update pytest

openbox/core/ea/adaptive_ea_advisor.py

@@ -51,7 +51,7 @@ def __init__(self,
         self.last_observations = []
 
     def get_suggestion(self):
-        if not self.last_suggestions:
+        while not self.last_suggestions:
             self.update_observations(self.last_observations)
             self.last_observations = []
             self.last_suggestions = self.get_suggestions()

test/acq_optimizer/test_random_configuration_chooser.py

@@ -13,26 +13,41 @@ def test_no_cool_down_chooser_returns_expected_result(self):
         self.assertTrue(chooser.check(2))
         self.assertFalse(chooser.check(3))
 
+        chooser2 = ChooserNoCoolDown(1.0)
+        self.assertTrue(chooser2.modulus == 1.0)
+
     def test_linear_cool_down_chooser_returns_expected_result(self):
         chooser = ChooserLinearCoolDown(2.0, 0.3, np.inf)
         self.assertTrue(chooser.check(2))
         self.assertFalse(chooser.check(3))
 
+        chooser.next_smbo_iteration()
+        self.assertTrue(chooser.check(3))
+
     def test_prob_chooser_returns_expected_result(self):
         chooser = ChooserProb(0.5, self.rng)
         self.assertTrue(chooser.check(1))
         self.assertFalse(chooser.check(2))
 
     def test_prob_cool_down_chooser_returns_expected_result(self):
-        chooser = ChooserProbCoolDown(0.5, 0.1, self.rng)
+        chooser = ChooserProbCoolDown(0.5, 2, self.rng)
         self.assertTrue(chooser.check(1))
         self.assertFalse(chooser.check(2))
 
+        chooser.next_smbo_iteration()
+        self.assertTrue(chooser.check(3))
+
     def test_cosine_annealing_chooser_returns_expected_result(self):
         chooser = ChooserCosineAnnealing(0.5, 0.1, 2, self.rng)
         self.assertTrue(chooser.check(1))
         self.assertFalse(chooser.check(2))
 
+        chooser.next_smbo_iteration()
+        self.assertTrue(chooser.iteration == 1)
+        chooser.next_smbo_iteration()
+        chooser.next_smbo_iteration()
+        self.assertTrue(chooser.iteration == 0)
+
 
 if __name__ == '__main__':
     unittest.main()

test/acquisition_function/test_acquisition.py

@@ -12,7 +12,15 @@
     LCB,
     Uncertainty,
 )
-
+'''
+加入其他函数的测试
+'''
+from openbox.acquisition_function.multi_objective_acquisition import (
+    MaxvalueEntropySearch,
+    MESMO,
+    MESMOC,
+    MESMOC2,
+)   
 class ConfigurationMock:
     def __init__(self, values=None):
         self.values = values
@@ -22,6 +30,13 @@ def get_array(self):
         return np.array(self.values, dtype=np.float64)
 
 class MockModel:
+    '''
+    其实axis的重点在于方向，而不是行和列。具体到各种用法而言也是如此。
+    当axis=1时，如果是求平均，那么是从左到右横向求平均；如果是拼接，那么也是左右横向拼接；如果是drop，那么也是横向发生变化，体现为列的减少。
+    mean_col_vector = np.mean(X, axis=1).reshape(-1, 1)
+return mean_col_vector, mean_col_vector
+    这里代码重复了？
+    '''
     def predict_marginalized_over_instances(self, X):
         return np.array([np.mean(X, axis=1).reshape((1, -1))]).reshape((-1, 1)), np.array(
             [np.mean(X, axis=1).reshape((1, -1))]
@@ -45,11 +60,30 @@ def __init__(self, num_targets=1):
         self.num_targets = num_targets
 
     def predict_marginalized_over_instances(self, X):
+        '''
+        这个乘法真的没有问题？不是引用？
+        np.full(X.shape[0], 0.0)：创建一个与 X 的行数相同的全 0 数组。
+        >>> b=[a]
+        >>> b
+        [array([1, 2])]
+        >>> b*2
+        [array([1, 2]), array([1, 2])]
+        >>> c=b*2
+        >>> c
+        [array([1, 2]), array([1, 2])]
+        >>> c[0][0]=2
+        >>> c
+        [array([2, 2]), array([2, 2])]
+        >>>
+        '''
         return np.array([np.mean(X, axis=1).reshape((1, -1))] * self.num_targets).reshape((-1, 2)), np.array(
             [np.full(X.shape[0], 0.0).reshape((1, -1))] * self.num_targets
         ).reshape((-1, 2))
 
