botorch.acquisition

Acquisition Function APIs

Abstract Acquisition Function APIs

Abstract base module for all botorch acquisition functions.

class botorch.acquisition.acquisition.AcquisitionFunction(model)[source]

Bases: Module, ABC

Abstract base class for acquisition functions.

Please note that if your acquisition requires a backwards call, you will need to wrap the backwards call inside of an enable_grad context to be able to optimize the acquisition. See #1164.

Constructor for the AcquisitionFunction base class.

Parameters:

model (Model) – A fitted model.

set_X_pending(X_pending=None)[source]

Informs the acquisition function about pending design points.

Parameters:

X_pending (Tensor | None) – n x d Tensor with n d-dim design points that have been submitted for evaluation but have not yet been evaluated.

Return type:

None

abstract forward(X)[source]

Evaluate the acquisition function on the candidate set X.

Parameters:

X (Tensor) – A (b) x q x d-dim Tensor of (b) t-batches with q d-dim design points each.

Returns:

A (b)-dim Tensor of acquisition function values at the given design points X.

Return type:

Tensor

class botorch.acquisition.acquisition.OneShotAcquisitionFunction(model)[source]

Bases: AcquisitionFunction, ABC

Abstract base class for acquisition functions using one-shot optimization

Constructor for the AcquisitionFunction base class.

Parameters:

model (Model) – A fitted model.

abstract get_augmented_q_batch_size(q)[source]

Get augmented q batch size for one-shot optimization.

Parameters:

q (int) – The number of candidates to consider jointly.

Returns:

The augmented size for one-shot optimization (including variables parameterizing the fantasy solutions).

Return type:

int

abstract extract_candidates(X_full)[source]

Extract the candidates from a full “one-shot” parameterization.

Parameters:

X_full (Tensor) – A b x q_aug x d-dim Tensor with b t-batches of q_aug design points each.

Returns:

A b x q x d-dim Tensor with b t-batches of q design points each.

Return type:

Tensor

class botorch.acquisition.acquisition.MCSamplerMixin(sampler=None)[source]

Bases: ABC

A mix-in for adding sampler functionality into an acquisition function class.

_default_sample_shape

The sample_shape for the default sampler.

Register the sampler on the acquisition function.

Parameters:

sampler (Optional[MCSampler]) – The sampler used to draw base samples for MC-based acquisition functions. If None, a sampler is generated on the fly within the get_posterior_samples method using get_sampler.

get_posterior_samples(posterior)[source]

Sample from the posterior using the sampler.

Parameters:

posterior (Posterior) – The posterior to sample from.

Return type:

Tensor

property sample_shape: Size
class botorch.acquisition.acquisition.MultiModelAcquisitionFunction(model_dict)[source]

Bases: AcquisitionFunction, ABC

Abstract base class for acquisition functions that require multiple types of models.

The intended use case for these acquisition functions are those where we have multiple models, each serving a distinct purpose. As an example, we can have a “regression” model that predicts one or more outcomes, and a “classification” model that predicts the probabilty that a given parameterization is feasible. The multi-model acquisition function can then weight the acquisition value computed with the “regression” model with the feasibility value predicted by the “classification” model to produce the composite acquisition value.

This is currently only a placeholder to help with some development in Ax. We plan to add some acquisition functions utilizing multiple models in the future.

Constructor for the MultiModelAcquisitionFunction base class.

Parameters:

model_dict (ModelDict) – A ModelDict mapping labels to models.

Analytic Acquisition Function API

class botorch.acquisition.analytic.AnalyticAcquisitionFunction(model, posterior_transform=None)[source]

Bases: AcquisitionFunction, ABC

Base class for analytic acquisition functions.

Base constructor for analytic acquisition functions.

Parameters:
  • model (Model) – A fitted single-outcome model.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

set_X_pending(X_pending=None)[source]

Informs the acquisition function about pending design points.

Parameters:

X_pending (Tensor | None) – n x d Tensor with n d-dim design points that have been submitted for evaluation but have not yet been evaluated.

Return type:

None

Cached Cholesky Acquisition Function API

Abstract class for acquisition functions leveraging a cached Cholesky decomposition of the posterior covariance over f(X_baseline).

botorch.acquisition.cached_cholesky.supports_cache_root(model)[source]

Checks if a model supports the cache_root functionality. The two criteria are that the model is not multi-task and the model produces a GPyTorchPosterior.

Parameters:

model (Model)

Return type:

bool

class botorch.acquisition.cached_cholesky.CachedCholeskyMCSamplerMixin(model, cache_root=False, sampler=None)[source]

Bases: MCSamplerMixin

Abstract Mixin class for acquisition functions using a cached Cholesky.

Specifically, this is for acquisition functions that require sampling from the posterior P(f(X_baseline, X) | D). The Cholesky of the posterior covariance over f(X_baseline) is cached.

Set class attributes and perform compatibility checks.

