Refactor recommender signature

CHANGELOG.md

@@ -4,6 +4,13 @@ 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
+- Passing an `Objective` to `Campaign` is now optional
+
+### Breaking Changes
+- Providing an explicit `batch_size` is now mandatory when asking for recommendations
+
 ## [0.9.1] - 2024-06-04
 ### Changed
 - Discrete searchspace memory estimate is now natively represented in bytes 
@@ -52,6 +59,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Environment variables `BAYBE_NUMPY_USE_SINGLE_PRECISION` and
   `BAYBE_TORCH_USE_SINGLE_PRECISION` to enforce single point precision usage
 
+### Breaking Changes
+- `RecommenderProtocol.recommend` now accepts an optional `Objective` 
+- `RecommenderProtocol.recommend` now expects training data to be provided as a single
+  dataframe in experimental representation instead of two separate dataframes in
+  computational representation
+
 ### Removed
 - `model_params` attribute from `Surrogate` base class, `GaussianProcessSurrogate` and
   `CustomONNXSurrogate`

baybe/campaign.py

@@ -3,11 +3,13 @@
 from __future__ import annotations
 
 import json
+from typing import Optional
 
 import cattrs
 import numpy as np
 import pandas as pd
 from attrs import define, field
+from attrs.converters import optional
 
 from baybe.exceptions import DeprecationError
 from baybe.objectives.base import Objective, to_objective
@@ -47,7 +49,9 @@ class Campaign(SerialMixin):
     searchspace: SearchSpace = field()
     """The search space in which the experiments are conducted."""
 
-    objective: Objective = field(converter=to_objective)
+    objective: Optional[Objective] = field(
+        default=None, converter=optional(to_objective)
+    )
     """The optimization objective.
     When passing a single :class:`baybe.targets.base.Target`, it gets automatically
     wrapped into a :class:`baybe.objectives.single.SingleTargetObjective`."""
@@ -127,21 +131,7 @@ def parameters(self) -> tuple[Parameter, ...]:
     @property
     def targets(self) -> tuple[Target, ...]:
         """The targets of the underlying objective."""
-        return self.objective.targets
-
-    @property
-    def _measurements_parameters_comp(self) -> pd.DataFrame:
-        """The computational representation of the measured parameters."""
-        if len(self._measurements_exp) < 1:
-            return pd.DataFrame()
-        return self.searchspace.transform(self._measurements_exp)
-
-    @property
-    def _measurements_targets_comp(self) -> pd.DataFrame:
-        """The computational representation of the measured targets."""
-        if len(self._measurements_exp) < 1:
-            return pd.DataFrame()
-        return self.objective.transform(self._measurements_exp)
+        return self.objective.targets if self.objective is not None else ()
 
     @classmethod
     def from_config(cls, config_json: str) -> Campaign:
@@ -258,7 +248,7 @@ def add_measurements(
 
     def recommend(
         self,
-        batch_size: int = 5,
+        batch_size: int,
         batch_quantity: int = None,  # type: ignore[assignment]
     ) -> pd.DataFrame:
         """Provide the recommendations for the next batch of experiments.
@@ -298,10 +288,10 @@ def recommend(
 
         # Get the recommended search space entries
         rec = self.recommender.recommend(
-            self.searchspace,
             batch_size,
-            self._measurements_parameters_comp,
-            self._measurements_targets_comp,
+            self.searchspace,
+            self.objective,
+            self._measurements_exp,
         )
 
         # Cache the recommendations

baybe/recommenders/base.py

@@ -5,6 +5,7 @@
 import cattrs
 import pandas as pd
 
+from baybe.objectives.base import Objective
 from baybe.recommenders.deprecation import structure_recommender_protocol
 from baybe.searchspace import SearchSpace
 from baybe.serialization import converter, unstructure_base
@@ -15,21 +16,29 @@ class RecommenderProtocol(Protocol):
 
     def recommend(
         self,
-        searchspace: SearchSpace,
         batch_size: int,
-        train_x: Optional[pd.DataFrame],
-        train_y: Optional[pd.DataFrame],
+        searchspace: SearchSpace,
+        objective: Optional[Objective],
+        measurements: Optional[pd.DataFrame],
     ) -> pd.DataFrame:
         """Recommend a batch of points from the given search space.
 
