Hotfix/random proposals until valid evals (#118)

* fix sampling of random specs until enough valid evals reported
* simplify num_initial_random check

---------

Signed-off-by: Grossberger Lukas (CR/AIR2.2) <Lukas.Grossberger@de.bosch.com>

Author: LGro

blackboxopt/optimizers/botorch_base.py

@@ -173,6 +173,9 @@ def __init__(
             num_initial_random_samples: Size of the initial space-filling design that
                 is used before starting BO. The points are sampled randomly in the
                 search space. If no random sampling is required, set it to 0.
+                When random sampling is enabled, but evaluations with missing objective
+                values are reported, more specifications are sampled until
+                `num_initial_random_samples` many valid evaluations were reported.
             max_pending_evaluations: Maximum number of parallel evaluations. For
                 sequential BO use the default value of 1. If no limit is required,
                 set it to None.
@@ -275,17 +278,18 @@ def generate_evaluation_specification(self) -> EvaluationSpecification:
         ):
             raise OptimizerNotReady
 
+        # Generate random samples until there are enough samples where at least one of
+        # the objective values is available
         if self.num_initial_random > 0 and (
-            self.X.size(-2) < self.num_initial_random
-            or torch.nonzero(~torch.any(self.losses.isnan(), dim=1)).numel() == 0
+            sum(~torch.any(self.losses.isnan(), dim=1)) < self.num_initial_random
         ):
-            # We keep generating random samples until there are enough samples, and
-            # at least one of them has a valid objective
             eval_spec = EvaluationSpecification(
                 configuration=self.search_space.sample(),
+                optimizer_info={"model_based_pick": False},
             )
         else:
             eval_spec = self._generate_evaluation_specification()
+            eval_spec.optimizer_info["model_based_pick"] = True
 
         eval_id = self.X.size(-2) + len(self.pending_specifications)
         eval_spec.optimizer_info["evaluation_id"] = eval_id

tests/optimizers/botorch_base_test.py

@@ -170,6 +170,56 @@ def test_find_optimum_in_1d_discrete_space(seed):
     assert opt.objective.name in best.objectives
 
 
+def test_propose_random_until_enough_evaluations_without_missing_objective_values(seed):
+    space = ps.ParameterSpace()
+    space.add(ps.IntegerParameter("integ", (0, 2)))
+    batch_shape = torch.Size()
+
+    opt = SingleObjectiveBOTorchOptimizer(
+        search_space=space,
+        objective=Objective("loss", greater_is_better=False),
+        model=SingleTaskGP(
+            torch.empty((*batch_shape, 0, len(space)), dtype=torch.float64),
+            torch.empty((*batch_shape, 0, 1), dtype=torch.float64),
+        ),
+        acquisition_function_factory=partial(
+            UpperConfidenceBound, beta=1.0, maximize=False
+        ),
+        num_initial_random_samples=2,
+        max_pending_evaluations=1,
+        seed=seed,
+    )
+
+    es = opt.generate_evaluation_specification()
+    assert not es.optimizer_info[
+        "model_based_pick"
+    ], "No evaluation reported, 0 < 2 initial random samples"
+    opt.report(
+        es.create_evaluation(objectives={"loss": es.configuration["integ"] ** 2}),
+    )
+
+    es = opt.generate_evaluation_specification()
+    assert not es.optimizer_info[
+        "model_based_pick"
+    ], "One evaluation reported, 1 < 2 initial random samples"
+    opt.report(
+        es.create_evaluation(objectives={"loss": None}),
+    )
+
+    es = opt.generate_evaluation_specification()
+    assert not es.optimizer_info[
+        "model_based_pick"
+    ], "One valid evaluation reported, 1 < 2 initial random samples"
+    opt.report(
+        es.create_evaluation(objectives={"loss": es.configuration["integ"] ** 2}),
+    )
+
+    es = opt.generate_evaluation_specification()
+    assert es.optimizer_info[
+        "model_based_pick"
+    ], "Two valid evaluations reported, 2 >= 2 initial random samples"
+
+
 def test_get_numerical_points_from_discrete_space():
     p0l, p0h = -5, 10
     p1 = ("small", "medium", "large")