Efficient NChooseKs in BO

bofire/benchmarks/single.py

@@ -1,13 +1,16 @@
 import math
+from typing import Optional
 
 import numpy as np
 import pandas as pd
 import torch
+from botorch.test_functions import Hartmann as botorch_hartmann
 from botorch.test_functions.synthetic import Branin as torchBranin
 from pydantic.types import PositiveInt
 
 from bofire.benchmarks.benchmark import Benchmark
-from bofire.data_models.domain.api import Domain, Inputs, Outputs
+from bofire.data_models.constraints.api import NChooseKConstraint
+from bofire.data_models.domain.api import Constraints, Domain, Inputs, Outputs
 from bofire.data_models.features.api import (
     CategoricalDescriptorInput,
     CategoricalInput,
@@ -156,6 +159,60 @@ def get_optima(self) -> pd.DataFrame:
         )
 
 
+class Hartmann(Benchmark):
+    def __init__(self, dim: int = 6, allowed_k: Optional[int] = None) -> None:
+        super().__init__()
+        self._domain = Domain(
+            input_features=Inputs(
+                features=[
+                    ContinuousInput(key=f"x_{i}", bounds=(0, 1)) for i in range(dim)
+                ]
+            ),
+            output_features=Outputs(
+                features=[ContinuousOutput(key="y", objective=MinimizeObjective())]
+            ),
+            constraints=Constraints(
+                constraints=[
+                    NChooseKConstraint(
+                        features=[f"x_{i}" for i in range(dim)],
+                        min_count=0,
+                        max_count=allowed_k,
+                        none_also_valid=True,
+                    )
+                ]
+            )
+            if allowed_k
+            else Constraints(),
+        )
+        self._hartmann = botorch_hartmann(dim=dim)
+
+    def get_optima(self) -> pd.DataFrame:
+        if self.dim != 6:
+            raise ValueError("Only available for dim==6.")
+        if len(self.domain.constraints) > 0:
+            raise ValueError("Not defined for NChooseK use case.")
+        return pd.DataFrame(
+            columns=[f"x_{i}" for i in range(self.dim)] + ["y"],
+            data=[[0.20169, 0.150011, 0.476874, 0.275332, 0.311652, 0.6573, -3.32237]],
+        )
+
+    @property
+    def dim(self) -> int:
+        return len(self.domain.inputs)
+
+    def _f(self, candidates: pd.DataFrame) -> pd.DataFrame:
+        return pd.DataFrame(
+            {
+                "y": self._hartmann(
+                    torch.from_numpy(
+                        candidates[[f"x_{i}" for i in range(self.dim)]].values
+                    )
+                ),
+                "valid_y": [1 for _ in range(len(candidates))],
+            }
+        )
+
+
 class Branin(Benchmark):
     def __init__(self) -> None:
         self._domain = Domain(

bofire/data_models/domain/domain.py

@@ -597,7 +597,7 @@ def describe_experiments(self, experiments: pd.DataFrame) -> pd.DataFrame:
         )
 
     def validate_candidates(
-        self, candidates: pd.DataFrame, only_inputs: bool = False, tol: float = 1e-6
+        self, candidates: pd.DataFrame, only_inputs: bool = False, tol: float = 1e-5
     ) -> pd.DataFrame:
         """Method to check the validty of porposed candidates
 
@@ -622,26 +622,41 @@ def validate_candidates(
         self.input_features.validate_inputs(candidates)
         # check if all constraints are fulfilled
         if not self.cnstrs.is_fulfilled(candidates, tol=tol).all():
-            raise ValueError("Constraints not fulfilled.")
+            raise ValueError(f"Constraints not fulfilled: {candidates}")
         # for each continuous output feature with an attached objective object
         if not only_inputs:
             assert isinstance(self.output_features, Outputs)
-            for key in self.output_features.get_keys_by_objective(Objective):
-                # check that pred, sd, and des cols are specified and numerical
-                for col in [f"{key}_pred", f"{key}_sd", f"{key}_des"]:
-                    if col not in candidates:
-                        raise ValueError(f"missing column {col}")
-                    if (not is_numeric(candidates[col])) and (
-                        not candidates[col].isnull().to_numpy().all()
-                    ):
-                        raise ValueError(
-                            f"not all values of output feature `{key}` are numerical"
+
+            cols = list(
+                itertools.chain.from_iterable(
+                    [
+                        [f"{key}_pred", f"{key}_sd", f"{key}_des"]
+                        for key in self.output_features.get_keys_by_objective(Objective)
+                    ]
+                    + [
+                        [f"{key}_pred", f"{key}_sd"]
+                        for key in self.output_features.get_keys_by_objective(
+                            excludes=Objective, includes=None  # type: ignore
                         )
+                    ]
+                )
+            )
+
+            # check that pred, sd, and des cols are specified and numerical
+            for col in cols:
+                if col not in candidates:
+                    raise ValueError(f"missing column {col}")
+                if (not is_numeric(candidates[col])) and (
+                    not candidates[col].isnull().to_numpy().all()
+                ):
+                    raise ValueError(f"not all values of column `{col}` are numerical")
+
             # validate no additional cols exist
             if_count = len(self.get_features(Input))
-            of_count = len(self.output_features.get_keys_by_objective(Objective))
+            of_count = len(self.outputs.get_by_objective(includes=Objective))
+            of_count_w = len(self.outputs.get_by_objective(excludes=Objective, includes=None))  # type: ignore
             # input features, prediction, standard deviation and reward for each output feature, 3 additional usefull infos: reward, aquisition function, strategy
-            if len(candidates.columns) != if_count + 3 * of_count:
+            if len(candidates.columns) != if_count + 3 * of_count + 2 * of_count_w:
                 raise ValueError("additional columns found")
         return candidates
 

