add feature importances (#42)

add feature importances (#42)

include/LightGBM/tree.h

@@ -83,6 +83,9 @@ public:
   /*! \brief Get depth of specific leaf*/
   inline int leaf_depth(int leaf_idx) const { return leaf_depth_[leaf_idx]; }
 
+  /*! \brief Get feature of specific split*/
+  inline int split_feature(int split_idx) const { return split_feature_[split_idx]; }
+
   /*!
   * \brief Shrinkage for the tree's output
   *        shrinkage rate (a.k.a learning rate) is used to tune the traning process

src/boosting/gbdt.cpp

@@ -209,6 +209,7 @@ void GBDT::Train() {
     if (is_early_stopping) {
         // close file with an early-stopping message
         Log::Info("Early stopping at iteration %d, the best iteration round is %d", iter + 1, iter + 1 - early_stopping_round_);
+        FeatureImportance(iter - early_stopping_round_ + 1);
         fclose(output_model_file);
         return;
     }
@@ -222,6 +223,7 @@ void GBDT::Train() {
       }
       fflush(output_model_file);
   }
+  FeatureImportance(models_.size());
   fclose(output_model_file);
 }
 
@@ -349,6 +351,19 @@ void GBDT::ModelsFromString(const std::string& model_str, int num_used_model) {
   Log::Info("%d models has been loaded\n", models_.size());
 }
 
+void GBDT::FeatureImportance(const int last_iter) {
+    size_t* feature_importances = new size_t[max_feature_idx_ + 1]{0};
+    for (int iter = 0; iter < last_iter; ++iter) {
+        for (int split_idx = 0; split_idx < models_.at(iter)->num_leaves() - 1; ++split_idx) {
+            ++feature_importances[models_.at(iter)->split_feature(split_idx)];
+        }
+    }
+    std::string ret = Common::ArrayToString(feature_importances, max_feature_idx_ + 1, ' ');
+    fprintf(output_model_file, "feature importances=%s\n", ret.c_str());
+    fflush(output_model_file);
+    delete[] feature_importances;
+}
+
 double GBDT::PredictRaw(const double* value) const {
   double ret = 0.0;
   for (size_t i = 0; i < models_.size(); ++i) {

src/boosting/gbdt.h

@@ -119,8 +119,11 @@ private:
   * \param iter Current interation
   */
   bool OutputMetric(int iter);
-  
-  int early_stopping_round_;
+  /*!
+  * \brief Calculate feature importances
+  * \param last_iter Last tree use to calculate
+  */
+  void FeatureImportance(const int last_iter);
   
   /*! \brief Pointer to training data */
   const Dataset* train_data_;
@@ -138,6 +141,8 @@ private:
   std::vector<ScoreUpdater*> valid_score_updater_;
   /*! \brief Metric for validation data */
   std::vector<std::vector<const Metric*>> valid_metrics_;
+  /*! \brief Number of rounds for early stopping */
+  int early_stopping_round_;
   /*! \brief Best score(s) for early stopping */
   std::vector<std::vector<int>> best_iter_;
   std::vector<std::vector<score_t>> best_score_;