Merge pull request #3 from ottenbreit-data-science/bugfix

fixed bug in bin splitting

Author: mathias-von-ottenbreit

cpp/main.cpp

@@ -18,7 +18,7 @@ int main()
     model.n_jobs=0;
     model.loss_function_mse=true;
     model.verbosity=3;
-    model.min_observations_in_split=30;
+    model.min_observations_in_split=10;
     //model.max_interaction_level=0;
     model.max_interaction_level=100;
     model.max_interactions=30;
@@ -46,10 +46,10 @@ int main()
     //Saving results
     save_data("output.csv",predictions);
     std::cout<<"min validation_error "<<model.validation_error_steps.minCoeff()<<"\n\n";
-    std::cout<<check_if_approximately_equal(model.validation_error_steps.minCoeff(),6.39607,0.00001)<<"\n";
+    std::cout<<check_if_approximately_equal(model.validation_error_steps.minCoeff(),6.01566,0.00001)<<"\n";
 
     std::cout<<"mean prediction "<<predictions.mean()<<"\n\n";
-    std::cout<<check_if_approximately_equal(predictions.mean(),23.7461,0.0001)<<"\n";
+    std::cout<<check_if_approximately_equal(predictions.mean(),23.8349,0.0001)<<"\n";
 
     std::cout<<"best_m: "<<model.m<<"\n";
 

cpp/term.h

@@ -361,30 +361,22 @@ void Term::setup_bins()
         auto ip{std::unique(values_sorted_unique.begin(),values_sorted_unique.end())};
         values_sorted_unique.resize(std::distance(values_sorted_unique.begin(),ip));
 
-        bins_start_index.reserve(bins+1);    
-        bins_end_index.reserve(bins+1);  
-        //Allocations
-        bool eligible_spacing{false};
-        bool eligible_unique_numbers{false};
+        bins_start_index.reserve(bins+1);
+        bins_end_index.reserve(bins+1);
         //Start_index
         bins_start_index.push_back(0);
         if(bins>1)
         {
-            for (size_t i = min_observations_in_split-1; i <= max_index; ++i) //for each observation in an allowable range
+            for (size_t i = min_observations_in_split-1; i <= max_index+1-min_observations_in_split; ++i) //for each observation in an allowable range
             {
-                //General eligibility
-                if(i>=min_observations_in_split-1 && (i%observations_in_bin==0 || values_sorted_unique.size()<=bins) && i<=max_index+1-min_observations_in_split) eligible_spacing=true;
+                size_t last_bin_start_index{bins_start_index[bins_start_index.size()-1]};
+                bool eligible_on_spacing_between_observations{i >= last_bin_start_index + observations_in_bin || values_sorted_unique.size()<=bins};
+                bool eligible_on_unique_numbers{i>0 && !check_if_approximately_equal(sorted_vectors.values_sorted[i],sorted_vectors.values_sorted[i-1])};
 
-                //Eligibility when considering unique numbers
-                if(i>0 && !check_if_approximately_equal(sorted_vectors.values_sorted[i],sorted_vectors.values_sorted[i-1])) eligible_unique_numbers=true;
-                else eligible_unique_numbers=false;
-
-                //Creating bin if possible and resetting eligibility for next iteration
-                if(eligible_spacing && eligible_unique_numbers) 
+                bool create_bin{eligible_on_spacing_between_observations && eligible_on_unique_numbers};
+                if(create_bin) 
                 {
                     bins_start_index.push_back(i);
-                    eligible_spacing=false;
-                    eligible_unique_numbers=false;
                 }
             }
         }

setup.py

@@ -15,7 +15,7 @@
 
 setuptools.setup(
     name='aplr',
-    version='1.0.3',
+    version='1.0.4',
     description='Automatic Piecewise Linear Regression',
     ext_modules=[sfc_module],
     author="Mathias von Ottenbreit",