Parameters:
  • model (Model) – A model.

  • cache_root (bool) – A boolean indicating whether to cache the Cholesky. This might be overridden in the model is not compatible.

  • sampler (Optional[MCSampler]) – An optional MCSampler object.

Decoupled Acquisition Function API

Abstract base module for decoupled acquisition functions.

class botorch.acquisition.decoupled.DecoupledAcquisitionFunction(model, X_evaluation_mask=None, **kwargs)[source]

Bases: AcquisitionFunction, ABC

Abstract base class for decoupled acquisition functions. A decoupled acquisition function where one may intend to evaluate a design on only a subset of the outcomes. Typically this would be handled by fantasizing, where one would fantasize as to what the partial observation would be if one were to evaluate a design on the subset of outcomes (e.g. you only fantasize at those outcomes). The X_evaluation_mask specifies which outcomes should be evaluated for each design. X_evaluation_mask is q x m, where there are q design points in the batch and m outcomes. In the asynchronous case, where there are n’ pending points, we need to track which outcomes each pending point should be evaluated on. In this case, we concatenate X_pending_evaluation_mask with X_evaluation_mask to obtain the full evaluation_mask.

This abstract class handles generating and updating an evaluation mask, which is a boolean tensor indicating which outcomes a given design is being evaluated on. The evaluation mask has shape (n’ + q) x m, where n’ is the number of pending points and the q represents the new candidates to be generated.

If X(_pending)_evaluation_mas`k is None, it is assumed that `X(_pending) will be evaluated on all outcomes.

Initialize.

Parameters:
  • model (ModelList) – A model

  • X_evaluation_mask (Optional[Tensor]) – A q x m-dim boolean tensor indicating which outcomes the decoupled acquisition function should generate new candidates for.

property X_evaluation_mask: Tensor | None

Get the evaluation indices for the new candidate.

set_X_pending(X_pending=None, X_pending_evaluation_mask=None)[source]

Informs the AF about pending design points for different outcomes.

Parameters:
  • X_pending (Tensor | None) – A n’ x d Tensor with n’ d-dim design points that have been submitted for evaluation but have not yet been evaluated.

  • X_pending_evaluation_mask (Tensor | None) – A n’ x m-dim tensor of booleans indicating for which outputs the pending point is being evaluated on. If X_pending_evaluation_mask is None, it is assumed that X_pending will be evaluated on all outcomes.

Return type:

None

construct_evaluation_mask(X)[source]

Construct the boolean evaluation mask for X and X_pending

Parameters:

X (Tensor) – A batch_shape x n x d-dim tensor of designs.

Returns:

A n + n’ x m-dim tensor of booleans indicating which outputs should be evaluated.

Return type:

Tensor | None

Monte-Carlo Acquisition Function API

class botorch.acquisition.monte_carlo.MCAcquisitionFunction(model, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Bases: AcquisitionFunction, MCSamplerMixin, ABC

Abstract base class for Monte-Carlo based batch acquisition functions.

Parameters:
  • model (Model) – A fitted model.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated on the fly within the get_posterior_samples method using botorch.sampling.get_sampler. NOTE: For posteriors that do not support base samples, a sampler compatible with intended use case must be provided. See ForkedRNGSampler and StochasticSampler as examples.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape, m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

abstract forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

Base Classes for Multi-Objective Acquisition Function API

Base classes for multi-objective acquisition functions.

class botorch.acquisition.multi_objective.base.MultiObjectiveAnalyticAcquisitionFunction(model, posterior_transform=None)[source]

Bases: AcquisitionFunction

Abstract base class for Multi-Objective batch acquisition functions.

Constructor for the MultiObjectiveAnalyticAcquisitionFunction base class.

Parameters:
  • model (Model) – A fitted model.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

abstract forward(X)[source]

Takes in a batch_shape x 1 x d X Tensor of t-batches with 1 d-dim design point each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X.

Parameters:

X (Tensor)

Return type:

Tensor

set_X_pending(X_pending=None)[source]

Informs the acquisition function about pending design points.

Parameters:

X_pending (Tensor | None) – n x d Tensor with n d-dim design points that have been submitted for evaluation but have not yet been evaluated.

Return type:

None

class botorch.acquisition.multi_objective.base.MultiObjectiveMCAcquisitionFunction(model, sampler=None, objective=None, constraints=None, eta=0.001, X_pending=None)[source]

Bases: AcquisitionFunction, MCSamplerMixin, ABC

Abstract base class for Multi-Objective batch acquisition functions.

NOTE: This does not inherit from MCAcquisitionFunction to avoid circular imports.

Parameters:
  • _default_sample_shape – The sample_shape for the default sampler.

  • model (Model)

  • sampler (Optional[MCSampler])

  • objective (Optional[MCMultiOutputObjective])

  • constraints (Optional[list[Callable[[Tensor], Tensor]]])

  • eta (Union[Tensor, float])

  • X_pending (Optional[Tensor])

Constructor for the MultiObjectiveMCAcquisitionFunction base class.

