Adaptive hyper-prior: CHENKernelFactory

CHANGELOG.md

@@ -16,7 +16,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Support for GPyTorch objects (kernels, means, likelihood) as Gaussian process
   components, enabling full low-level customization
 - Factories for all Gaussian process components
-- `EDBO` and `EDBO_SMOOTHED` presets for `GaussianProcessSurrogate`
+- `CHEN`, `EDBO` and `EDBO_SMOOTHED` presets for `GaussianProcessSurrogate`
 - `TypeSelector` and `NameSelector` classes for parameter selection in kernel factories
 - `parameter_names` attribute to basic kernels for controlling the considered parameters
 - `ParameterKind` flag enum for classifying parameters by their role and automatic

CONTRIBUTORS.md

@@ -39,4 +39,10 @@
 - Kathrin Skubch (Merck KGaA, Darmstadt, Germany):\
   Transfer learning regression benchmarks infrastructure
 - Myra Zmarsly (Merck Life Science KGaA, Darmstadt, Germany):\
-  Identification of non-dominated parameter configurations
+  Identification of non-dominated parameter configurations
+- Thijs Stuyver (PSL University, Paris, France):\
+  Adaptive hyper-prior tailored for reaction yield optimization tasks
+- Maximilian Fleck (PSL University, Paris, France):\
+  Adaptive hyper-prior tailored for reaction yield optimization tasks
+- Guanming Chen (PSL University, Paris, France):\
+  Adaptive hyper-prior tailored for reaction yield optimization tasks

baybe/surrogates/gaussian_process/presets/__init__.py

@@ -7,6 +7,9 @@
     BayBEMeanFactory,
 )
 
+# Chen preset
+from baybe.surrogates.gaussian_process.presets.chen import CHENKernelFactory
+
 # Core
 from baybe.surrogates.gaussian_process.presets.core import GaussianProcessPreset
 
@@ -31,6 +34,8 @@
     "BayBEKernelFactory",
     "BayBELikelihoodFactory",
     "BayBEMeanFactory",
+    # Chen preset
+    "CHENKernelFactory",
     # EDBO preset
     "EDBOKernelFactory",
     "EDBOLikelihoodFactory",

baybe/surrogates/gaussian_process/presets/chen.py

@@ -0,0 +1,77 @@
+"""Preset for adaptive kernel hyperpriors proposed by :cite:p:`Chen2026`."""
+
+from __future__ import annotations
+
+import gc
+import math
+from typing import TYPE_CHECKING, ClassVar
+
+from attrs import define, field
+from typing_extensions import override
+
+from baybe.kernels.basic import MaternKernel
+from baybe.kernels.composite import ScaleKernel
+from baybe.parameters.categorical import TaskParameter
+from baybe.parameters.selectors import (
+    ParameterSelectorProtocol,
+    TypeSelector,
+    to_parameter_selector,
+)
+from baybe.priors.basic import GammaPrior
+from baybe.surrogates.gaussian_process.components.kernel import (
+    _PureKernelFactory,
+)
+from baybe.surrogates.gaussian_process.presets.baybe import (
+    BayBELikelihoodFactory,
+    BayBEMeanFactory,
+)
+
+if TYPE_CHECKING:
+    from torch import Tensor
+
+    from baybe.kernels.base import Kernel
+    from baybe.searchspace.core import SearchSpace
+
+
+@define
+class CHENKernelFactory(_PureKernelFactory):
+    """A factory providing adaptive hyperprior kernels as proposed by :cite:p:`Chen2026`."""  # noqa: E501
+
+    _uses_parameter_names: ClassVar[bool] = True
+    # See base class.
+
+    parameter_selector: ParameterSelectorProtocol | None = field(
+        factory=lambda: TypeSelector([TaskParameter], exclude=True),
+        converter=to_parameter_selector,
+    )
+    # TODO: Reuse base attribute (https://github.com/python-attrs/attrs/pull/1429)
+
+    @override
+    def _make(
+        self, searchspace: SearchSpace, train_x: Tensor, train_y: Tensor
+    ) -> Kernel:
+        lengthscale = 0.4 * math.sqrt(train_x.shape[-1]) + 4.0
+        lengthscale_prior = GammaPrior(2.0 * lengthscale, 2.0)
+        lengthscale_initial_value = lengthscale
+        outputscale_prior = GammaPrior(1.0 * lengthscale, 1.0)
+        outputscale_initial_value = lengthscale
+
+        return ScaleKernel(
+            MaternKernel(
+                nu=2.5,
+                lengthscale_prior=lengthscale_prior,
+                lengthscale_initial_value=lengthscale_initial_value,
+                parameter_names=self.get_parameter_names(searchspace),
+            ),
+            outputscale_prior=outputscale_prior,
+            outputscale_initial_value=outputscale_initial_value,
+        )
+
+
+# Collect leftover original slotted classes processed by `attrs.define`
+gc.collect()
+
+# Aliases for generic preset imports
+PresetKernelFactory = CHENKernelFactory
+PresetMeanFactory = BayBEMeanFactory
+PresetLikelihoodFactory = BayBELikelihoodFactory

baybe/surrogates/gaussian_process/presets/core.py

@@ -11,8 +11,11 @@ class GaussianProcessPreset(Enum):
     BAYBE = "BAYBE"
     """The default BayBE settings of the Gaussian process surrogate class."""
 
