Botorch Continuous Search

BotorchContinuousSearch is a search function for continuous optimisation with BoTorch. It returns an empty candidate list, which signals BoTorchAcquisition to generate candidates by optimising the acquisition function directly in continuous space (via optimize_acqf) rather than scoring a discrete pool.

Use it with continuous search spaces where gradient-based candidate generation is desired; the acquisition function must be constructed with bounds.

Search functions for BoTorch-based continuous optimization.

This module provides search functions designed to work with BoTorch acquisition functions that can directly optimize candidates in continuous spaces.

class alf_tools.optimizer.search.botorch_continuous_search.BotorchContinuousSearch

Bases: BaseSearch

Search function for continuous optimization with BoTorch.

This search function returns an empty list of candidates, signaling to BoTorch acquisition functions that they should perform continuous optimization (via optimize_acqf) rather than scoring a discrete pool.

This is typically used with: - BoTorchSyntheticDataset (continuous test functions) - Any continuous search space where gradient-based optimization is desired

Example

>>> from alf_tools.optimizer.search import BotorchContinuousSearch
>>> from alf_tools.optimizer.acquisition_functions import BoTorchAcquisition
>>> from alf_core import Optimizer
>>>
>>> optimizer = Optimizer(
...     acquisition_fn=BoTorchAcquisition(
...         acquisition_type="qEI",
...         bounds=[[0, 1], [0, 1]]
...     ),
...     search_fn=BotorchContinuousSearch()
... )

get_metrics(task_state)

Return empty metrics dict.

Parameters:

task_state (State) – Current task state.

Return type:

dict[str, float]

Returns:

Empty dictionary (no search-specific metrics for continuous optimization).