Parameters:
  • model (Model) – A fitted model.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler. NOTE: For posteriors that do not support base samples, a sampler compatible with intended use case must be provided. See ForkedRNGSampler and StochasticSampler as examples.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMCMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

abstract forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

Acquisition Functions

Analytic Acquisition Functions

Analytic Acquisition Functions that evaluate the posterior without performing Monte-Carlo sampling.

class botorch.acquisition.analytic.LogProbabilityOfImprovement(model, best_f, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Log Probability of Improvement.

Logarithm of the probability of improvement over the current best observed value, computed using the analytic formula under a Normal posterior distribution. Only supports the case of q=1. Requires the posterior to be Gaussian. The model must be single-outcome.

The logarithm of the probability of improvement is numerically better behaved than the original function, which can lead to significantly improved optimization of the acquisition function. This is analogous to the common practice of optimizing the log likelihood of a probabilistic model - rather the likelihood - for the sake of maximium likelihood estimation.

logPI(x) = log(P(y >= best_f)), y ~ f(x)

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> LogPI = LogProbabilityOfImprovement(model, best_f=0.2)
>>> log_pi = LogPI(test_X)

Single-outcome Probability of Improvement.

Parameters:
  • model (Model) – A fitted single-outcome model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best function value observed so far (assumed noiseless).

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate the Log Probability of Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk)-dim tensor of Log Probability of Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.ProbabilityOfImprovement(model, best_f, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Probability of Improvement.

Probability of improvement over the current best observed value, computed using the analytic formula under a Normal posterior distribution. Only supports the case of q=1. Requires the posterior to be Gaussian. The model must be single-outcome.

PI(x) = P(y >= best_f), y ~ f(x)

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> PI = ProbabilityOfImprovement(model, best_f=0.2)
>>> pi = PI(test_X)

Single-outcome Probability of Improvement.

Parameters:
  • model (Model) – A fitted single-outcome model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best function value observed so far (assumed noiseless).

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate the Probability of Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk)-dim tensor of Probability of Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.qAnalyticProbabilityOfImprovement(model, best_f, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Approximate, single-outcome batch Probability of Improvement using MVNXPB.

This implementation uses MVNXPB, a bivariate conditioning algorithm for approximating P(a <= Y <= b) for multivariate normal Y. See [Trinh2015bivariate]. This (analytic) approximate q-PI is given by approx-qPI(X) = P(max Y >= best_f) = 1 - P(Y < best_f), Y ~ f(X), X = (x_1,…,x_q), where P(Y < best_f) is estimated using MVNXPB.

qPI using an analytic approximation.

Parameters:
  • model (Model) – A fitted single-outcome model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best function value observed so far (assumed noiseless).

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate approximate qPI on the candidate set X.

Parameters:

X (Tensor) – A batch_shape x q x d-dim Tensor of t-batches with q d-dim design points each

Returns:

A batch_shape-dim Tensor of approximate Probability of Improvement values at the given design points X, where batch_shape’ is the broadcasted batch shape of model and input X.

Return type:

Tensor

class botorch.acquisition.analytic.ExpectedImprovement(model, best_f, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Expected Improvement (analytic).

Computes classic Expected Improvement over the current best observed value, using the analytic formula for a Normal posterior distribution. Unlike the MC-based acquisition functions, this relies on the posterior at single test point being Gaussian (and require the posterior to implement mean and variance properties). Only supports the case of q=1. The model must be single-outcome.

EI(x) = E(max(f(x) - best_f, 0)),

where the expectation is taken over the value of stochastic function f at x.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> EI = ExpectedImprovement(model, best_f=0.2)
>>> ei = EI(test_X)

NOTE: It is strongly recommended to use LogExpectedImprovement instead of regular EI, as it can lead to substantially improved BO performance through improved numerics. See https://arxiv.org/abs/2310.20708 for details.

Single-outcome Expected Improvement (analytic).

Parameters:
  • model (Model) – A fitted single-outcome model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best function value observed so far (assumed noiseless).

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points. Expected Improvement is computed for each point individually, i.e., what is considered are the marginal posteriors, not the joint.

Returns:

A (b1 x … bk)-dim tensor of Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.LogExpectedImprovement(model, best_f, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Log Expected Improvement (analytic).

Computes the logarithm of the classic Expected Improvement acquisition function, in a numerically robust manner. In particular, the implementation takes special care to avoid numerical issues in the computation of the acquisition value and its gradient in regions where improvement is predicted to be virtually impossible.

See [Ament2023logei] for details. Formally,

LogEI(x) = log(E(max(f(x) - best_f, 0))),

where the expectation is taken over the value of stochastic function f at x.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> LogEI = LogExpectedImprovement(model, best_f=0.2)
>>> ei = LogEI(test_X)

Logarithm of single-outcome Expected Improvement (analytic).

