1 // Copyright 2017 The TensorFlow Authors. All Rights Reserved. 2 // 3 // Licensed under the Apache License, Version 2.0 (the "License"); 4 // you may not use this file except in compliance with the License. 5 // You may obtain a copy of the License at 6 // 7 // http://www.apache.org/licenses/LICENSE-2.0 8 // 9 // Unless required by applicable law or agreed to in writing, software 10 // distributed under the License is distributed on an "AS IS" BASIS, 11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 // See the License for the specific language governing permissions and 13 // limitations under the License. 14 // ============================================================================= 15 #ifndef TENSORFLOW_CONTRIB_BOOSTED_TREES_LIB_UTILS_DROPOUT_UTILS_H_ 16 #define TENSORFLOW_CONTRIB_BOOSTED_TREES_LIB_UTILS_DROPOUT_UTILS_H_ 17 18 #include <unordered_set> 19 #include <vector> 20 21 #include "tensorflow/contrib/boosted_trees/proto/learner.pb.h" // NOLINT 22 #include "tensorflow/core/lib/core/status.h" 23 #include "tensorflow/core/platform/types.h" 24 25 namespace tensorflow { 26 namespace boosted_trees { 27 namespace utils { 28 29 // Utils for deciding on what trees to be/were dropped when building a new tree. 30 class DropoutUtils { 31 public: 32 // This method determines what trees should be dropped and returns their 33 // indices and the weights they had when this method ran. 34 // seed: random seed to be used 35 // config: dropout config, that defines the probability of dropout etc 36 // trees_not_to_drop: indices of trees that can't be dropped, for example bias 37 // (0) and the last tree in the batch mode. 38 // number_of_trees_to_consider: how many trees are currently in the ensemble 39 // weights: weights of those trees 40 // Returns sorted vector of indices of trees to be dropped and their original 41 // weights. 42 static tensorflow::Status DropOutTrees( 43 const uint64 seed, const learner::LearningRateDropoutDrivenConfig& config, 44 const std::unordered_set<int32>& trees_not_to_drop, 45 const std::vector<float>& weights, std::vector<int32>* dropped_trees, 46 std::vector<float>* original_weights); 47 48 // Recalculates the weights of the trees when the new trees are added to 49 // ensemble. 50 // dropped_trees: ids of trees that were dropped when trees to add were built. 51 // dropped_trees_original_weights: the weight dropped trees had during dropout 52 // new_trees_first_index: index of the last tree. If it is already in the 53 // ensemble, its weight and num updates are adjusted. Otherwise, its weight 54 // and num updates are added as new entries to current_weights and 55 // num_updates. num_trees_to_add: how many trees are being added to the 56 // ensemble. Returns current_weights: updated vector of the tree weights. 57 // Weights of dropped trees are updated. Note that the size of returned vector 58 // will be total_num_trees + num_trees_to_add (the last elements are the 59 // weights of the new trees to be added) if new_trees_first_index 60 // >=current_weights.size num_updates: updated vector with increased number of 61 // updates for dropped trees. 62 static void GetTreesWeightsForAddingTrees( 63 const std::vector<int32>& dropped_trees, 64 const std::vector<float>& dropped_trees_original_weights, 65 const int32 new_trees_first_index, const int32 num_trees_to_add, 66 // Current weights and num_updates will be updated as a result of this 67 // func 68 std::vector<float>* current_weights, 69 // How many weight assignments have been done for each tree already. 70 std::vector<int32>* num_updates); 71 }; 72 73 } // namespace utils 74 } // namespace boosted_trees 75 } // namespace tensorflow 76 77 #endif // TENSORFLOW_CONTRIB_BOOSTED_TREES_LIB_UTILS_DROPOUT_UTILS_H_ 78