Changeset 12697 for branches/GBT-trunkintegration/HeuristicLab.Algorithms.DataAnalysis/3.4/GradientBoostedTrees/RegressionTreeBuilder.cs

Timestamp:

07/09/15 16:39:37 (9 years ago)

Author:

gkronber

Message:

#2261 removed line search closure (binding y.ToArray(), and pred.ToArray())

File:

• branches/GBT-trunkintegration/HeuristicLab.Algorithms.DataAnalysis/3.4/GradientBoostedTrees/RegressionTreeBuilder.cs (modified) (10 diffs)

branches/GBT-trunkintegration/HeuristicLab.Algorithms.DataAnalysis/3.4/GradientBoostedTrees/RegressionTreeBuilder.cs

@@ -42 +42 @@
     private readonly int nCols;
     private readonly double[][] x; // all training data (original order from problemData), x is constant
+    private double[] originalY; // the original target labels (from problemData), originalY is constant
+    private double[] curPred; // current predictions for originalY (in case we are using gradient boosting, otherwise = zeros), only necessary for line search
+
     private double[] y; // training labels (original order from problemData), y can be changed
 
@@ -102 +105 @@
 
       x = new double[nCols][];
-      y = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices).ToArray();
-
+      originalY = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices).ToArray();
+      y = new double[originalY.Length];
+      Array.Copy(originalY, y, y.Length); // copy values (originalY is fixed, y is changed in gradient boosting)
+      curPred = Enumerable.Repeat(0.0, y.Length).ToArray(); // zeros
 
       int col = 0;
@@ -127 +132 @@
 
       var seLoss = new SquaredErrorLoss();
-      var zeros = Enumerable.Repeat(0.0, y.Length);
-
-      var model = CreateRegressionTreeForGradientBoosting(y, maxSize, problemData.TrainingIndices.ToArray(), seLoss.GetLineSearchFunc(y, zeros), r, m);
+
+      var model = CreateRegressionTreeForGradientBoosting(y, curPred, maxSize, problemData.TrainingIndices.ToArray(), seLoss, r, m);
 
       return new GradientBoostedTreesModel(new[] { new ConstantRegressionModel(yAvg), model }, new[] { 1.0, 1.0 });
@@ -135 +139 @@
 
     // specific interface that allows to specify the target labels and the training rows which is necessary when for gradient boosted trees
-    public IRegressionModel CreateRegressionTreeForGradientBoosting(double[] y, int maxSize, int[] idx, LineSearchFunc lineSearch, double r = 0.5, double m = 0.5) {
+    public IRegressionModel CreateRegressionTreeForGradientBoosting(double[] y, double[] curPred, int maxSize, int[] idx, ILossFunction lossFunction, double r = 0.5, double m = 0.5) {
       Debug.Assert(maxSize > 0);
       Debug.Assert(r > 0);
@@ -143 +147 @@
       Debug.Assert(m <= 1.0);
 
-      this.y = y; // y is changed in gradient boosting
+      // y and curPred are changed in gradient boosting
+      this.y = y; 
+      this.curPred = curPred; 
 
       // shuffle row idx
@@ -157 +163 @@
       effectiveVars = (int)Math.Ceiling(nCols * m);
 
-      // the which array is used for partining row idxs 
+      // the which array is used for partitioing row idxs  
       Array.Clear(which, 0, which.Length);
 
@@ -184 +190 @@
       // and calculate the best split for this root node and enqueue it into a queue sorted by improvement throught the split
       // start and end idx are inclusive
-      CreateLeafNode(0, effectiveRows - 1, lineSearch);
+      CreateLeafNode(0, effectiveRows - 1, lossFunction);
 
       // process the priority queue to complete the tree
-      CreateRegressionTreeFromQueue(maxSize, lineSearch);
+      CreateRegressionTreeFromQueue(maxSize, lossFunction);
 
       return new RegressionTreeModel(tree.ToArray());
@@ -194 +200 @@
 
     // processes potential splits from the queue as long as splits are left and the maximum size of the tree is not reached
-    private void CreateRegressionTreeFromQueue(int maxNodes, LineSearchFunc lineSearch) {
+    private void CreateRegressionTreeFromQueue(int maxNodes, ILossFunction lossFunction) {
       while (queue.Any() && curTreeNodeIdx + 1 < maxNodes) { // two nodes are created in each loop
         var f = queue[queue.Count - 1]; // last element has the largest improvement
@@ -213 +219 @@
 
         // create two leaf nodes (and enqueue best splits for both)
-        var leftTreeIdx = CreateLeafNode(startIdx, splitIdx, lineSearch);
-        var rightTreeIdx = CreateLeafNode(splitIdx + 1, endIdx, lineSearch);
+        var leftTreeIdx = CreateLeafNode(startIdx, splitIdx, lossFunction);
+        var rightTreeIdx = CreateLeafNode(splitIdx + 1, endIdx, lossFunction);
 
         // overwrite existing leaf node with an internal node
@@ -223 +229 @@
 
     // returns the index of the newly created tree node
-    private int CreateLeafNode(int startIdx, int endIdx, LineSearchFunc lineSearch) {
+    private int CreateLeafNode(int startIdx, int endIdx, ILossFunction lossFunction) {
       // write a leaf node
-      var val = lineSearch(internalIdx, startIdx, endIdx);
+      var val = lossFunction.LineSearch(originalY, curPred, internalIdx, startIdx, endIdx);
       tree[curTreeNodeIdx] = new RegressionTreeModel.TreeNode(RegressionTreeModel.TreeNode.NO_VARIABLE, val);