Parameters:
  • model (Model) – A fitted single-outcome model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best function value observed so far (assumed noiseless).

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate logarithm of Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points. Expected Improvement is computed for each point individually, i.e., what is considered are the marginal posteriors, not the joint.

Returns:

A (b1 x … bk)-dim tensor of the logarithm of the Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.LogConstrainedExpectedImprovement(model, best_f, objective_index, constraints, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Log Constrained Expected Improvement (feasibility-weighted).

Computes the logarithm of the analytic expected improvement for a Normal posterior distribution weighted by a probability of feasibility. The objective and constraints are assumed to be independent and have Gaussian posterior distributions. Only supports non-batch mode (i.e. q=1). The model should be multi-outcome, with the index of the objective and constraints passed to the constructor.

See [Ament2023logei] for details. Formally,

LogConstrainedEI(x) = log(EI(x)) + Sum_i log(P(y_i in [lower_i, upper_i])),

where y_i ~ constraint_i(x) and lower_i, upper_i are the lower and upper bounds for the i-th constraint, respectively.

Example

# example where the 0th output has a non-negativity constraint and # the 1st output is the objective >>> model = SingleTaskGP(train_X, train_Y) >>> constraints = {0: (0.0, None)} >>> LogCEI = LogConstrainedExpectedImprovement(model, 0.2, 1, constraints) >>> cei = LogCEI(test_X)

Analytic Log Constrained Expected Improvement.

Parameters:
  • model (Model) – A fitted multi-output model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best feasible function value observed so far (assumed noiseless).

  • objective_index (int) – The index of the objective.

  • constraints (dict[int, tuple[Optional[float], Optional[float]]]) – A dictionary of the form {i: [lower, upper]}, where i is the output index, and lower and upper are lower and upper bounds on that output (resp. interpreted as -Inf / Inf if None)

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate Constrained Log Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b) x 1 x d-dim Tensor of (b) t-batches of d-dim design points each.

Returns:

A (b)-dim Tensor of Log Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.ConstrainedExpectedImprovement(model, best_f, objective_index, constraints, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Constrained Expected Improvement (feasibility-weighted).

Computes the analytic expected improvement for a Normal posterior distribution, weighted by a probability of feasibility. The objective and constraints are assumed to be independent and have Gaussian posterior distributions. Only supports non-batch mode (i.e. q=1). The model should be multi-outcome, with the index of the objective and constraints passed to the constructor.

Constrained_EI(x) = EI(x) * Product_i P(y_i in [lower_i, upper_i]), where y_i ~ constraint_i(x) and lower_i, upper_i are the lower and upper bounds for the i-th constraint, respectively.

Example

# example where the 0th output has a non-negativity constraint and # 1st output is the objective >>> model = SingleTaskGP(train_X, train_Y) >>> constraints = {0: (0.0, None)} >>> cEI = ConstrainedExpectedImprovement(model, 0.2, 1, constraints) >>> cei = cEI(test_X)

NOTE: It is strongly recommended to use LogConstrainedExpectedImprovement instead of regular CEI, as it can lead to substantially improved BO performance through improved numerics. See https://arxiv.org/abs/2310.20708 for details.

Analytic Constrained Expected Improvement.

Parameters:
  • model (Model) – A fitted multi-output model.

  • best_f (Union[float, Tensor]) – Either a scalar or a b-dim Tensor (batch mode) representing the best feasible function value observed so far (assumed noiseless).

  • objective_index (int) – The index of the objective.

  • constraints (dict[int, tuple[Optional[float], Optional[float]]]) – A dictionary of the form {i: [lower, upper]}, where i is the output index, and lower and upper are lower and upper bounds on that output (resp. interpreted as -Inf / Inf if None)

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate Constrained Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A (b) x 1 x d-dim Tensor of (b) t-batches of d-dim design points each.

Returns:

A (b)-dim Tensor of Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.LogNoisyExpectedImprovement(model, X_observed, num_fantasies=20, maximize=True, posterior_transform=None)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Log Noisy Expected Improvement (via fantasies).

This computes Log Noisy Expected Improvement by averaging over the Expected Improvement values of a number of fantasy models. Only supports the case q=1. Assumes that the posterior distribution of the model is Gaussian. The model must be single-outcome.

See [Ament2023logei] for details. Formally,

LogNEI(x) = log(E(max(y - max Y_base), 0))), (y, Y_base) ~ f((x, X_base)),

where X_base are previously observed points.

Note: This acquisition function currently relies on using a SingleTaskGP with known observation noise. In other words, train_Yvar must be passed to the model. (required for noiseless fantasies).

Example

>>> model = SingleTaskGP(train_X, train_Y, train_Yvar=train_Yvar)
>>> LogNEI = LogNoisyExpectedImprovement(model, train_X)
>>> nei = LogNEI(test_X)

Single-outcome Noisy Log Expected Improvement (via fantasies).

