Merge pull request #1 from ottenbreit-data-science/finishing

Modified numpy version requirement. Better cleanup after fit.

Author: mathias-von-ottenbreit

cpp/APLRRegressor.h

@@ -422,6 +422,7 @@ void APLRRegressor::determine_interactions_to_consider()
                 if(!(terms[sorted_latest_split_point_errors_indices[i]]==interaction))
                 {
                     interaction.given_terms.push_back(terms[sorted_latest_split_point_errors_indices[i]]);
+                    interaction.given_terms[interaction.given_terms.size()-1].cleanup_when_this_term_was_added_as_a_given_predictor();
                     bool already_exists{false};
                     for (size_t k = 0; k < terms_eligible_current.size(); ++k)
                     {
@@ -836,6 +837,10 @@ void APLRRegressor::cleanup_after_fit()
     distributed_terms.clear();
     interactions_to_consider.clear();
     error_index_for_interactions_to_consider.resize(0);
+    for (size_t i = 0; i < terms.size(); ++i)
+    {
+        terms[i].cleanup_after_fit();
+    }
 }
 
 VectorXd APLRRegressor::predict(const MatrixXd &X)

cpp/term.h

@@ -53,7 +53,9 @@ class Term
     void estimate_split_point_on_discretized_data();
     void calculate_coefficient_and_error_on_discretized_data(bool direction_right, double split_point);
     void estimate_coefficient_and_error_on_all_data();
-    void clean_up_after_estimate_split_point();
+    void cleanup_after_estimate_split_point();
+    void cleanup_after_fit();
+    void cleanup_when_this_term_was_added_as_a_given_predictor();
 
 public:
     //fields
@@ -171,7 +173,7 @@ void Term::estimate_split_point(const MatrixXd &X,const VectorXd &y,const Vector
     discretize_data_by_bin();
     estimate_split_point_on_discretized_data();
     estimate_coefficient_and_error_on_all_data();
-    clean_up_after_estimate_split_point();
+    cleanup_after_estimate_split_point();
 }
 
 //Calculate indices that get zeroed out during calculate() because of given terms. Also calculates indices of those observations that do not.
@@ -582,7 +584,7 @@ void Term::estimate_coefficient_and_error_on_all_data()
     }
 }
 
-void Term::clean_up_after_estimate_split_point()
+void Term::cleanup_after_estimate_split_point()
 {
     given_terms_indices.not_zeroed.resize(0);
     given_terms_indices.zeroed.resize(0);
@@ -593,6 +595,22 @@ void Term::clean_up_after_estimate_split_point()
     errors_initial.resize(0);
 }
 
+void Term::cleanup_after_fit()
+{
+    bins_start_index.clear();
+    bins_end_index.clear();
+    bins_split_points_left.clear();
+    bins_split_points_right.clear();
+    values_discretized.resize(0);
+    sample_weight_discretized.resize(0);
+}
+
+void Term::cleanup_when_this_term_was_added_as_a_given_predictor()
+{
+    cleanup_after_fit();
+    coefficient_steps.resize(0);
+}
+
 VectorXd Term::calculate_prediction_contribution(const MatrixXd &X)
 {
     VectorXd values{calculate(X)};

setup.py

@@ -15,15 +15,15 @@
 
 setuptools.setup(
     name='aplr',
-    version='1.0.1',
+    version='1.0.2',
     description='Automatic Piecewise Linear Regression',
     ext_modules=[sfc_module],
     author="Mathias von Ottenbreit",
     author_email="[email protected]",
     long_description="Build predictive and interpretable parametric machine learning models in Python based on the Automatic Piecewise Linear Regression methodology developed by Mathias von Ottenbreit.",
     long_description_content_type="text/markdown",
     packages=['aplr'],
-    install_requires=["numpy"],       
+    install_requires=["numpy>=1.20"],       
     python_requires='>=3.8',
     classifiers=["License :: OSI Approved :: MIT License"],
     license="MIT",