fix Dart::NormalizeTrees (#1265)

dmlc · Jun 9, 2016 · 7cfeb5f · 7cfeb5f
1 parent 949d1e3
commit 7cfeb5f
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/doc/parameter.md b/doc/parameter.md
@@ -84,10 +84,10 @@ Additional parameters for Dart Booster
 * normalize_type [default="tree]
   - type of normalization algorithm.
     - "tree": New trees have the same weight of each of dropped trees.
-              weight of new trees are learning_rate / (k + learnig_rate)
+              weight of new trees are 1 / (k + learnig_rate)
               dropped trees are scaled by a factor of k / (k + learning_rate)
     - "forest": New trees have the same weight of sum of dropped trees (forest).
-                weight of new trees are learning_rate / (1 + learning_rate)
+                weight of new trees are 1 / (1 + learning_rate)
                 dropped trees are scaled by a factor of 1 / (1 + learning_rate)
 * rate_drop [default=0.0]
   - dropout rate.

diff --git a/src/gbm/gbtree.cc b/src/gbm/gbtree.cc
@@ -703,7 +703,7 @@ class Dart : public GBTree {
           weight_drop[i] *= factor;
         }
         for (size_t i = 0; i < size_new_trees; ++i) {
-          weight_drop.push_back(lr * factor);
+          weight_drop.push_back(factor);
         }
       } else {
         // normalize_type 0
@@ -712,7 +712,7 @@ class Dart : public GBTree {
           weight_drop[i] *= factor;
         }
         for (size_t i = 0; i < size_new_trees; ++i) {
-          weight_drop.push_back(1.0 * lr / (num_drop + lr));
+          weight_drop.push_back(1.0 / (num_drop + lr));
         }
       }
     }