS

@@ -1,3 +1,6 @@
+##### Version 0.4.1
+* Runtime improvement for regression forests on classification data
+
 ##### Version 0.4.0
 * Reduce memory usage of savest forest objects (changed child.nodeIDs interface)
 * Add keep.inbag option to track in-bag counts

source/src/Tree/TreeRegression.cpp

@@ -81,7 +81,23 @@ bool TreeRegression::splitNodeInternal(size_t nodeID, std::vector<size_t>& possi
     return true;
   }
 
-// Find best split, stop if no decrease of impurity
+  // Check if node is pure and set split_value to estimate and stop if pure
+  bool pure = true;
+  double pure_value = 0;
+  for (size_t i = 0; i < sampleIDs[nodeID].size(); ++i) {
+    double value = data->get(sampleIDs[nodeID][i], dependent_varID);
+    if (i != 0 && value != pure_value) {
+      pure = false;
+      break;
+    }
+    pure_value = value;
+  }
+  if (pure) {
+    split_values[nodeID] = pure_value;
+    return true;
+  }
+
+  // Find best split, stop if no decrease of impurity
   bool stop = findBestSplit(nodeID, possible_split_varIDs);
   if (stop) {
     split_values[nodeID] = estimate(nodeID);

source/src/version.h

@@ -1,3 +1,3 @@
 #ifndef RANGER_VERSION
-#define RANGER_VERSION "0.4.0"
+#define RANGER_VERSION "0.4.1"
 #endif