Fix initial point sampling

pyPDP/__init__.py

@@ -3,6 +3,6 @@
 author = "Yannik Mahlau and Dominik Woiwode"
 author_email = "[email protected]"
 description = "A python implementation of 'Explaining Hyperparameter Optimization via Partial Dependence Plots' by Moosbauer et al."
-version = "0.1.7"
+version = "0.1.8"
 license = "MIT"
 url = "https://github.com/dwoiwode/py-pdp-partitioner"

pyPDP/sampler/bayesian_optimization.py

@@ -55,11 +55,12 @@ def _sample_initial_points(self, max_sampled_points=None):
         else:
             sampled_points = min(self.initial_points, max_sampled_points)
 
-        self.config_list = self.config_space.sample_configuration(sampled_points)
-        if self.initial_points == 1:  # for a single value, the sampling does not return a list
-            self.config_list = [self.config_list]
+        configs = self.config_space.sample_configuration(sampled_points)
+        if sampled_points == 1:  # for a single value, the sampling does not return a list
+            configs = [configs]
 
-        self.y_list = [self.obj_func(**config) for config in self.config_list]
+        self.y_list += [self.obj_func(**config) for config in configs]
+        self.config_list += configs
         self.fit_surrogate(force=True)
 
     def fit_surrogate(self, force: bool = False):

tests/sampler/test_optimizers.py

@@ -213,3 +213,35 @@ def test_cache_different_acq_class(self):
 
         self.assertGreater(t_dif_1, 0.1)  # If values are loaded from cache, it is pretty fast
         self.assertGreater(t_dif_2, 0.1)  # If values are loaded from cache, it is pretty fast
+
+    def test_sample_1_init(self):
+        f = Square.for_n_dimensions(5, seed=0)
+        cs = f.config_space
+        # First try with seed
+        bo = BayesianOptimizationSampler(
+            obj_func=f,
+            config_space=cs,
+            initial_points=10,
+        )
+        bo.clear_cache()
+
+        bo.sample(1)
+        bo.sample(1)
+        bo.sample(1)
+
+        self.assertEqual(3, len(bo))
+
+    def test_sample_1_after_init(self):
+        f = Square.for_n_dimensions(5, seed=0)
+        cs = f.config_space
+        # First try with seed
+        bo = BayesianOptimizationSampler(
+            obj_func=f,
+            config_space=cs,
+            initial_points=5,
+        )
+
+        bo.sample(5)
+        bo.sample(1)
+        bo.sample(1)
+        self.assertEqual(7, len(bo))