Merge: Add Continuous Optimization Controls to BotorchRecommender (#389)

BotorchRecommender now has the optional keywords `n_restarts` and
`n_raw_samples` which control continuous optimization behavior and can
be increased if the user desires. The defaults have been slighlty
increased as well as previous values seemed extremely low.

CHANGELOG.md

@@ -5,6 +5,10 @@ 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]
+### Added
+- `n_restarts` and `n_raw_samples` keywords to configure continuous optimization
+  behavior for `BotorchRecommender`
+
 ### Fixed
 - Leftover attrs-decorated classes are garbage collected before the subclass tree is
   traversed, avoiding sporadic serialization problems 
@@ -23,7 +27,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Leftover attrs-decorated classes are garbage collected before the subclass tree is
   traversed, avoiding sporadic serialization problems 
 
-### Deprecated
+### Deprecations
 - `ContinuousLinearEqualityConstraint` and `ContinuousLinearInequalityConstraint`
   replaced by `ContinuousLinearConstraint` with the corresponding `operator` keyword
 

baybe/recommenders/pure/bayesian/botorch.py

@@ -5,8 +5,9 @@
 from typing import Any, ClassVar
 
 import pandas as pd
-from attr.converters import optional
 from attrs import define, field
+from attrs.converters import optional as optional_c
+from attrs.validators import gt, instance_of
 
 from baybe.acquisition.acqfs import qThompsonSampling
 from baybe.exceptions import (
@@ -50,12 +51,12 @@ class BotorchRecommender(BayesianRecommender):
     # Object variables
     sequential_continuous: bool = field(default=False)
     """Flag defining whether to apply sequential greedy or batch optimization in
-    **continuous** search spaces. (In discrete/hybrid spaces, sequential greedy
-    optimization is applied automatically.)
+    **continuous** search spaces. In discrete/hybrid spaces, sequential greedy
+    optimization is applied automatically.
     """
 
     hybrid_sampler: DiscreteSamplingMethod | None = field(
-        converter=optional(DiscreteSamplingMethod), default=None
+        converter=optional_c(DiscreteSamplingMethod), default=None
     )
     """Strategy used for sampling the discrete subspace when performing hybrid search
     space optimization."""
@@ -64,6 +65,16 @@ class BotorchRecommender(BayesianRecommender):
     """Percentage of discrete search space that is sampled when performing hybrid search
     space optimization. Ignored when ``hybrid_sampler="None"``."""
 
+    n_restarts: int = field(validator=[instance_of(int), gt(0)], default=10)
+    """Number of times gradient-based optimization is restarted from different initial
+    points. **Does not affect purely discrete optimization**.
+    """
+
+    n_raw_samples: int = field(validator=[instance_of(int), gt(0)], default=64)
+    """Number of raw samples drawn for the initialization heuristic in gradient-based
+    optimization. **Does not affect purely discrete optimization**.
+    """
+
     @sampling_percentage.validator
     def _validate_percentage(  # noqa: DOC101, DOC103
         self, _: Any, value: float
@@ -168,8 +179,8 @@ def _recommend_continuous(
             acq_function=self._botorch_acqf,
             bounds=torch.from_numpy(subspace_continuous.comp_rep_bounds.values),
             q=batch_size,
-            num_restarts=5,  # TODO make choice for num_restarts
-            raw_samples=10,  # TODO make choice for raw_samples
+            num_restarts=self.n_restarts,
+            raw_samples=self.n_raw_samples,
             equality_constraints=[
                 c.to_botorch(subspace_continuous.parameters)
                 for c in subspace_continuous.constraints_lin_eq
@@ -252,8 +263,8 @@ def _recommend_hybrid(
             acq_function=self._botorch_acqf,
             bounds=torch.from_numpy(searchspace.comp_rep_bounds.values),
             q=batch_size,
-            num_restarts=5,  # TODO make choice for num_restarts
-            raw_samples=10,  # TODO make choice for raw_samples
+            num_restarts=self.n_restarts,
+            raw_samples=self.n_raw_samples,
             fixed_features_list=fixed_features_list,
             equality_constraints=[
                 c.to_botorch(

docs/userguide/recommenders.md

@@ -38,16 +38,21 @@ for various acquisition functions.
 * The **[`BotorchRecommender`](baybe.recommenders.pure.bayesian.botorch.BotorchRecommender)**
   is a powerful recommender based on BoTorch's optimization engine that can be applied
   to all kinds of search spaces. In continuous spaces, its `sequential_continuous` flag
-  allows to chose between greedy sequential optimization and batch optimization as the
+  allows to choose between greedy sequential optimization and batch optimization as the
   underlying point generation mode. In discrete/hybrid spaces, sequential greedy
   selection is the only available mode and is thus activated automatically.
 
   Note that the recommender performs a brute-force search when applied to hybrid search
-  spaces, as it optimizes the continuous part of the space while exhaustively searching
-  choices in the discrete subspace. You can customize this behavior to only sample a
-  certain percentage of the discrete subspace via the `sample_percentage` attribute and
-  to choose different sampling algorithms via the `hybrid_sampler` attribute.
-
+  spaces, as it does gradient-based optimization in the continuous part of the space
+  while exhaustively evaluating configurations of the discrete subspace. You can customize this
+  behavior to only sample a certain percentage of the discrete subspace via the
+  `sample_percentage` attribute and to choose different sampling algorithms via the
+  `hybrid_sampler` attribute.
+
+  The gradient-based optimization part can also further be controlled by the
+  `n_restarts` and `n_raw_samples` keywords. For details, please refer
+  to [BotorchRecommender](baybe.recommenders.pure.bayesian.botorch.BotorchRecommender).
+
   An example on using this recommender in a hybrid space can be found
   [here](./../../examples/Backtesting/hybrid).