Parameters:
  • model (GPyTorchModel) – A fitted single-outcome model. Only SingleTaskGP models with known observation noise are currently supported.

  • X_observed (Tensor) – A n x d Tensor of observed points that are likely to be the best observed points so far.

  • num_fantasies (int) – The number of fantasies to generate. The higher this number the more accurate the model (at the expense of model complexity and performance).

  • maximize (bool) – If True, consider the problem a maximization problem.

  • posterior_transform (Optional[PosteriorTransform])

forward(X)[source]

Evaluate logarithm of the mean Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A b1 x … bk x 1 x d-dim batched tensor of d-dim design points.

Returns:

A b1 x … bk-dim tensor of Log Noisy Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.NoisyExpectedImprovement(model, X_observed, num_fantasies=20, maximize=True)[source]

Bases: ExpectedImprovement

Single-outcome Noisy Expected Improvement (via fantasies).

This computes Noisy Expected Improvement by averaging over the Expected Improvement values of a number of fantasy models. Only supports the case q=1. Assumes that the posterior distribution of the model is Gaussian. The model must be single-outcome.

NEI(x) = E(max(y - max Y_baseline), 0)), (y, Y_baseline) ~ f((x, X_baseline)), where X_baseline are previously observed points.

Note: This acquisition function currently relies on using a SingleTaskGP with known observation noise. In other words, train_Yvar must be passed to the model. (required for noiseless fantasies).

Example

>>> model = SingleTaskGP(train_X, train_Y, train_Yvar=train_Yvar)
>>> NEI = NoisyExpectedImprovement(model, train_X)
>>> nei = NEI(test_X)

NOTE: It is strongly recommended to use LogNoisyExpectedImprovement instead of regular NEI, as it can lead to substantially improved BO performance through improved numerics. See https://arxiv.org/abs/2310.20708 for details.

Single-outcome Noisy Expected Improvement (via fantasies).

Parameters:
  • model (GPyTorchModel) – A fitted single-outcome model. Only SingleTaskGP models with known observation noise are currently supported.

  • X_observed (Tensor) – A n x d Tensor of observed points that are likely to be the best observed points so far.

  • num_fantasies (int) – The number of fantasies to generate. The higher this number the more accurate the model (at the expense of model complexity and performance).

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate Expected Improvement on the candidate set X.

Parameters:

X (Tensor) – A b1 x … bk x 1 x d-dim batched tensor of d-dim design points.

Returns:

A b1 x … bk-dim tensor of Noisy Expected Improvement values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.UpperConfidenceBound(model, beta, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Upper Confidence Bound (UCB).

Analytic upper confidence bound that comprises of the posterior mean plus an additional term: the posterior standard deviation weighted by a trade-off parameter, beta. Only supports the case of q=1 (i.e. greedy, non-batch selection of design points). The model must be single-outcome.

UCB(x) = mu(x) + sqrt(beta) * sigma(x), where mu and sigma are the posterior mean and standard deviation, respectively.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> UCB = UpperConfidenceBound(model, beta=0.2)
>>> ucb = UCB(test_X)

Single-outcome Upper Confidence Bound.

Parameters:
  • model (Model) – A fitted single-outcome GP model (must be in batch mode if candidate sets X will be)

  • beta (Union[float, Tensor]) – Either a scalar or a one-dim tensor with b elements (batch mode) representing the trade-off parameter between mean and covariance

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem.

forward(X)[source]

Evaluate the Upper Confidence Bound on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk)-dim tensor of Upper Confidence Bound values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.PosteriorMean(model, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Posterior Mean.

Only supports the case of q=1. Requires the model’s posterior to have a mean property. The model must be single-outcome.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> PM = PosteriorMean(model)
>>> pm = PM(test_X)

Single-outcome Posterior Mean.

Parameters:
  • model (Model) – A fitted single-outcome GP model (must be in batch mode if candidate sets X will be)

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem. Note that if maximize=False, the posterior mean is negated. As a consequence optimize_acqf(PosteriorMean(gp, maximize=False)) actually returns -1 * minimum of the posterior mean.

forward(X)[source]

Evaluate the posterior mean on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk)-dim tensor of Posterior Mean values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.ScalarizedPosteriorMean(model, weights, posterior_transform=None)[source]

Bases: AnalyticAcquisitionFunction

Scalarized Posterior Mean.

This acquisition function returns a scalarized (across the q-batch) posterior mean given a vector of weights.

Scalarized Posterior Mean.

Parameters:
  • model (Model) – A fitted single-outcome model.

  • weights (Tensor) – A tensor of shape q for scalarization. In order to minimize the scalarized posterior mean, pass -weights.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

forward(X)[source]

Evaluate the scalarized posterior mean on the candidate set X.

Parameters:

X (Tensor) – A (b) x q x d-dim Tensor of (b) t-batches of d-dim design points each.

