SimonBlanke / Gradient-Free-Optimizers

from ._parametrize import optimizers_local

@pytest.mark.parametrize(*optimizers_local)

def test_convex_convergence_singleOpt(Optimizer):

    def objective_function(para):

        score = -(para["x1"] * para["x1"])

        return score

    search_space = {

        "x1": np.arange(-100, 101, 1),

    }

    init1 = {

        "x1": -1000,

    }

    initialize = {"warm_start": [init1]}

    n_opts = 33

    scores = []

    for rnd_st in range(n_opts):

        opt = Optimizer(search_space, initialize=initialize, rand_rest_p=1)

        opt.search(

            objective_function,

            n_iter=20,

            random_state=rnd_st,

            memory=False,

            verbosity=False,

        )

        scores.append(opt.best_score)

    score_mean = np.array(scores).mean()

    print("score_mean", score_mean)

    assert score_mean > -400

)

@pytest.mark.parametrize(*opt_local_l)

def test_local_perf(Optimizer):

    def objective_function(para):

        score = -para["x1"] * para["x1"]

        return score

    search_space = {"x1": np.arange(-100, 101, 1)}

    initialize = {"vertices": 2}

    n_opts = 33

    n_iter = 100

    scores = []

    for rnd_st in tqdm(range(n_opts)):

        opt = Optimizer(search_space, initialize=initialize)

        opt.search(

            objective_function,

            n_iter=n_iter,

            random_state=rnd_st,

            memory=False,

            verbosity=False,

        )

        scores.append(opt.best_score)

    score_mean = np.array(scores).mean()

    print("\n score_mean", score_mean)

    assert score_mean > -5