 class MockConstraintModel:
+    '''
+    np.full(m.shape, 0.1)：创建一个与 m 形状相同的数组，数组中的每个元素都为 0.1
+    '''
     def predict_marginalized_over_instances(self, X):
         m = -np.abs(np.mean(X, axis=1)).reshape((-1, 1))
         v = np.full(m.shape, 0.1)
@@ -60,6 +94,7 @@ def predict_marginalized_over_instances(self, X):
         return np.full(X.shape[0], np.nan).reshape((-1, 1)), np.full(X.shape[0], np.nan).reshape((-1, 1))
 
 class MockModelWrongVShape:
+    #错的v shape？
     def predict_marginalized_over_instances(self, X):
         m = np.mean(X, axis=1).reshape((-1, 2))
         v = np.var(X, axis=1).reshape((-1, 1)) 
@@ -150,6 +185,12 @@ def test_ei_eta_fail(model, acquisition_function):
     ei.update(model=model)
     configurations = [ConfigurationMock([1.0])]
     with pytest.raises(ValueError):
+        '''
+        1. with pytest.raises(ValueError):
+pytest.raises 是 pytest 提供的一个上下文管理器，用于检查某段代码是否抛出特定类型的异常。
+ValueError 是期望的异常类型。
+这段代码的含义是：在 with 语句的块内，pytest 期望执行的代码会引发 ValueError。如果代码没有抛出 ValueError，测试将会失败。
+        '''
         ei(configurations)
 
 def test_ei_1x1(model, acquisition_function):
@@ -426,7 +467,7 @@ def test_pi_1xD(model, acq_pi):
 def test_pi_NxD(model, acq_pi):
     pi = acq_pi
     pi.update(model=model, eta=1.0)
-    configurations = [
+    configurations = [  
         ConfigurationMock([0.0001, 0.0001, 0.0001]),
         ConfigurationMock([0.1, 0.1, 0.1]),
         ConfigurationMock([1.0, 1.0, 1.0]),
@@ -549,4 +590,72 @@ def test_uncertainty_with_nan(model_with_nan, acq_uncer):
     configurations = [ConfigurationMock([0.5])]
     acq = uncertainty(configurations)
 
-    assert np.all(acq == 0)
+    assert np.all(acq == 0)
+
+
+
+
+'''
+import numpy as np
+
+# 创建一个形状为 (2, 3, 4) 的三维数组
+data = np.array([
+    [[1, 2, 3, 4],   # 实验 1
+     [5, 6, 7, 8],
+     [9, 10, 11, 12]],
+
+    [[13, 14, 15, 16],  # 实验 2
+     [17, 18, 19, 20],
+     [21, 22, 23, 24]]
+])
+
+print("原始数组:")
+print(data)
+mean_axis_0 = np.mean(data, axis=0)
+print("沿 axis=0 计算的均值:")
+print(mean_axis_0)
+# 输出:
+# [[ 7.  8.  9. 10.]
+#  [11. 12. 13. 14.]
+#  [15. 16. 17. 18.]]
+mean_axis_1 = np.mean(data, axis=1)
+
+print("沿 axis=1 计算的均值:")
+print(mean_axis_1)
+# 输出:
+# [[ 5.  6.  7.  8.]
+#  [19. 20. 21. 22.]]
+
+mean_axis_neg1 = np.mean(data, axis=-1)
+print("沿 axis=-1 计算的均值:")
+print(mean_axis_neg1)
+# 输出:
+# [[ 2.5  6.5 10.5]
+#  [14.5 18.5 22.5]]
+总结
+'''
+
+'''
+使用 np.expand_dims
+当我们调用：
+expanded = np.expand_dims(self.cell_lower_bounds, axis=(1, 2))
+axis=(1, 2) 指定在第 1 和第 2 维的位置插入新维度。
+变化步骤
+第一次扩展 (axis=1):
+
+原始数组形状为 (m,)，在第 1 维插入一个新维度后，形状变为 (1, m)：
+python
+复制代码
+[[1, 2, 3]]  # 形状为 (1, 3)
+第二次扩展 (axis=2):
+
+在新数组的第 2 维插入新维度，形状变为 (1, m, 1)：
+python
+复制代码
+[[[1],
+  [2],
+  [3]]]  # 形状为 (1, 3, 1)
+总结
+因此，经过两次扩展后，self.cell_lower_bounds 的形状最终变为 (1, m, 1)，表示有一个批次（第一维为 1），m 个目标（第二维为 m），
+以及每个目标的单个值（第三维为 1）。这种形状使得该数组可以与其他三维数组（如 Y_samples）进行有效的广播和运算。
+'''

