Use BetaPrior in MultiTask Botorch Preset

CHANGELOG.md

@@ -4,6 +4,14 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+### Changed
+- `BOTORCH` GP preset now includes `BetaPrior(2.5, 1.5)` for the task covariance
+  kernel in multi-task scenarios, matching BoTorch's `MultiTaskGP` defaults introduced
+  in version `0.18.0`
+- The `BOTORCH` GP preset now requires BoTorch `>= 0.18.0` and raises an
+  `IncompatibilityError` if an older version is installed
+
 ## [0.15.0] - 2026-06-11
 ### Breaking Changes
 - `GaussianProcessSurrogate` no longer automatically adds a task kernel in multi-task

baybe/surrogates/gaussian_process/presets/botorch.py

@@ -1,16 +1,26 @@
-"""BoTorch preset for Gaussian process surrogates.
+"""BoTorch preset for Gaussian process surrogates."""
 
-Currently mimics the Hvarfner preset :cite:p:`Hvarfner2024`.
-"""
+from __future__ import annotations
 
-from baybe.surrogates.gaussian_process.presets.hvarfner import (
-    FIT_CRITERION_FACTORY,
-    KERNEL_FACTORY,
-    LIKELIHOOD_FACTORY,
-    MEAN_FACTORY,
+import gc
+from itertools import chain
+from typing import TYPE_CHECKING, ClassVar
+
+import pandas as pd
+from attrs import define
+from typing_extensions import override
+
+from baybe.kernels.base import Kernel
+from baybe.objectives.base import Objective
+from baybe.parameters.enum import _ParameterKind
+from baybe.searchspace.core import SearchSpace
+from baybe.surrogates.gaussian_process.components.fit_criterion import (
+    FitCriterion,
+    PlainFitCriterionFactory,
 )
-from baybe.surrogates.gaussian_process.presets.hvarfner import (
-    HvarfnerKernelFactory as BotorchKernelFactory,
+from baybe.surrogates.gaussian_process.components.kernel import (
+    ICMKernelFactory,
+    _PureKernelFactory,
 )
 from baybe.surrogates.gaussian_process.presets.hvarfner import (
     HvarfnerLikelihoodFactory as BotorchLikelihoodFactory,
@@ -19,6 +29,101 @@
     HvarfnerMeanFactory as BotorchMeanFactory,
 )
 
+if TYPE_CHECKING:
+    from gpytorch.kernels import Kernel as GPyTorchKernel
+
+# The minimum BoTorch version required for the preset
+_MIN_BOTORCH_VERSION = "0.18.0"
+
+
+@define
+class BotorchKernelFactory(_PureKernelFactory):
+    """A factory providing kernels matching BoTorch's :class:`~botorch.models.MultiTaskGP` defaults."""  # noqa: E501
+
+    _uses_parameter_names: ClassVar[bool] = True
+    # See base class.
+
+    _supported_parameter_kinds: ClassVar[_ParameterKind] = (
+        _ParameterKind.REGULAR | _ParameterKind.TASK
+    )
+    # See base class.
+
+    @override
+    def _make(
+        self, searchspace: SearchSpace, objective: Objective, measurements: pd.DataFrame
+    ) -> Kernel | GPyTorchKernel:
+        self._validate_botorch_version()
+
+        from botorch.models.kernels.positive_index import PositiveIndexKernel
+        from botorch.models.utils.gpytorch_modules import (
+            get_covar_module_with_dim_scaled_prior,
+        )
+        from botorch.models.utils.priors import BetaPrior
+
+        parameter_names = self.get_parameter_names(searchspace)
+
+        # For regular parameters, resolve parameter names to active dimension indices
+        active_dims = list(
+            chain.from_iterable(
+                searchspace.get_comp_rep_parameter_indices(name)
+                for name in parameter_names
+                if searchspace.get_parameters_by_name([name])[0]._kind
+                is _ParameterKind.REGULAR
+            )
+        )
+        ard_num_dims = len(active_dims)
+
+        # Create the base kernel for the regular parameters
+        base_kernel = get_covar_module_with_dim_scaled_prior(
+            ard_num_dims=ard_num_dims, active_dims=active_dims
+        )
+
+        # Single-task case
+        if (task_idx := searchspace.task_idx) is None:
+            return base_kernel
+
+        task_prior = BetaPrior(concentration1=2.5, concentration0=1.5)
+        index_kernel = PositiveIndexKernel(
+            num_tasks=searchspace.n_tasks,
+            rank=searchspace.n_tasks,
+            task_prior=task_prior,
+            active_dims=[task_idx],
+        )
+        return ICMKernelFactory(base_kernel, index_kernel)(
+            searchspace, objective, measurements
+        )
+
+    def _validate_botorch_version(self) -> None:
+        """Verify that the installed BoTorch version meets the minimum requirement.
+
+        Raises:
+            IncompatibilityError: If the installed BoTorch version is too old.
+        """
+        from importlib.metadata import version
+
+        from packaging.version import Version
+
+        from baybe.exceptions import IncompatibilityError
+
+        installed = version("botorch")
+        if Version(installed) < Version(_MIN_BOTORCH_VERSION):
+            raise IncompatibilityError(
+                f"The '{self.__class__.__name__}' requires botorch>="
+                f"{_MIN_BOTORCH_VERSION}, but version {installed} is installed. "
+                f"Please upgrade: pip install 'botorch>="
+                f"{_MIN_BOTORCH_VERSION}'."
+            )
+
+
+# Collect leftover original slotted classes processed by `attrs.define`
+gc.collect()
+
+# Aliases for generic preset imports
+KERNEL_FACTORY = BotorchKernelFactory()
+MEAN_FACTORY = BotorchMeanFactory()
+LIKELIHOOD_FACTORY = BotorchLikelihoodFactory()
+FIT_CRITERION_FACTORY = PlainFitCriterionFactory(FitCriterion.MARGINAL_LOG_LIKELIHOOD)
+
 __all__ = [
     "BotorchKernelFactory",
     "BotorchLikelihoodFactory",