bofire/data_models/strategies/predictives/botorch.py

@@ -1,7 +1,12 @@
-from typing import Optional
+from typing import Optional, Type
 
 from pydantic import PositiveInt, root_validator, validator
 
+from bofire.data_models.constraints.api import (
+    Constraint,
+    NonlinearEqualityConstraint,
+    NonlinearInequalityConstraint,
+)
 from bofire.data_models.domain.api import Domain, Outputs
 from bofire.data_models.enum import CategoricalEncodingEnum, CategoricalMethodEnum
 from bofire.data_models.features.api import CategoricalDescriptorInput, CategoricalInput
@@ -26,6 +31,20 @@ class BotorchStrategy(PredictiveStrategy):
     discrete_method: CategoricalMethodEnum = CategoricalMethodEnum.EXHAUSTIVE
     surrogate_specs: Optional[BotorchSurrogates] = None
 
+    @classmethod
+    def is_constraint_implemented(cls, my_type: Type[Constraint]) -> bool:
+        """Method to check if a specific constraint type is implemented for the strategy
+
+        Args:
+            my_type (Type[Constraint]): Constraint class
+
+        Returns:
+            bool: True if the constraint type is valid for the strategy chosen, False otherwise
+        """
+        if my_type in [NonlinearInequalityConstraint, NonlinearEqualityConstraint]:
+            return False
+        return True
+
     @validator("num_sobol_samples")
     def validate_num_sobol_samples(cls, v):
         if is_power_of_two(v) is False:

bofire/data_models/strategies/predictives/qehvi.py

@@ -2,7 +2,6 @@
 
 from pydantic import validator
 
-from bofire.data_models.constraints.api import Constraint, NChooseKConstraint
 from bofire.data_models.features.api import Feature
 from bofire.data_models.objectives.api import (
     MaximizeObjective,
@@ -33,20 +32,6 @@ def validate_ref_point(cls, v, values):
             )
         return v
 
-    @classmethod
-    def is_constraint_implemented(cls, my_type: Type[Constraint]) -> bool:
-        """Method to check if a specific constraint type is implemented for the strategy
-
-        Args:
-            my_type (Type[Constraint]): Constraint class
-
-        Returns:
-            bool: True if the constraint type is valid for the strategy chosen, False otherwise
-        """
-        if my_type == NChooseKConstraint:
-            return False
-        return True
-
     @classmethod
     def is_feature_implemented(cls, my_type: Type[Feature]) -> bool:
         """Method to check if a specific feature type is implemented for the strategy

bofire/data_models/strategies/predictives/qnehvi.py

@@ -3,6 +3,7 @@
 from pydantic import confloat
 
 from bofire.data_models.objectives.api import (
+    CloseToTargetObjective,
     MaximizeObjective,
     MaximizeSigmoidObjective,
     MinimizeObjective,
@@ -33,4 +34,5 @@ def is_objective_implemented(cls, my_type: Type[Objective]) -> bool:
             MinimizeSigmoidObjective,
             MaximizeSigmoidObjective,
             TargetObjective,
+            CloseToTargetObjective,
         ]

bofire/data_models/strategies/predictives/qparego.py

@@ -1,8 +1,8 @@
 from typing import Literal, Type
 
-from bofire.data_models.constraints.api import Constraint, NChooseKConstraint
 from bofire.data_models.features.api import Feature
 from bofire.data_models.objectives.api import (
+    CloseToTargetObjective,
     MaximizeObjective,
     MaximizeSigmoidObjective,
     MinimizeObjective,
@@ -18,12 +18,6 @@
 class QparegoStrategy(MultiobjectiveStrategy):
     type: Literal["QparegoStrategy"] = "QparegoStrategy"
 
-    @classmethod
-    def is_constraint_implemented(cls, my_type: Type[Constraint]) -> bool:
-        if my_type == NChooseKConstraint:
-            return False
-        return True
-
     @classmethod
     def is_feature_implemented(cls, my_type: Type[Feature]) -> bool:
         return True
@@ -36,6 +30,7 @@ def is_objective_implemented(cls, my_type: Type[Objective]) -> bool:
             TargetObjective,
             MinimizeSigmoidObjective,
             MaximizeSigmoidObjective,
+            CloseToTargetObjective,
         ]:
             return False
         return True

bofire/data_models/strategies/predictives/sobo.py

@@ -3,11 +3,6 @@
 from pydantic import validator
 
 from bofire.data_models.acquisition_functions.api import AnyAcquisitionFunction
-from bofire.data_models.constraints.api import (
-    Constraint,
-    NonlinearEqualityConstraint,
-    NonlinearInequalityConstraint,
-)
 from bofire.data_models.features.api import Feature
 from bofire.data_models.objectives.api import BotorchConstrainedObjective, Objective
 from bofire.data_models.strategies.predictives.botorch import BotorchStrategy
@@ -16,20 +11,6 @@
 class SoboBaseStrategy(BotorchStrategy):
     acquisition_function: AnyAcquisitionFunction
 
-    @classmethod
-    def is_constraint_implemented(cls, my_type: Type[Constraint]) -> bool:
-        """Method to check if a specific constraint type is implemented for the strategy
-
-        Args:
-            my_type (Type[Constraint]): Constraint class
-
-        Returns:
-            bool: True if the constraint type is valid for the strategy chosen, False otherwise
-        """
-        if my_type in [NonlinearInequalityConstraint, NonlinearEqualityConstraint]:
-            return False
-        return True
-
     @classmethod
     def is_feature_implemented(cls, my_type: Type[Feature]) -> bool:
         """Method to check if a specific feature type is implemented for the strategy