         Args:
-            searchspace: The search space from which to recommend the points.
             batch_size: The number of points to be recommended.
-            train_x: Optional training inputs for training a model.
-            train_y: Optional training labels for training a model.
+            searchspace: The search space from which to recommend the points.
+            objective: An optional objective to be optimized.
+            measurements: Optional experimentation data that can be used for model
+                training. The data is to be provided in "experimental representation":
+                It needs to contain one column for each parameter spanning the search
+                space (column name matching the parameter name) and one column for each
+                target tracked by the objective (column name matching the target name).
+                Each row corresponds to one conducted experiment, where the parameter
+                columns define the experimental setting and the target columns report
+                the measured outcomes.
 
         Returns:
-            A dataframe containing the recommendations as individual rows.
+            A dataframe containing the recommendations in experimental representation
+            as individual rows.
         """
         ...
 

baybe/recommenders/meta/base.py

@@ -8,6 +8,7 @@
 from attrs import define, field
 
 from baybe.exceptions import DeprecationError
+from baybe.objectives.base import Objective
 from baybe.recommenders.base import RecommenderProtocol
 from baybe.recommenders.deprecation import structure_recommender_protocol
 from baybe.recommenders.pure.base import PureRecommender
@@ -50,35 +51,47 @@ def _validate_allow_recommending_already_measured(self, _, value):
     @abstractmethod
     def select_recommender(
         self,
+        batch_size: int,
         searchspace: SearchSpace,
-        batch_size: int = 1,
-        train_x: Optional[pd.DataFrame] = None,
-        train_y: Optional[pd.DataFrame] = None,
+        objective: Optional[Objective] = None,
+        measurements: Optional[pd.DataFrame] = None,
     ) -> PureRecommender:
         """Select a pure recommender for the given experimentation context.
 
         Args:
+            batch_size:
+                See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
             searchspace:
                 See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
-            batch_size:
+            objective:
+                See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
+            measurements:
                 See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
-            train_x: See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
-            train_y: See :func:`baybe.recommenders.meta.base.MetaRecommender.recommend`.
 
         Returns:
             The selected recommender.
         """
 
     def recommend(
         self,
+        batch_size: int,
         searchspace: SearchSpace,
-        batch_size: int = 1,
-        train_x: Optional[pd.DataFrame] = None,
-        train_y: Optional[pd.DataFrame] = None,
+        objective: Optional[Objective] = None,
+        measurements: Optional[pd.DataFrame] = None,
     ) -> pd.DataFrame:
         """See :func:`baybe.recommenders.base.RecommenderProtocol.recommend`."""
-        recommender = self.select_recommender(searchspace, batch_size, train_x, train_y)
-        return recommender.recommend(searchspace, batch_size, train_x, train_y)
+        recommender = self.select_recommender(
+            batch_size=batch_size,
+            searchspace=searchspace,
+            objective=objective,
+            measurements=measurements,
+        )
+        return recommender.recommend(
+            batch_size=batch_size,
+            searchspace=searchspace,
+            objective=objective,
+            measurements=measurements,
+        )
 
 
 # Register (un-)structure hooks

baybe/recommenders/meta/sequential.py

@@ -11,12 +11,12 @@
 from attrs.validators import deep_iterable, in_, instance_of
 
 from baybe.exceptions import NoRecommendersLeftError
+from baybe.objectives.base import Objective
 from baybe.recommenders.meta.base import MetaRecommender
 from baybe.recommenders.pure.base import PureRecommender
 from baybe.recommenders.pure.bayesian.sequential_greedy import (
     SequentialGreedyRecommender,
 )
-from baybe.recommenders.pure.nonpredictive.base import NonPredictiveRecommender
 from baybe.recommenders.pure.nonpredictive.sampling import RandomRecommender
 from baybe.searchspace import SearchSpace
 from baybe.serialization import (
@@ -25,12 +25,6 @@
     converter,
 )
 
-# TODO: Make bayesian recommenders handle empty training data
-_unsupported_recommender_error = ValueError(
-    f"For cases where no training is available, the selected recommender "
-    f"must be a subclass of '{NonPredictiveRecommender.__name__}'."
-)
-
 
 @define
 class TwoPhaseMetaRecommender(MetaRecommender):
@@ -59,22 +53,16 @@ class TwoPhaseMetaRecommender(MetaRecommender):
 
     def select_recommender(  # noqa: D102
         self,
-        searchspace: SearchSpace,
-        batch_size: int = 1,
-        train_x: Optional[pd.DataFrame] = None,
-        train_y: Optional[pd.DataFrame] = None,
+        batch_size: int,
+        searchspace: Optional[SearchSpace] = None,
+        objective: Optional[Objective] = None,
+        measurements: Optional[pd.DataFrame] = None,
     ) -> PureRecommender:
         # See base class.
 
