SimonBlanke /
Gradient-Free-Optimizers
| @@ 7-41 (lines=35) @@ | ||
| 4 | from ._parametrize import optimizers_local |
|
| 5 | ||
| 6 | ||
| 7 | @pytest.mark.parametrize(*optimizers_local) |
|
| 8 | def test_convex_convergence_singleOpt(Optimizer): |
|
| 9 | def objective_function(para): |
|
| 10 | score = -(para["x1"] * para["x1"]) |
|
| 11 | return score |
|
| 12 | ||
| 13 | search_space = { |
|
| 14 | "x1": np.arange(-1000, 1, 1), |
|
| 15 | } |
|
| 16 | ||
| 17 | init1 = { |
|
| 18 | "x1": -1000, |
|
| 19 | } |
|
| 20 | initialize = {"warm_start": [init1]} |
|
| 21 | ||
| 22 | n_opts = 33 |
|
| 23 | ||
| 24 | scores = [] |
|
| 25 | for rnd_st in range(n_opts): |
|
| 26 | opt = Optimizer(search_space, rand_rest_p=1) |
|
| 27 | opt.search( |
|
| 28 | objective_function, |
|
| 29 | n_iter=30, |
|
| 30 | random_state=rnd_st, |
|
| 31 | memory=False, |
|
| 32 | verbosity=False, |
|
| 33 | initialize=initialize, |
|
| 34 | ) |
|
| 35 | ||
| 36 | scores.append(opt.best_score) |
|
| 37 | score_mean = np.array(scores).mean() |
|
| 38 | ||
| 39 | print("score_mean", score_mean) |
|
| 40 | ||
| 41 | assert score_mean > -10000 |
|
| 42 | ||
| 43 | ||
| @@ 41-67 (lines=27) @@ | ||
| 38 | assert score_mean > -25 |
|
| 39 | ||
| 40 | ||
| 41 | @pytest.mark.parametrize(*optimizers_PopBased) |
|
| 42 | def test_convex_convergence_popBased(Optimizer): |
|
| 43 | def objective_function(para): |
|
| 44 | score = -para["x1"] * para["x1"] |
|
| 45 | return score |
|
| 46 | ||
| 47 | search_space = {"x1": np.arange(-100, 101, 1)} |
|
| 48 | initialize = {"vertices": 2, "grid": 2} |
|
| 49 | ||
| 50 | n_opts = 33 |
|
| 51 | ||
| 52 | scores = [] |
|
| 53 | for rnd_st in tqdm(range(n_opts)): |
|
| 54 | opt = Optimizer(search_space) |
|
| 55 | opt.search( |
|
| 56 | objective_function, |
|
| 57 | n_iter=80, |
|
| 58 | random_state=rnd_st, |
|
| 59 | memory=False, |
|
| 60 | verbosity=False, |
|
| 61 | initialize=initialize, |
|
| 62 | ) |
|
| 63 | ||
| 64 | scores.append(opt.best_score) |
|
| 65 | score_mean = np.array(scores).mean() |
|
| 66 | ||
| 67 | assert score_mean > -25 |
|
| 68 | ||
| 69 | ||
| 70 | @pytest.mark.parametrize(*optimizers_SBOM) |
|
| @@ 12-38 (lines=27) @@ | ||
| 9 | ) |
|
| 10 | ||
| 11 | ||
| 12 | @pytest.mark.parametrize(*optimizers_singleOpt) |
|
| 13 | def test_convex_convergence_singleOpt(Optimizer): |
|
| 14 | def objective_function(para): |
|
| 15 | score = -para["x1"] * para["x1"] |
|
| 16 | return score |
|
| 17 | ||
| 18 | search_space = {"x1": np.arange(-100, 101, 1)} |
|
| 19 | initialize = {"vertices": 1} |
|
| 20 | ||
| 21 | n_opts = 33 |
|
| 22 | ||
| 23 | scores = [] |
|
| 24 | for rnd_st in tqdm(range(n_opts)): |
|
| 25 | opt = Optimizer(search_space) |
|
| 26 | opt.search( |
|
| 27 | objective_function, |
|
| 28 | n_iter=100, |
|
| 29 | random_state=rnd_st, |
|
| 30 | memory=False, |
|
| 31 | verbosity=False, |
|
| 32 | initialize=initialize, |
|
| 33 | ) |
|
| 34 | ||
| 35 | scores.append(opt.best_score) |
|
| 36 | score_mean = np.array(scores).mean() |
|
| 37 | ||
| 38 | assert score_mean > -25 |
|
| 39 | ||
| 40 | ||
| 41 | @pytest.mark.parametrize(*optimizers_PopBased) |
|
