Convergence issue: good solution appears to be forgotten, resulting best solution is not even valid

[AndreyKolomiets]

The following is my code for pygad:

    num_parents_mating = 4
    parent_selection_type = "random"
    keep_parents = -1
    crossover_type = "single_point"
    mutation_type = "random"
    mutation_percent_genes = 10
    num_edge_pairs_per_edges = 5.0
    drop_single_nodes = True
    ga_instance = pygad.GA(num_generations=num_generations,
                           num_parents_mating=num_parents_mating,
                           fitness_func=problem.fitness_func,
                           parent_selection_type=parent_selection_type,
                           keep_parents=keep_parents,
                           crossover_type=crossover_type,
                           mutation_type=mutation_type,
                           mutation_percent_genes=mutation_percent_genes,
                           random_seed=100500,
                           gene_type=int,
                           keep_elitism=0,
                           initial_population=initial_population,
                           stop_criteria=['saturate_15'],
                           logger=logger,
                           save_solutions=False,
                           on_fitness=on_fitness,
                           )

problem is class instance, it is computing fitness as well as logging and saving some data. Best solution and best fitness are stored in best_solution_cls and best_fitness_cls attributes respectively. There is a strange bug in my code, that best_solution_cls doesn't match to best_fitness_cls. While best_fitness_cls is indeed the best fitness across all tries, best_fitness_cls is some other solution, it is usually not even valid (for invalid solutions, fitness=-1000.0 is expected).
The following is conda_env.yml used to create environment:

channels:
  - pytorch
  - anaconda
  - conda-forge
  - defaults
dependencies:
  - python=3.7.15=haa1d7c7_0
  - pip=22.2.2=py37h06a4308_0
  - setuptools=65.5.0=py37h06a4308_0
  - pip:
    - numpy==1.21.6
    - pandas==1.3.5
    - scipy==1.7.3
    - tqdm==4.64.1
    - networkx==2.6.3
    - numba==0.56.4
    - pygad==3.2.0

[ahmedfgad]

Hi @AndreyKolomiets,
Since such issues cannot easily reproduced without having a full working code, I appreciate if you would share a sample that reproduces the issue. This will heavily help debug the issue.