test/core/ea/test_adaptive_ea_advisor.py

@@ -7,7 +7,12 @@
 
 def test_adaptive_ea_advisor(configspace_tiny):
     config_space = configspace_tiny
-    advisor = AdaptiveEAAdvisor(config_space, population_size=4, subset_size=2, num_objectives=2, pc=1, pm=1)
+    advisor = AdaptiveEAAdvisor(config_space, population_size=8, subset_size=2, num_objectives=2, pc=1, pm=1)
+    #换一个oldest策略再来一次?
+    '''
+    目前看来这个只支持单目标优化，这里的多目标似乎是假的
+    '''
+    advisor_oldest = AdaptiveEAAdvisor(config_space, strategy='oldest', population_size=8, subset_size=2, num_objectives=2, pc=1, pm=1)
     assert advisor.config_space == config_space
     assert advisor.subset_size == 2
     assert advisor.epsilon == 0.2
@@ -20,13 +25,23 @@ def test_adaptive_ea_advisor(configspace_tiny):
     assert advisor.k4 == 0.3
     assert advisor.last_suggestions == []
     assert advisor.last_observations == []
-
-    perfs = [[2, 1], [1, 2], [2, 2], [1, 1], [1, 2]]
-    for i in range(4):
+    '''
+    多assert
+    '''
+    perfs = [[2, 1], [1, 2], [2, 2], [1, 1], [1, 2], [1, 1], [1, 1], [1, 1], [1, 1], [2, 3], [4, 5], [6, 7], [3, 6]]
+    #修改为10次 ref_point多给点
+    for i in range(13):
         suggestion1 = advisor.get_suggestion()
         assert isinstance(suggestion1, Configuration)
         observation1 = Observation(suggestion1, perfs[i], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
         advisor.update_observation(observation1)
+        '''
+        oldest strategy
+        '''
+        suggestion1 = advisor_oldest.get_suggestion()
+        assert isinstance(suggestion1, Configuration)
+        observation1 = Observation(suggestion1, perfs[i], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
+        advisor_oldest.update_observation(observation1)
 
     config_a = config_space.sample_configuration()
     config_b = config_space.sample_configuration()
@@ -35,4 +50,7 @@ def test_adaptive_ea_advisor(configspace_tiny):
 
     config = config_space.sample_configuration()
     next_config = advisor.mutation(config)
-    assert isinstance(next_config, Configuration)
+    ## 测试None
+    # config = None
+    # next_config = advisor.mutation(config)
+    # assert isinstance(next_config, Configuration)