Returns:

A (b)-dim Tensor of Posterior Mean values at the given design points X.

Return type:

Tensor

class botorch.acquisition.analytic.PosteriorStandardDeviation(model, posterior_transform=None, maximize=True)[source]

Bases: AnalyticAcquisitionFunction

Single-outcome Posterior Standard Deviation.

An acquisition function for pure exploration. Only supports the case of q=1. Requires the model’s posterior to have mean and variance properties. The model must be either single-outcome or combined with a posterior_transform to produce a single-output posterior.

Example

>>> import torch
>>> from botorch.models.gp_regression import SingleTaskGP
>>> from botorch.models.transforms.input import Normalize
>>> from botorch.models.transforms.outcome import Standardize
>>>
>>> # Set up a model
>>> train_X = torch.rand(20, 2, dtype=torch.float64)
>>> train_Y = torch.sin(train_X).sum(dim=1, keepdim=True)
>>> model = SingleTaskGP(
...     train_X, train_Y, outcome_transform=Standardize(m=1),
...     input_transform=Normalize(d=2),
... )
>>> # Now set up the acquisition function
>>> PSTD = PosteriorStandardDeviation(model)
>>> test_X = torch.zeros((1, 2), dtype=torch.float64)
>>> std = PSTD(test_X)
>>> std.item()
0.16341639895667773

Single-outcome Posterior Mean.

Parameters:
  • model (Model) – A fitted single-outcome GP model (must be in batch mode if candidate sets X will be)

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • maximize (bool) – If True, consider the problem a maximization problem. Note that if maximize=False, the posterior standard deviation is negated. As a consequence, optimize_acqf(PosteriorStandardDeviation(gp, maximize=False)) actually returns -1 * minimum of the posterior standard deviation.

forward(X)[source]

Evaluate the posterior standard deviation on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk)-dim tensor of Posterior Mean values at the given design points X.

Return type:

Tensor

Monte-Carlo Acquisition Functions

Batch acquisition functions using the reparameterization trick in combination with (quasi) Monte-Carlo sampling. See [Rezende2014reparam], [Wilson2017reparam] and [Balandat2020botorch].

References

[Rezende2014reparam]

D. J. Rezende, S. Mohamed, and D. Wierstra. Stochastic backpropagation and approximate inference in deep generative models. ICML 2014.

[Wilson2017reparam]

J. T. Wilson, R. Moriconi, F. Hutter, and M. P. Deisenroth. The reparameterization trick for acquisition functions. ArXiv 2017.

class botorch.acquisition.monte_carlo.SampleReductionProtocol(*args, **kwargs)[source]

Bases: Protocol

For static type check of SampleReducingMCAcquisitionFunction’s mc_reduction.

class botorch.acquisition.monte_carlo.SampleReducingMCAcquisitionFunction(model, sampler=None, objective=None, posterior_transform=None, X_pending=None, sample_reduction=<built-in method mean of type object>, q_reduction=<built-in method amax of type object>, constraints=None, eta=0.001, fat=False)[source]

Bases: MCAcquisitionFunction

MC-based batch acquisition function that reduces across samples and implements a general treatment of outcome constraints.

This class’s forward computes the - possibly constrained - acquisition value by (1) computing the unconstrained utility for each MC sample using _sample_forward, (2) weighing the utility values by the constraint indicator per MC sample, and (3) reducing (e.g. averaging) the weighted utility values over the MC dimension.

NOTE: Do NOT override the forward method, unless you have thought about it well.

forward is implemented generically to incorporate constraints in a principled way, and takes care of reducing over the Monte Carlo and batch dimensions via the sample_reduction and q_reduction arguments, which default to torch.mean and torch.max, respectively.

In order to implement a custom SampleReducingMCAcquisitionFunction, we only need to implement the _sample_forward(obj: Tensor) -> Tensor method, which maps objective samples to acquisition utility values without reducing the Monte Carlo and batch (i.e. q) dimensions (see details in the docstring of _sample_forward).

A note on design choices:

The primary purpose of SampleReducingMCAcquisitionFunction`is to support outcome constraints. On the surface, designing a wrapper `ConstrainedMCAcquisitionFunction could be an elegant solution to this end, but it would still require the acquisition functions to implement a _sample_forward method to weigh acquisition utilities at the sample level. Further, qNoisyExpectedImprovement is a special case that is hard to encompass in this pattern, since it requires the computation of the best feasible objective, which requires access to the constraint functions. However, if the constraints are stored in a wrapper class, they will be inaccessible to the forward pass. These problems are circumvented by the design of this class.

Constructor of SampleReducingMCAcquisitionFunction.

