Working on improving SPEA2

Author: ajnebro

…ples/multiobjective/spea2/spea2_dtlz1.py …ples/multiobjective/spea2/spea2_dtlz2.pyexamples/multiobjective/spea2/spea2_dtlz1.py renamed to examples/multiobjective/spea2/spea2_dtlz2.py examples/multiobjective/spea2/spea2_dtlz1.py renamed to examples/multiobjective/spea2/spea2_dtlz2.py

@@ -1,7 +1,7 @@
 from jmetal.algorithm.multiobjective.spea2 import SPEA2
 from jmetal.operator.crossover import SBXCrossover
 from jmetal.operator.mutation import PolynomialMutation
-from jmetal.problem import DTLZ2
+from jmetal.problem import DTLZ2, DTLZ1, DTLZ3
 from jmetal.util.solution import (
     print_function_values_to_file,
     print_variables_to_file,
@@ -11,11 +11,11 @@
 if __name__ == "__main__":
     problem = DTLZ2()
 
-    max_evaluations = 20000
+    max_evaluations = 30000
     algorithm = SPEA2(
         problem=problem,
-        population_size=20,
-        offspring_population_size=20,
+        population_size=100,
+        offspring_population_size=100,
         mutation=PolynomialMutation(probability=1.0 / problem.number_of_variables(), distribution_index=20),
         crossover=SBXCrossover(probability=1.0, distribution_index=20),
         termination_criterion=StoppingByEvaluations(max_evaluations=max_evaluations),

examples/multiobjective/spea2/spea2_zdt1.py

@@ -2,20 +2,21 @@
 from jmetal.operator.crossover import SBXCrossover
 from jmetal.operator.mutation import PolynomialMutation
 from jmetal.problem import ZDT1
+from jmetal.algorithm.multiobjective.spea2 import SPEA2
 from jmetal.util.solution import (
     print_function_values_to_file,
     print_variables_to_file,
 )
 from jmetal.util.termination_criterion import StoppingByEvaluations
 
-if __name__ == "__main__":
+def main():
     problem = ZDT1()
 
     max_evaluations = 20000
     algorithm = SPEA2(
         problem=problem,
-        population_size=40,
-        offspring_population_size=40,
+        population_size=100,
+        offspring_population_size=100,
         mutation=PolynomialMutation(probability=1.0 / problem.number_of_variables(), distribution_index=20),
         crossover=SBXCrossover(probability=1.0, distribution_index=20),
         termination_criterion=StoppingByEvaluations(max_evaluations=max_evaluations),
@@ -31,3 +32,6 @@
     print(f"Algorithm: {algorithm.get_name()}")
     print(f"Problem: {problem.name()}")
     print(f"Computing time: {algorithm.total_computing_time}")
+
+if __name__ == "__main__":
+    main()

src/jmetal/algorithm/multiobjective/spea2.py

@@ -80,8 +80,8 @@ def replacement(self, population: List[S], offspring_population: List[S]) -> Lis
         ranking = StrengthRanking(self.dominance_comparator)
         density_estimator = KNearestNeighborDensityEstimator()
 
-        r = RankingAndDensityEstimatorReplacement(ranking, density_estimator, RemovalPolicyType.SEQUENTIAL)
-        solutions = r.replace(population, offspring_population)
+        replacement = RankingAndDensityEstimatorReplacement(ranking, density_estimator, RemovalPolicyType.SEQUENTIAL)
+        solutions = replacement.replace(population, offspring_population)
 
         return solutions
 

src/jmetal/util/density_estimator.py

@@ -4,7 +4,8 @@
 
 import numpy
 from moocore import hv_contributions
-from scipy.spatial.distance import euclidean
+import numpy as np
+from scipy.spatial.distance import cdist
 
 from jmetal.logger import get_logger
 from jmetal.util.comparator import Comparator, SolutionAttributeComparator
@@ -110,18 +111,19 @@ def compute_density_estimator(self, solutions: List[S]):
         if solutions_size <= self.k:
             return
 
-        # Compute distance matrix
-        self.distance_matrix = numpy.zeros(shape=(solutions_size, solutions_size))
-        for i in range(solutions_size):
-            for j in range(solutions_size):
-                self.distance_matrix[i, j] = self.distance_matrix[j, i] = euclidean(
-                    solutions[i].objectives, solutions[j].objectives
-                )
+        # Extract objectives as a 2D numpy array for vectorized operations
+        objectives = np.array([s.objectives for s in solutions])
+        
+        # Compute pairwise distances using cdist (much faster than nested loops)
+        self.distance_matrix = cdist(objectives, objectives, 'euclidean')
+        
+        # Get k-th nearest neighbor distance for each solution
+        # Using np.partition which is O(n) instead of full sort O(n log n)
+        k_distances = np.partition(self.distance_matrix, kth=self.k, axis=1)[:, self.k]
+        
         # Assign knn_density attribute
-        for i in range(solutions_size):
-            distances = list(self.distance_matrix[i])
-            distances.sort()
-            solutions[i].attributes["knn_density"] = distances[self.k]
+        for i, dist in enumerate(k_distances):
+            solutions[i].attributes["knn_density"] = dist
 
     def sort(self, solutions: List[S]) -> List[S]:
         """