-        # TODO: enable bayesian recommenders for empty training data
-        if (train_x is None or len(train_x) == 0) and not isinstance(
-            self.initial_recommender, NonPredictiveRecommender
-        ):
-            raise _unsupported_recommender_error
-
         return (
             self.recommender
-            if len(train_x) >= self.switch_after
+            if (measurements is not None) and (len(measurements) >= self.switch_after)
             else self.initial_recommender
         )
 
@@ -95,6 +83,8 @@ class SequentialMetaRecommender(MetaRecommender):
         instead.
 
     Raises:
+        RuntimeError: If the training dataset size decreased compared to the previous
+            call.
         NoRecommendersLeftError: If more recommenders are requested than there are
             recommenders available and ``mode="raise"``.
     """
@@ -134,21 +124,24 @@ class SequentialMetaRecommender(MetaRecommender):
 
     def select_recommender(  # noqa: D102
         self,
-        searchspace: SearchSpace,
-        batch_size: int = 1,
-        train_x: Optional[pd.DataFrame] = None,
-        train_y: Optional[pd.DataFrame] = None,
+        batch_size: int,
+        searchspace: Optional[SearchSpace] = None,
+        objective: Optional[Objective] = None,
+        measurements: Optional[pd.DataFrame] = None,
     ) -> PureRecommender:
         # See base class.
 
+        n_data = len(measurements) if measurements is not None else 0
+
         # If the training dataset size has increased, move to the next recommender
-        if len(train_x) > self._n_last_measurements:
+        if n_data > self._n_last_measurements:
             self._step += 1
+
         # If the training dataset size has decreased, something went wrong
-        elif len(train_x) < self._n_last_measurements:
+        elif n_data < self._n_last_measurements:
             raise RuntimeError(
                 f"The training dataset size decreased from {self._n_last_measurements} "
-                f"to {len(train_x)} since the last function call, which indicates that "
+                f"to {n_data} since the last function call, which indicates that "
                 f"'{self.__class__.__name__}' was not used as intended."
             )
 
@@ -169,13 +162,7 @@ def select_recommender(  # noqa: D102
             ) from ex
 
         # Remember the training dataset size for the next call
-        self._n_last_measurements = len(train_x)
-
-        # TODO: enable bayesian recommenders for empty training data
-        if (train_x is None or len(train_x) == 0) and not isinstance(
-            recommender, NonPredictiveRecommender
-        ):
-            raise _unsupported_recommender_error
+        self._n_last_measurements = n_data
 
         return recommender
 
@@ -219,24 +206,26 @@ def default_iterator(self):
 
     def select_recommender(  # noqa: D102
         self,
-        searchspace: SearchSpace,
-        batch_size: int = 1,
-        train_x: Optional[pd.DataFrame] = None,
-        train_y: Optional[pd.DataFrame] = None,
+        batch_size: int,
+        searchspace: Optional[SearchSpace] = None,
+        objective: Optional[Objective] = None,
+        measurements: Optional[pd.DataFrame] = None,
     ) -> PureRecommender:
         # See base class.
 
         use_last = True
+        n_data = len(measurements) if measurements is not None else 0
 
         # If the training dataset size has increased, move to the next recommender
-        if len(train_x) > self._n_last_measurements:
+        if n_data > self._n_last_measurements:
             self._step += 1
             use_last = False
+
         # If the training dataset size has decreased, something went wrong
-        elif len(train_x) < self._n_last_measurements:
+        elif n_data < self._n_last_measurements:
             raise RuntimeError(
                 f"The training dataset size decreased from {self._n_last_measurements} "
-                f"to {len(train_x)} since the last function call, which indicates that "
+                f"to {n_data} since the last function call, which indicates that "
                 f"'{self.__class__.__name__}' was not used as intended."
             )
 
@@ -251,13 +240,7 @@ def select_recommender(  # noqa: D102
             ) from ex
 
         # Remember the training dataset size for the next call
-        self._n_last_measurements = len(train_x)
-
-        # TODO: enable bayesian recommenders for empty training data
-        if (train_x is None or len(train_x) == 0) and not isinstance(
-            self._last_recommender, NonPredictiveRecommender
-        ):
-            raise _unsupported_recommender_error
+        self._n_last_measurements = n_data
 
         return self._last_recommender  # type: ignore[return-value]