Parameters:
  • model (Model) – A fitted model.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated on the fly within the get_posterior_samples method using botorch.sampling.get_sampler. NOTE: For posteriors that do not support base samples, a sampler compatible with intended use case must be provided. See ForkedRNGSampler and StochasticSampler as examples.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective(). NOTE: ConstrainedMCObjective for outcome constraints is deprecated in favor of passing the constraints directly to this constructor.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape, m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • sample_reduction (SampleReductionProtocol) – A callable that takes in a sample_shape x batch_shape Tensor of acquisition utility values, a keyword-argument dim that specifies the sample dimensions to reduce over, and returns a batch_shape-dim Tensor of acquisition values.

  • q_reduction (SampleReductionProtocol) – A callable that takes in a sample_shape x batch_shape x q Tensor of acquisition utility values, a keyword-argument dim that specifies the q dimension to reduce over (i.e. -1), and returns a sample_shape x batch_shape-dim Tensor of acquisition values.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero. NOTE: Constraint-weighting is only compatible with non-negative acquistion utilities, e.g. all improvement-based acquisition functions.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • fat (bool) – Wether to apply a fat-tailed smooth approximation to the feasibility indicator or the canonical sigmoid approximation.

forward(X)[source]

Computes the acquisition value associated with the input X. Weighs the acquisition utility values by smoothed constraint indicators if constraints was passed to the constructor of the class. Applies self.sample_reduction and self.q_reduction to reduce over the Monte Carlo and batch (q) dimensions.

NOTE: Do NOT override the forward method for a custom acquisition function. Instead, implement the _sample_forward method. See the docstring of this class for details.

Parameters:

X (Tensor) – A batch_shape x q x d Tensor of t-batches with q d-dim design points each.

Returns:

A Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X.

Return type:

Tensor

class botorch.acquisition.monte_carlo.qExpectedImprovement(model, best_f, sampler=None, objective=None, posterior_transform=None, X_pending=None, constraints=None, eta=0.001)[source]

Bases: SampleReducingMCAcquisitionFunction

MC-based batch Expected Improvement.

This computes qEI by (1) sampling the joint posterior over q points (2) evaluating the improvement over the current best for each sample (3) maximizing over q (4) averaging over the samples

qEI(X) = E(max(max Y - best_f, 0)), Y ~ f(X), where X = (x_1,…,x_q)

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> best_f = train_Y.max()[0]
>>> sampler = SobolQMCNormalSampler(1024)
>>> qEI = qExpectedImprovement(model, best_f, sampler)
>>> qei = qEI(test_X)

NOTE: It is strongly recommended to use qLogExpectedImprovement instead of regular qEI, as it can lead to substantially improved BO performance through improved numerics. See https://arxiv.org/abs/2310.20708 for details.

q-Expected Improvement.

Parameters:
  • model (Model) – A fitted model.

  • best_f (Union[float, Tensor]) – The best objective value observed so far (assumed noiseless). Can be a scalar, or a batch_shape-dim tensor. In case of a batched model, the tensor can specify different values for each element of the batch.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective(). NOTE: ConstrainedMCObjective for outcome constraints is deprecated in favor of passing the constraints directly to this constructor.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

class botorch.acquisition.monte_carlo.qNoisyExpectedImprovement(model, X_baseline, sampler=None, objective=None, posterior_transform=None, X_pending=None, prune_baseline=True, cache_root=True, constraints=None, eta=0.001, marginalize_dim=None)[source]

Bases: SampleReducingMCAcquisitionFunction, CachedCholeskyMCSamplerMixin

MC-based batch Noisy Expected Improvement.

This function does not assume a best_f is known (which would require noiseless observations). Instead, it uses samples from the joint posterior over the q test points and previously observed points. The improvement over previously observed points is computed for each sample and averaged.

qNEI(X) = E(max(max Y - max Y_baseline, 0)), where (Y, Y_baseline) ~ f((X, X_baseline)), X = (x_1,…,x_q)

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> sampler = SobolQMCNormalSampler(1024)
>>> qNEI = qNoisyExpectedImprovement(model, train_X, sampler)
>>> qnei = qNEI(test_X)

NOTE: It is strongly recommended to use qLogNoisyExpectedImprovement instead of regular qNEI, as it can lead to substantially improved BO performance through improved numerics. See https://arxiv.org/abs/2310.20708 for details.

q-Noisy Expected Improvement.

Parameters:
  • model (Model) – A fitted model.

  • X_baseline (Tensor) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective(). NOTE: ConstrainedMCObjective for outcome constraints is deprecated in favor of passing the constraints directly to this constructor.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • prune_baseline (bool) – If True, remove points in X_baseline that are highly unlikely to be the best point. This can significantly improve performance and is generally recommended. In order to customize pruning parameters, instead manually call botorch.acquisition.utils.prune_inferior_points on X_baseline before instantiating the acquisition function.

  • cache_root (bool) – A boolean indicating whether to cache the root decomposition over X_baseline and use low-rank updates.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • marginalize_dim (Optional[int]) – The dimension to marginalize over.

