qThompsonSampling

class baybe.acquisition.acqfs.qThompsonSampling

Bases: qSimpleRegret

Thomson sampling, implemented via simple regret. Inherently Monte Carlo based.

This implementation exploits the fact that one-sample-based Thompson sampling (i.e. where the action probability is approximated using a single posterior sample) is equivalent to optimizing the Monte Carlo approximated posterior mean with sample size one. The latter can be achieved via qSimpleRegret and controlling its sample shape attribute.

Public methods

__init__()	Method generated by attrs for class qThompsonSampling.
evaluate(candidates, surrogate, searchspace, ...)	Get the acquisition values for the given candidates.
from_dict(dictionary)	Create an object from its dictionary representation.
from_json(string)	Create an object from its JSON representation.
to_botorch(surrogate, searchspace, ...[, ...])	Create the botorch-ready representation of the function.
to_dict()	Create an object's dictionary representation.
to_json()	Create an object's JSON representation.

Public attributes and properties

n_mc_samples	Number of Monte Carlo samples drawn from the posterior at each design point.
abbreviation	An alternative name for type resolution.
supports_multi_output	Whether this acquisition function can handle models with multiple outputs.

__init__()

Method generated by attrs for class qThompsonSampling.

For details on the parameters, see Public attributes and properties.

evaluate(candidates: DataFrame, surrogate: SurrogateProtocol, searchspace: SearchSpace, objective: Objective, measurements: DataFrame, pending_experiments: DataFrame | None = None, *, jointly: bool = False)

Get the acquisition values for the given candidates.

Parameters:

• candidates (DataFrame) – The candidate points in experimental representation. For details, see baybe.surrogates.base.Surrogate.posterior().

• surrogate (SurrogateProtocol) – The surrogate model to use for the acquisition function.

• searchspace (SearchSpace) – The search space. See baybe.recommenders.base.RecommenderProtocol.recommend().

• objective (Objective) – The objective. See baybe.recommenders.base.RecommenderProtocol.recommend().

• measurements (DataFrame) – Available experimentation data. See baybe.recommenders.base.RecommenderProtocol.recommend().

• pending_experiments (Optional[DataFrame]) – Optional pending experiments. See baybe.recommenders.base.RecommenderProtocol.recommend().

• jointly (bool) – If False, the acquisition values are computed independently for each candidate. If True, a single joint acquisition value is computed for the entire candidate set.

Return type:

Series | float

Returns:

Depending on the joint mode, either a single batch acquisition value or a series of individual acquisition values.

classmethod from_dict(dictionary: dict)

Create an object from its dictionary representation.

Parameters:

dictionary (dict) – The dictionary representation.

Return type:

TypeVar(_T)

Returns:

The reconstructed object.

classmethod from_json(string: str)

Create an object from its JSON representation.

Parameters:

string (str) – The JSON representation of the object.

Return type:

TypeVar(_T)

Returns:

The reconstructed object.

to_botorch(surrogate: SurrogateProtocol, searchspace: SearchSpace, objective: Objective, measurements: DataFrame, pending_experiments: DataFrame | None = None)

Create the botorch-ready representation of the function.

The required structure of measurements is specified in baybe.recommenders.base.RecommenderProtocol.recommend().

Return type:

AcquisitionFunction

to_dict()

Create an object’s dictionary representation.

Return type:

dict

to_json()

Create an object’s JSON representation.

Return type:

str

Returns:

The JSON representation as a string.

abbreviation: ClassVar[str] = 'qTS'

An alternative name for type resolution.

n_mc_samples: int

Number of Monte Carlo samples drawn from the posterior at each design point.

Restring the the sample size to one allows us to emulate (one-sample based) Thompson sampling using the regular acquisition function machinery.

supports_multi_output: ClassVar[bool] = False

Whether this acquisition function can handle models with multiple outputs.