improved convergence when family is mae

Author: mathias-von-ottenbreit

cpp/APLRRegressor.h

@@ -63,7 +63,7 @@ class APLRRegressor
     void initialize(const std::vector<size_t> &prioritized_predictors_indexes, const std::vector<int> &monotonic_constraints);
     bool check_if_base_term_has_only_one_unique_value(size_t base_term);
     void add_term_to_terms_eligible_current(Term &term);
-    VectorXd calculate_neg_gradient_current();
+    VectorXd calculate_neg_gradient_current(const VectorXd &sample_weight_train);
     void execute_boosting_steps();
     void execute_boosting_step(size_t boosting_step);
     std::vector<size_t> find_terms_eligible_current_indexes_for_a_base_term(size_t base_term);
@@ -580,7 +580,7 @@ void APLRRegressor::add_term_to_terms_eligible_current(Term &term)
     terms_eligible_current.push_back(term);
 }
 
-VectorXd APLRRegressor::calculate_neg_gradient_current()
+VectorXd APLRRegressor::calculate_neg_gradient_current(const VectorXd &sample_weight_train)
 {
     VectorXd output;
     if(family=="gaussian")
@@ -609,7 +609,10 @@ VectorXd APLRRegressor::calculate_neg_gradient_current()
         }
     }
     else if(family=="mae")
-        output=(y_train.array() - predictions_current.array()).sign();
+    {
+        double mae{calculate_errors(y_train,predictions_current,sample_weight_train,"mae").mean()};
+        output=(y_train.array() - predictions_current.array()).sign()*mae;
+    }
 
     if(link_function!="identity")
         output=output.array()*differentiate_predictions().array();
@@ -700,7 +703,7 @@ void APLRRegressor::update_linear_predictor_and_predictions()
 
 void APLRRegressor::update_gradient_and_errors()
 {
-    neg_gradient_current=calculate_neg_gradient_current();
+    neg_gradient_current=calculate_neg_gradient_current(sample_weight_train);
     neg_gradient_nullmodel_errors_sum=calculate_sum_error(calculate_errors(neg_gradient_current,linear_predictor_null_model,sample_weight_train,FAMILY_GAUSSIAN));
 }
 

cpp/test ALRRegressor mae.cpp

@@ -40,9 +40,8 @@ int main()
 
     //Fitting
     //model.fit(X_train,y_train);
-    //model.fit(X_train,y_train,sample_weight);
-    //model.fit(X_train,y_train,sample_weight,{},{0,1,2,3,4,5,10,static_cast<size_t>(y_train.size()-1)});
     model.fit(X_train,y_train,sample_weight);
+    //model.fit(X_train,y_train,sample_weight,{},{0,1,2,3,4,5,10,static_cast<size_t>(y_train.size()-1)});
     std::cout<<"feature importance\n"<<model.feature_importance<<"\n\n";
 
     VectorXd predictions{model.predict(X_test)};
@@ -52,7 +51,7 @@ int main()
     save_as_csv_file("data/output.csv",predictions);
 
     std::cout<<predictions.mean()<<"\n\n";
-    tests.push_back(is_approximately_equal(predictions.mean(),23.5594,0.00001));
+    tests.push_back(is_approximately_equal(predictions.mean(),23.6839,0.00001));
 
     //Test summary
     std::cout<<"\n\nTest summary\n"<<"Passed "<<std::accumulate(tests.begin(),tests.end(),0)<<" out of "<<tests.size()<<" tests.";

setup.py

@@ -15,7 +15,7 @@
 
 setuptools.setup(
     name='aplr',
-    version='2.4.0',
+    version='2.5.0',
     description='Automatic Piecewise Linear Regression',
     ext_modules=[sfc_module],
     author="Mathias von Ottenbreit",