test/core/ea/test_differential_ea_advisor.py

@@ -6,9 +6,19 @@
 from openbox.utils.constants import MAXINT, SUCCESS
 
 
-def test_cmaes_ea_advisor_initialization(configspace_tiny):
+def test_cmaes_ea_advisor_initialization(configspace_tiny, configspace_cat):
     config_space = configspace_tiny
     advisor = DifferentialEAAdvisor(config_space, num_objectives=2, population_size=4)
+    config_cat = configspace_cat
+    #添加变化的f以及cr
+
+    '''
+    在单目标优化中，允许使用动态的 f 和 cr 来提升算法的表现。
+    在多目标优化中，为了保持多个目标的平衡，限制动态调整 f 和 cr，因此在这种情况下需要固定这些参数。
+    断言的目的是避免算法在不合适的情况下使用动态参数，确保差分进化算法在不同优化问题上的一致性和正确性。
+    '''
+    advisor_vary = DifferentialEAAdvisor(config_space, f=(0,1), cr=(0,1), num_objectives=1, population_size=4)
+    advisor_cat = DifferentialEAAdvisor(config_cat, num_objectives=1, population_size=4)
     assert advisor.config_space == config_space
     assert advisor.num_objectives == 2
     assert advisor.num_constraints == 0
@@ -24,13 +34,23 @@ def test_cmaes_ea_advisor_initialization(configspace_tiny):
     assert advisor.filter_gen_population is None
     assert advisor.keep_unexpected_population is True
     assert advisor.save_cached_configuration is True
-
-    perfs = [[2, 1], [1, 2], [2, 2], [1, 1], [1, 2]]
-    for i in range(5):
+    perfs = [[2, 1], [1, 2], [2, 2], [1, 1], [1, 2], [1, 1], [1, 1], [-1, 1]]
+    for i in range(8):
         suggestion1 = advisor.get_suggestion()
         assert isinstance(suggestion1, Configuration)
         observation1 = Observation(suggestion1, perfs[i], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
+        #这个update内有next_population的更新，会把suggestion1也就是gen的config加入sub，让sel能使用sub
         advisor.update_observation(observation1)
+
+        suggestion2 = advisor_vary.get_suggestion()
+        assert isinstance(suggestion2, Configuration)
+        observation2 = Observation(suggestion2, [perfs[i][0]], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
+        advisor_vary.update_observation(observation2)
+
+    # suggestion3 = advisor_cat.get_suggestion()
+    # assert isinstance(suggestion3, Configuration)
+    # observation3 = Observation(suggestion3, [1], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
+    # advisor_cat.update_observation(observation3)
 
     config_a = config_space.sample_configuration()
     config_b = config_space.sample_configuration()
@@ -40,3 +60,10 @@ def test_cmaes_ea_advisor_initialization(configspace_tiny):
 
     config = advisor.cross_over(config_a, config_b, 0.5)
     assert isinstance(config, Configuration)
+
+
+    config_a = config_cat.sample_configuration()
+    config_b = config_cat.sample_configuration()
+    config_c = config_cat.sample_configuration()
+    config = advisor_cat.mutate(config_a, config_b, config_c, 0.5)
+        #这里的prefs对应ref_point吧？传到Observation中对应objectives属性

test/core/ea/test_nsga2_ea_advisor.py

@@ -6,7 +6,7 @@
 from openbox.utils.constants import MAXINT, SUCCESS
 
 
-def test_nsga2_ea_advisor_initialization(configspace_tiny):
+def test_nsga2_ea_advisor_initialization(configspace_tiny, configspace_huge):
     config_space = configspace_tiny
     advisor = NSGA2EAdvisor(config_space, num_objectives=2, subset_size=2, population_size=4)
     assert advisor.config_space == config_space
@@ -25,4 +25,18 @@ def test_nsga2_ea_advisor_initialization(configspace_tiny):
         suggestion1 = advisor.get_suggestion()
         assert isinstance(suggestion1, Configuration)
         observation1 = Observation(suggestion1, perfs[i], trial_state=SUCCESS, elapsed_time=2.0, extra_info={})
-        advisor.update_observation(observation1)
+        advisor.update_observation(observation1)
+
+    obs = list()
+    for i in range(5):
+        configs = advisor.get_suggestions()
+        for config in configs:
+            obs.append(Observation(config, perfs[i], trial_state=SUCCESS, elapsed_time=2.0, extra_info={}))
+    advisor.update_observations(obs)
+
+
+    #test cross_over
+    advisor_2 = NSGA2EAdvisor(config_space=configspace_huge)
+    for i in range(5):
+        config = advisor_2.cross_over(configspace_huge.sample_configuration(), configspace_huge.sample_configuration())
+        assert config.configuration_space == configspace_huge