Changed parametric problems. Fixed csv export

parametric_problems/lasso.py

@@ -4,6 +4,7 @@
 import numpy as np
 import pandas as pd
 import os
+from solvers.solvers import SOLVER_MAP  # AVOID CIRCULAR DEPENDENCY
 from problem_classes.lasso import LassoExample
 from utils.general import make_sure_path_exists
 # import osqppurepy as osqp
@@ -131,7 +132,7 @@ def solve(self):
 
                 # Solve problem
                 r = m.solve()
-                
+
                 # DEBUG
                 #  print("Lambda = %.4e,\t niter = %d" % (lambda_val, r.info.iter))
 

run_parametric_problems.py

@@ -5,12 +5,11 @@
 
 '''
 
-from utils.parametric import print_results_parametric
 
 from parametric_problems.lasso import LassoParametric
 from parametric_problems.mpc import MPCParametric
 from parametric_problems.portfolio import PortfolioParametric
-
+from utils.parametric import print_results_parametric, compute_results_parametric
 
 PROBLEMS_MAP = {'Lasso': LassoParametric,
                 'MPC': MPCParametric,
@@ -23,9 +22,9 @@
             ]
 
 # Problem dimensions
-dimensions = {'Lasso': 50,
-              'MPC': 20,
-              'Portfolio': 100
+dimensions = {'Lasso': [50, 100, 150, 200],
+              'MPC': [20, 40, 60, 80],
+              'Portfolio': [100, 200, 300, 400]
               }
 
 # OSQP solver settings
@@ -35,12 +34,16 @@
 
 # Solve all problems
 for problem in problems:
-    problem_instance = PROBLEMS_MAP[problem](osqp_settings,
-                                             dimensions[problem])
-    problem_instance.solve()
-
-# Extract results info
-print("Results")
-print("-------")
-for problem in problems:
-    print_results_parametric(problem, dimensions[problem])
+    for dim in dimensions[problem]:
+        problem_instance = PROBLEMS_MAP[problem](osqp_settings,
+                                                 dim)
+        problem_instance.solve()
+
+# Compute results
+compute_results_parametric(problems, dimensions)
+
+#  # Extract results info
+#  print("Results")
+#  print("-------")
+#  for problem in problems:
+#      print_results_parametric(problem, dimensions[problem])

utils/benchmark.py

@@ -163,7 +163,7 @@ def compute_shifted_geometric_means(solvers, problems_type):
     g_mean_file = os.path.join('.', 'results',
                                problems_type,
                                'geom_mean.csv')
-    df_g_mean.to_csv(g_mean_file, header=False, index=True)
+    df_g_mean.to_frame().transpose().to_csv(g_mean_file, index=False)
 
 
     # r = {}  # Dictionary of relative times for each solver/problem
@@ -224,7 +224,7 @@ def compute_failure_rates(solvers, problems_type):
 
     # Write csv file
     df_failure_rates = pd.Series(failure_rates)
-    df_failure_rates.to_csv(failure_rates_file, header=False, index=True)
+    df_failure_rates.to_frame().transpose().to_csv(failure_rates_file, index=False)
 
 
 def compute_polish_statistics(problems_type, high_accuracy=False):
@@ -262,7 +262,7 @@ def compute_polish_statistics(problems_type, high_accuracy=False):
                          'percentage_of_success': (polish_success * 100)}
 
     df_polish = pd.Series(polish_statistics)
-    df_polish.to_csv(polish_file, header=False, index=True)
+    df_polish.to_frame().transpose().to_csv(polish_file, index=False)
 
 
 def compute_ratio_setup_solve(problems_type, high_accuracy=False):
@@ -292,7 +292,7 @@ def compute_ratio_setup_solve(problems_type, high_accuracy=False):
                    'median_ratio': np.median(ratios)}
 
     df_ratio = pd.Series(ratio_stats)
-    df_ratio.to_csv(ratio_file, header=False, index=True)
+    df_ratio.to_frame().transpose().to_csv(ratio_file, index=False)
 
 
 def compute_stats_info(solvers, benchmark_type,

utils/parametric.py

@@ -46,3 +46,54 @@ def print_results_parametric(problem, dimension):
     f.close()
 
     print("")
+
+def compute_results_parametric(problems, dimensions):
+
+    row_list = []
+    statistics_file = os.path.join('.', 'results', 'parametric_problems',
+                                   "statistics.csv")
+    for p in problems:
+
+        statistics_prob_file = os.path.join('.', 'results', 'parametric_problems',
+                                            "statistics_%s.csv" % p.lower())
+        row_list_prob = []
+        for d in dimensions[p]:
+            ws_file = os.path.join('.', 'results', 'parametric_problems',
+                                   'OSQP warmstart',
+                                   p,
+                                   'n%i.csv' % d
+                                   )
+            no_ws_file = os.path.join('.', 'results', 'parametric_problems',
+                                      'OSQP no warmstart',
+                                      p,
+                                      'n%i.csv' % d
+                                      )
+            no_ws_df = pd.read_csv(no_ws_file)
+            ws_df = pd.read_csv(ws_file)
+            dict_stats = {
+                    'problem': p,
+                    'dimension': d,
+                    'OSQP_ws_mean_time': ws_df['run_time'].mean(),
+                    'OSQP_ws_median_time': ws_df['run_time'].median(),
+                    'OSQP_ws_mean_iter': ws_df['iter'].mean(),
+                    'OSQP_ws_median_iter': ws_df['iter'].median(),
+                    'OSQP_nows_mean_time': no_ws_df['run_time'].mean(),
+                    'OSQP_nows_median_time': no_ws_df['run_time'].median(),
+                    'OSQP_nows_mean_iter': no_ws_df['iter'].mean(),
+                    'OSQP_nows_median_iter': no_ws_df['iter'].median(),
+                    'time_speedup_mean': no_ws_df['run_time'].mean()/ws_df['run_time'].mean(),
+                    'time_speedup_median': no_ws_df['run_time'].median()/ws_df['run_time'].median(),
+                    'iter_reduction_mean': no_ws_df['iter'].mean()/ws_df['iter'].mean(),
+                    'iter_reduction_median': no_ws_df['iter'].median()/ws_df['iter'].median()
+                    }
+            row_list_prob.append(dict_stats)
+
+        # Store prob statistics
+        prob_statistics = pd.DataFrame(row_list_prob)
+        prob_statistics.to_csv(statistics_prob_file)
+        row_list += row_list_prob
+
+
+    # Create statistics dict
+    statistics = pd.DataFrame(row_list)
+    statistics.to_csv(statistics_file)