.3106580201476295:57e7867912cbe2694e6be58bfe7e8206_69f5c75ede92085ec783325d.69f5cf02de92085ec7833419.69f5cf01fbc5f8462304873a:Trae CN.T(2026/5/2 18:16:34)

pygad/islandga.py

@@ -237,7 +237,9 @@ def _replace_individuals(self, ga_instance, fitness_values, new_individuals, new
 
     def _perform_migration(self):
         """执行迁移操作"""
-        all_migrants = []
+        from collections import defaultdict
+
+        migrants_by_destination = defaultdict(list)
         modified_islands = set()
 
         for island_idx, ga_instance in enumerate(self.islands):
@@ -248,24 +250,30 @@ def _perform_migration(self):
             for dest_idx in destinations:
                 modified_islands.add(dest_idx)
                 for migrant_idx, migrant in enumerate(migrants):
-                    all_migrants.append({
+                    migrants_by_destination[dest_idx].append({
                         'source': island_idx,
-                        'destination': dest_idx,
                         'individual': migrant.copy(),
                         'fitness': migrant_fitness[migrant_idx]
                     })
 
-        for migration in all_migrants:
-            dest_idx = migration['destination']
+        for dest_idx, incoming_migrants in migrants_by_destination.items():
             dest_ga = self.islands[dest_idx]
-            dest_fitness = dest_ga.last_generation_fitness
+            dest_fitness = dest_ga.last_generation_fitness.copy()
 
-            self._replace_individuals(
-                dest_ga, 
-                dest_fitness, 
-                [migration['individual']],
-                [migration['fitness']]
-            )
+            new_individuals = [m['individual'] for m in incoming_migrants]
+            new_fitness = [m['fitness'] for m in incoming_migrants]
+
+            num_to_replace = min(len(new_individuals), len(dest_fitness))
+
+            if self.replacement == 'worst':
+                sorted_indices = numpy.argsort(dest_fitness)
+                replace_indices = sorted_indices[:num_to_replace]
+            else:
+                replace_indices = random.sample(range(len(dest_fitness)), num_to_replace)
+
+            for i, idx in enumerate(replace_indices):
+                if i < len(new_individuals):
+                    dest_ga.population[idx] = new_individuals[i].copy()
 
         for island_idx in modified_islands:
             self.islands[island_idx].last_generation_fitness = self.islands[island_idx].cal_pop_fitness()

tests/test_islandga.py

@@ -541,6 +541,76 @@ def test_fully_connected_topology_paths():
 
     print("  ✓ 全连接拓扑路径正确 (每个岛屿连接到所有其他岛屿)")
 
+def test_multiple_sources_to_single_destination_no_overwrite():
+    """
+    关键测试：验证全连接拓扑下多个源岛向同一个目标岛迁移时，
+    移民不会互相覆盖。
+
+    测试逻辑：
+    1. 创建 4 个岛屿，全连接拓扑
+    2. 在岛屿 0、1、2 中分别放入具有唯一标识的高适应度个体
+    3. 执行一次迁移，这 3 个岛的移民都会进入岛屿 3
+    4. 验证岛屿 3 的种群中确实存在这 3 个不同的移民，
+       而不是只保留最后到达的那一个
+    """
+    print("\n测试: 多源到单目标迁移不覆盖...")
+
+    island_ga = pygad.IslandGA(
+        num_islands=4,
+        num_generations=5,
+        num_parents_mating=2,
+        fitness_func=fitness_func,
+        sol_per_pop=10,
+        num_genes=4,
+        migration_frequency=10,
+        num_migrants=1,
+        migration_topology='fully_connected',
+        migrant_selection='best',
+        replacement='worst',
+        random_seed=random_seed,
+        suppress_warnings=True
+    )
+
+    for ga_instance in island_ga.islands:
+        ga_instance.initialize_population(
+            allow_duplicate_genes=ga_instance.allow_duplicate_genes,
+            gene_type=ga_instance.gene_type,
+            gene_constraint=ga_instance.gene_constraint
+        )
+
+    unique_solution_0 = numpy.array([100.0, 0.0, 0.0, 0.0])
+    unique_solution_1 = numpy.array([0.0, 100.0, 0.0, 0.0])
+    unique_solution_2 = numpy.array([0.0, 0.0, 100.0, 0.0])
+
+    island_ga.islands[0].population[0] = unique_solution_0.copy()
+    island_ga.islands[1].population[0] = unique_solution_1.copy()
+    island_ga.islands[2].population[0] = unique_solution_2.copy()
+
+    for i in range(4):
+        island_ga.islands[i].last_generation_fitness = island_ga.islands[i].cal_pop_fitness()
+
+    island_3_before = [list(sol) for sol in island_ga.islands[3].population]
+
+    island_ga._perform_migration()
+
+    island_3_after = [list(sol) for sol in island_ga.islands[3].population]
+
+    found_0 = any(numpy.allclose(sol, unique_solution_0, atol=1e-10) for sol in island_ga.islands[3].population)
+    found_1 = any(numpy.allclose(sol, unique_solution_1, atol=1e-10) for sol in island_ga.islands[3].population)
+    found_2 = any(numpy.allclose(sol, unique_solution_2, atol=1e-10) for sol in island_ga.islands[3].population)
+
+    found_count = sum([found_0, found_1, found_2])
+
+    print(f"  岛屿 3 迁移前种群:\n    {[list(s) for s in island_3_before[:3]]}...")
+    print(f"  岛屿 3 迁移后种群 (前5个):\n    {[list(s) for s in island_ga.islands[3].population[:5]]}")
+    print(f"  找到的唯一移民: 岛屿0={found_0}, 岛屿1={found_1}, 岛屿2={found_2}")
+
+    assert found_count >= 2, \
+        f"全连接拓扑下 3 个源岛向目标岛迁移，至少应该找到 2 个不同的移民，但只找到 {found_count} 个。" \
+        f"这说明移民可能被覆盖了！"
+
+    print("  ✓ 多源到单目标迁移无覆盖问题")
+
 if __name__ == "__main__":
     print("=" * 60)
     print("IslandGA 单元测试")
@@ -560,6 +630,7 @@ def test_fully_connected_topology_paths():
     test_ring_topology_paths()
     test_star_topology_paths()
     test_fully_connected_topology_paths()
+    test_multiple_sources_to_single_destination_no_overwrite()
 
     print("\n" + "=" * 60)
     print("所有测试通过! ✓")