add regression example

generate_cpp_code.py

@@ -2,13 +2,16 @@
 import argparse
 import re
 
-def get_single_booster_cpp_code(booster_tree, branch_id, class_index, indentation_level=0):
+def get_single_booster_cpp_code(booster_tree, branch_id, class_index, indentation_level=0, model_type='classification'):
     level = booster_tree[branch_id].split()
 
     booster_code = ""
 
     if 'leaf' in level[0]:
-        booster_code += "{0}sum[{1}] += {2};\n".format("  " * indentation_level, class_index, float(level[0].split('=')[1]))
+        if model_type == 'classification':
+            booster_code += "{0}sum[{1}] += {2};\n".format("  " * indentation_level, class_index, float(level[0].split('=')[1]))
+        elif model_type == 'regression':
+            booster_code += "{0}sum += {1};\n".format("  " * indentation_level, float(level[0].split('=')[1]))
         return booster_code
 
     branch_ids = level[1].split(',')
@@ -21,37 +24,46 @@ def get_single_booster_cpp_code(booster_tree, branch_id, class_index, indentatio
     comparison = re.search('[^0-9a-zA-Z:[]+[0-9]*[0-9.]*', level[0]).group(0)
 
     booster_code += "{0}if (sample[{1}] {2}) {{\n".format("  " * indentation_level, feature_index, comparison)
-    booster_code += get_single_booster_cpp_code(booster_tree, yes_branch_id, class_index, indentation_level + 1)
+    booster_code += get_single_booster_cpp_code(booster_tree, yes_branch_id, class_index, indentation_level + 1, model_type)
     booster_code += "{0}}} else {{\n".format("  " * indentation_level)
-    booster_code += get_single_booster_cpp_code(booster_tree, no_branch_id, class_index, indentation_level + 1)
+    booster_code += get_single_booster_cpp_code(booster_tree, no_branch_id, class_index, indentation_level + 1, model_type)
     booster_code += "{0}}}\n".format("  " * indentation_level)
 
     return booster_code
 
-def generate_single_booster_cpp_code(booster, class_index):
+def generate_single_booster_cpp_code(booster, class_index, model_type):
     booster_tree = dict()
     for line in booster:
         branch_id = int(line.split(':')[0].strip())
         booster_tree[branch_id] = line
 
-    return get_single_booster_cpp_code(booster_tree, 0, class_index, 1)
+    return get_single_booster_cpp_code(booster_tree, 0, class_index, 1, model_type)
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
     parser.add_argument('--xgb_dump', type=str, default='dump.raw.txt', help='Raw boosters dump. Created without passing feature map file to XGBoost dump() function.')
     parser.add_argument('--num_classes', type=int, required=True, help='number of classes this model is classyfing')
+    parser.add_argument('--model_type', type=str, required=True, help='classification/regression')
 
     args = parser.parse_args()
 
     result = ""
-
-    result += "#include \"xgboost_classifier.h\"\n"
+    if args.model_type == 'classification':
+        result += "#include \"xgboost_classifier.h\"\n"
+    else:
+        result += "#include \"xgboost_regressor.h\"\n"
     result += "#include <iostream>\n"
     result += "#include <fstream>\n"
     result += "#include <vector>\n"
     result += "using namespace std;\n\n"
-    result += "std::vector<float> xgb_classify(std::vector<float> &sample) {\n\n"
-    result += "  std::vector<float> sum ({0}, 0.0);\n\n".format(args.num_classes)
+    if args.model_type == 'classification':
+        result += "std::vector<float> xgb_classify(std::vector<float> &sample) {\n\n"
+        result += "  std::vector<float> sum ({0}, 0.0);\n\n".format(args.num_classes)
+    elif args.model_type == 'regression':
+        result += "float xgb_regress(std::vector<float> &sample) {\n\n"
+        result += "  float sum = 0;"
+    else:
+        raise Exception("Please use model_type as classification/regression")
 
     booster_counter = 0
     boosters = []
@@ -64,12 +76,17 @@ def generate_single_booster_cpp_code(booster, class_index):
                 boosters[booster_counter - 1].append(line.strip())
 
     for index, booster in enumerate(boosters):
-        class_index = index % args.num_classes
-        result += generate_single_booster_cpp_code(booster, class_index)
+        if args.model_type == 'classification':
+            class_index = index % args.num_classes
+        else:
+            class_index = 1
+        # result += generate_single_booster_cpp_code(booster, class_index, model_type='classification')
+        result += generate_single_booster_cpp_code(booster, class_index, model_type=args.model_type)
         result += "\n\n"
 
     result += "  return sum;\n"
     result += "}\n\n"
 
-    with open('xgboost_classifier.cpp', 'w') as f:
+    cpp_file_name = 'xgboost_classifier.cpp' if args.model_type == 'classification' else 'xgboost_regressor.cpp'
+    with open(cpp_file_name, 'w') as f:
         f.write(result)

sample/codeblocks_project/main.cpp

@@ -1,5 +1,11 @@
+// g++ -c xgboost_classifier.cpp -o xgb_c.o
+// g++ -c xgboost_regressor.cpp -o xgb_r.o
+// g++ -std=c++11 main.cpp -I./ -o test xgb_c.o xgb_r.o
+// ./test
+
 #include <iostream>
 #include "xgboost_classifier.h"
+#include "xgboost_regressor.h"
 #include <chrono>
 #include <math.h>
 
@@ -60,5 +66,30 @@ int main()
     std::cout << "Classified " << NUM_SAMPLES << " samples in " << time_diff << " miliseconds." << std::endl;
     std::cout << time_diff / NUM_SAMPLES << "ms per sample." << std::endl;
 
+    // Regression test
+    cout << "Start regression." << endl;
+
+    test_sample = {
+        -0.0016666666666666904, -0.0668518518518519, 0.018124999999999978,
+       0.10351851851851848, -0.013125000000000024, 0.09703703703703699,
+       0.021249999999999977, -0.06129629629629634, -0.024583333333333356,
+       -0.07240740740740745, 1.250996173937358, 1.5632783693080987,
+       0.3273181103443364, 0.23735363332356418, 1.6726289334343811,
+       1.2316100868318476, 0.20377132592362593, 1.027838760908221,
+       1.9099825667579455, 0.2446893370421753, 1.3379751193714342,
+       1.5589281971761841, 0.24330398118549879, 1.315624215990686,
+       1.5826644564136092, 1.4901161193847656e-08, 1.4901161193847656e-08,
+       3.141592638688632, 1.3185890322659237, 0.3346538140629468,
+       1.4883498072609227, 1.8065823504935967, 0.2877854550824648,
+       1.0472248480137314, 1.6726289334343811, 1.027838760908221,
+       0.44112495924719125, 1.3379751193714342, 1.315624215990686,
+       0.48799331822767306, 0.03125, 0.006481481481481481,
+       0.04583333333333333, 0.011111111111111112, 0.9662725445360024,
+       1.4702680482603403, 1, 1, 1, 1
+    };
+
+    float reg_result = xgb_regress(test_sample);
+    cout << "Regression result:" << reg_result << endl;
+
     return 1;
 }