Hotfix: Minimization with bounds properly minimize targets (#462)

This PR hot-fixes #460 by multiplying the output of the
`objective.transform` with -1 if the target is bounded and to be
minimized.

This PR also introduces a test and a small example for verifying the
desired behavior. The solution implemented here is only temporary and
will be replaced with a proper mechanism soon.

CHANGELOG.md

@@ -41,6 +41,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   `NumericalTarget` now raises an error
 - Crash when using `ContinuousCardinalityConstraint` caused by an unintended interplay
   between constraints and dropped parameters yielding empty parameter sets
+- Minimizing a single `NumericalTarget` with specified bounds/transformation via
+  `SingleTargetObjective` no longer erroneously maximizes it
 
 ### Removed
 - `botorch_function_wrapper` utility for creating lookup callables

baybe/objectives/single.py

@@ -10,6 +10,8 @@
 
 from baybe.objectives.base import Objective
 from baybe.targets.base import Target
+from baybe.targets.enum import TargetMode
+from baybe.targets.numerical import NumericalTarget
 from baybe.utils.dataframe import get_transform_objects, pretty_print_df
 from baybe.utils.plotting import to_string
 
@@ -86,7 +88,16 @@ def transform(
 
         target_data = df[self._target.name].copy()
 
-        return self._target.transform(target_data).to_frame()
+        out = self._target.transform(target_data).to_frame()
+
+        # TODO: Remove hotfix (https://github.com/emdgroup/baybe/issues/460)
+        if (
+            isinstance(t := self._target, NumericalTarget)
+            and t.mode is TargetMode.MIN
+            and t.bounds.is_bounded
+        ):
+            out = -out
+        return out
 
 
 # Collect leftover original slotted classes processed by `attrs.define`

tests/test_objective.py

@@ -1,12 +1,15 @@
 """Tests for the objective module."""
 
 import numpy as np
+import pandas as pd
 import pytest
 from cattrs import IterableValidationError
 
 from baybe.objectives.desirability import DesirabilityObjective, scalarize
 from baybe.objectives.enum import Scalarizer
 from baybe.objectives.single import SingleTargetObjective
+from baybe.parameters.numerical import NumericalContinuousParameter
+from baybe.recommenders import BotorchRecommender
 from baybe.targets import NumericalTarget
 
 
@@ -93,3 +96,19 @@ def test_desirability_scalarization(values, scalarizer, weights, expected):
     """The desirability scalarization yields the expected result."""
     actual = scalarize(values, scalarizer, weights)
     assert np.array_equal(actual, expected), (expected, actual)
+
+
+@pytest.mark.parametrize(
+    ("mode", "bounds", "opt"),
+    [("MIN", None, 0), ("MAX", None, 1), ("MIN", (0, 1), 0), ("MAX", (0, 1), 1)],
+)
+def test_single_objective(mode, bounds, opt):
+    """Recommendations yield expected results with and without bounded objective."""
+    searchspace = NumericalContinuousParameter("p", [0, 1]).to_searchspace()
+    objective = NumericalTarget("t", mode=mode, bounds=bounds).to_objective()
+    recommender = BotorchRecommender()
+    measurements = pd.DataFrame(
+        {"p": np.linspace(0, 1, 100), "t": np.linspace(0, 1, 100)}
+    )
+    rec = recommender.recommend(1, searchspace, objective, measurements)
+    assert np.isclose(rec["p"].item(), opt)