tests/test_gp.py

@@ -1,5 +1,7 @@
 """Tests for the Gaussian Process surrogate."""
 
+import sys
+
 import pandas as pd
 import pytest
 import torch
@@ -195,13 +197,16 @@ def test_invalid_components():
         GaussianProcessSurrogate(fit_criterion_or_factory=MaternKernel())
 
 
-# NOTE: BOTORCH and HVARFNER presets coincide at the moment but BOTORCH settings can
-#   change in the future. If that happens, the test below will start to fail and the
-#   HVARFNER parametrization needs to be dropped.
-@pytest.mark.parametrize("preset", ["BOTORCH", "HVARFNER"], ids=["botorch", "hvarfner"])
+# NOTE: The BOTORCH preset tracks BoTorch's GP defaults while the HVARFNER preset
+#   implements BoTorch's static Hvarfner et al. (2024) parametrization. Therefore, the
+#   presets diverge as BoTorch evolves (e.g., BetaPrior added in 0.18.0).
+@pytest.mark.skipif(
+    sys.version_info < (3, 11),
+    reason="BoTorch >=0.18.0 requires Python >=3.11.",
+)
 @pytest.mark.parametrize("multitask", [False, True], ids=["single-task", "multi-task"])
-def test_botorch_preset(multitask: bool, preset: str):
-    """The BoTorch/Hvarfner presets exactly mimic BoTorch's behavior."""
+def test_botorch_preset(multitask: bool):
+    """The BoTorch preset exactly mimics BoTorch's MultiTaskGP/SingleTaskGP behavior."""
     if multitask:
         sp = searchspace_mt
         data = measurements_mt
@@ -210,7 +215,7 @@ def test_botorch_preset(multitask: bool, preset: str):
         data = measurements
 
     active_settings.random_seed = 1337
-    gp = GaussianProcessSurrogate.from_preset(preset)
+    gp = GaussianProcessSurrogate.from_preset("BOTORCH")
     gp.fit(sp, objective, data)
     posterior1 = gp.posterior_stats(data)
 

tox.ini

@@ -6,6 +6,8 @@ isolated_build = True
 [testenv:fulltest,fulltest-py{310,311,312,313,314}]
 description = Run PyTest with all extra functionality
 extras = extras,examples,lint,test
+deps =
+    py311,py312,py313,py314: botorch>=0.18.0,<1
 passenv =
     CI
     BAYBE_USE_SINGLE_PRECISION_NUMPY 
@@ -21,6 +23,8 @@ commands =
 [testenv:gputest,gputest-py{310,311,312,313,314}]
 description = Runs GPU tests
 extras = test
+deps =
+    py311,py312,py313,py314: botorch>=0.18.0,<1
 passenv =
     CI
     BAYBE_USE_SINGLE_PRECISION_NUMPY
@@ -35,6 +39,8 @@ commands =
 [testenv:coretest,coretest-py{310,311,312,313,314}]
 description = Run PyTest with core functionality
 extras = test
+deps =
+    py311,py312,py313,py314: botorch>=0.18.0,<1
 passenv =
     CI
     BAYBE_USE_SINGLE_PRECISION_NUMPY