+    CHEN = "CHEN"
+    """The adaptive kernel hyperprior settings proposed by :cite:p:`Chen2026`."""
+
     EDBO = "EDBO"
-    """The EDBO settings."""
+    """The EDBO settings proposed by :cite:p:`Shields2021`."""
 
     EDBO_SMOOTHED = "EDBO_SMOOTHED"
-    """A smoothed version of the EDBO settings."""
+    """A smoothed version of the EDBO settings (adapted from :cite:p:`Shields2021`)."""

baybe/surrogates/gaussian_process/presets/edbo.py

@@ -1,4 +1,4 @@
-"""EDBO preset for Gaussian process surrogates."""
+"""EDBO preset :cite:p:`Shields2021`."""
 
 from __future__ import annotations
 
@@ -58,11 +58,9 @@ def _contains_encoding(
 
 @define
 class EDBOKernelFactory(_PureKernelFactory):
-    """A factory providing EDBO kernels.
+    """A factory providing EDBO kernels, as proposed by :cite:p:`Shields2021`.
 
-    References:
-        * https://github.com/b-shields/edbo/blob/master/edbo/bro.py#L664
-        * https://doi.org/10.1038/s41586-021-03213-y
+    GitHub repository: https://github.com/b-shields/edbo
     """
 
     _uses_parameter_names: ClassVar[bool] = True
@@ -130,11 +128,9 @@ def _make(
 
 @define
 class EDBOLikelihoodFactory(LikelihoodFactoryProtocol):
-    """A factory providing EDBO likelihoods.
+    """A factory providing EDBO likelihoods, as proposed by :cite:p:`Shields2021`.
 
-    References:
-        * https://github.com/b-shields/edbo/blob/master/edbo/bro.py#L664
-        * https://doi.org/10.1038/s41586-021-03213-y
+    GitHub repository: https://github.com/b-shields/edbo
     """
 
     @override

baybe/surrogates/gaussian_process/presets/edbo_smoothed.py

@@ -1,4 +1,4 @@
-"""Smoothed EDBO preset for Gaussian process surrogates."""
+"""Smoothed EDBO preset (adapted from :cite:p:`Shields2021`)."""
 
 from __future__ import annotations
 
@@ -40,12 +40,12 @@
 
 @define
 class SmoothedEDBOKernelFactory(_PureKernelFactory):
-    """A factory providing smoothed versions of EDBO kernels.
+    """A factory providing smoothed versions of EDBO kernels (adapted from :cite:p:`Shields2021`).
 
     Takes the low and high dimensional limits of
     :class:`baybe.surrogates.gaussian_process.presets.edbo.EDBOKernelFactory`
     and interpolates the prior moments linearly in between.
-    """
+    """  # noqa: E501
 
     _uses_parameter_names: ClassVar[bool] = True
     # See base class.
@@ -94,12 +94,12 @@ def _make(
 
 @define
 class SmoothedEDBOLikelihoodFactory(LikelihoodFactoryProtocol):
-    """A factory providing smoothed versions of EDBO likelihoods.
+    """A factory providing smoothed versions of EDBO likelihoods (adapted from :cite:p:`Shields2021`).
 
     Takes the low and high dimensional limits of
     :class:`baybe.surrogates.gaussian_process.presets.edbo.EDBOLikelihoodFactory`
     and interpolates the prior moments linearly in between.
-    """
+    """  # noqa: E501
 
     @override
     def __call__(

docs/index.md

@@ -40,6 +40,7 @@ Contribute <misc/contributing_link>
 Contributors <misc/contributors_link>
 Known Issues <known_issues>
 Changelog <misc/changelog_link>
+References <misc/references>
 Github <https://github.com/emdgroup/baybe/>
 License <misc/license_link>
 ```

docs/misc/references.md

@@ -0,0 +1,4 @@
+# References
+
+```{bibliography}
+```

docs/references.bib

@@ -8,3 +8,22 @@ @inproceedings{NIPS2007_66368270
  volume = {20},
  year = {2007}
 }
+
+@article{Chen2026,
+ author = {Chen, Guanming and Fleck, Maximilian and Stuyver, Thijs},
+ title = {Leveraging Hidden-Space Representations Effectively in Bayesian Optimization for Experiment Design through Dimension-Aware Hyperpriors},
+ journal = {ChemRxiv},
+ year = {2026},
+ doi = {10.26434/chemrxiv.10001986/v2}
+}
+
+@article{Shields2021,
+ author = {Shields, Benjamin J. and Stevens, Jason and Li, Jun and Parasram, Marvin and Damani, Farhan and Alvarado, Jesus I. Martinez and Janey, Jacob M. and Adams, Ryan P. and Doyle, Abigail G.},
+ title = {Bayesian reaction optimization as a tool for chemical synthesis},
+ journal = {Nature},
+ volume = {590},
+ number = {7844},
+ pages = {89--96},
+ year = {2021},
+ doi = {10.1038/s41586-021-03213-y}
+}

docs/userguide/transfer_learning.md

@@ -185,7 +185,4 @@ on the optimization:
 :class: only-dark
 ```
 
-```{bibliography}
-```
-
 [`TaskParameter`]: baybe.parameters.categorical.TaskParameter