TODO: similar to qNEHVI, when we are using sequential greedy candidate selection, we could incorporate pending points X_baseline and compute the incremental qNEI from the new point. This would greatly increase efficiency for large batches.

compute_best_f(obj)[source]

Computes the best (feasible) noisy objective value.

Parameters:

obj (Tensor) – sample_shape x batch_shape x q-dim Tensor of objectives in forward.

Returns:

A sample_shape x batch_shape-dim Tensor of best feasible objectives.

Return type:

Tensor

class botorch.acquisition.monte_carlo.qProbabilityOfImprovement(model, best_f, sampler=None, objective=None, posterior_transform=None, X_pending=None, tau=0.001, constraints=None, eta=0.001)[source]

Bases: SampleReducingMCAcquisitionFunction

MC-based batch Probability of Improvement.

Estimates the probability of improvement over the current best observed value by sampling from the joint posterior distribution of the q-batch. MC-based estimates of a probability involves taking expectation of an indicator function; to support auto-differentiation, the indicator is replaced with a sigmoid function with temperature parameter tau.

qPI(X) = P(max Y >= best_f), Y ~ f(X), X = (x_1,…,x_q)

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> best_f = train_Y.max()[0]
>>> sampler = SobolQMCNormalSampler(1024)
>>> qPI = qProbabilityOfImprovement(model, best_f, sampler)
>>> qpi = qPI(test_X)

q-Probability of Improvement.

Parameters:
  • model (Model) – A fitted model.

  • best_f (Union[float, Tensor]) – The best objective value observed so far (assumed noiseless). Can be a batch_shape-shaped tensor, which in case of a batched model specifies potentially different values for each element of the batch.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective(). NOTE: ConstrainedMCObjective for outcome constraints is deprecated in favor of passing the constraints directly to this constructor.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • tau (float) – The temperature parameter used in the sigmoid approximation of the step function. Smaller values yield more accurate approximations of the function, but result in gradients estimates with higher variance.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map posterior samples to a scalar. The associated constraint is considered satisfied if this scalar is less than zero.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

class botorch.acquisition.monte_carlo.qSimpleRegret(model, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Bases: SampleReducingMCAcquisitionFunction

MC-based batch Simple Regret.

Samples from the joint posterior over the q-batch and computes the simple regret.

qSR(X) = E(max Y), Y ~ f(X), X = (x_1,…,x_q)

Constraints should be provided as a ConstrainedMCObjective. Passing constraints as an argument is not supported. This is because SampleReducingMCAcquisitionFunction computes the acquisition values on the sample level and then weights the sample-level acquisition values by a soft feasibility indicator. Hence, it expects non-log acquisition function values to be non-negative. qSimpleRegret acquisition values can be negative, so we instead use a ConstrainedMCObjective which applies constraints to the objectives (e.g. before computing the acquisition function) and shifts negative objective values using by an infeasible cost to ensure non-negativity (before applying constraints and shifting them back).

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> sampler = SobolQMCNormalSampler(1024)
>>> qSR = qSimpleRegret(model, sampler)
>>> qsr = qSR(test_X)

q-Simple Regret.

Parameters:
  • model (Model) – A fitted model.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

class botorch.acquisition.monte_carlo.qUpperConfidenceBound(model, beta, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Bases: SampleReducingMCAcquisitionFunction

MC-based batch Upper Confidence Bound.

Uses a reparameterization to extend UCB to qUCB for q > 1 (See Appendix A of [Wilson2017reparam].)

qUCB = E(max(mu + |Y_tilde - mu|)), where Y_tilde ~ N(mu, beta pi/2 Sigma) and f(X) has distribution N(mu, Sigma).

Constraints should be provided as a ConstrainedMCObjective. Passing constraints as an argument is not supported. This is because SampleReducingMCAcquisitionFunction computes the acquisition values on the sample level and then weights the sample-level acquisition values by a soft feasibility indicator. Hence, it expects non-log acquisition function values to be non-negative. qSimpleRegret acquisition values can be negative, so we instead use a ConstrainedMCObjective which applies constraints to the objectives (e.g. before computing the acquisition function) and shifts negative objective values using by an infeasible cost to ensure non-negativity (before applying constraints and shifting them back).

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> sampler = SobolQMCNormalSampler(1024)
>>> qUCB = qUpperConfidenceBound(model, 0.1, sampler)
>>> qucb = qUCB(test_X)

q-Upper Confidence Bound.

Parameters:
  • model (Model) – A fitted model.

  • beta (float) – Controls tradeoff between mean and standard deviation in UCB.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

class botorch.acquisition.monte_carlo.qLowerConfidenceBound(model, beta, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Bases: qUpperConfidenceBound

MC-based batched lower confidence bound.

This acquisition function is useful for confident/risk-averse decision making. This acquisition function is intended to be maximized as with qUpperConfidenceBound, but the qLowerConfidenceBound will be pessimistic in the face of uncertainty and lead to conservative candidates.

q-Upper Confidence Bound.

Parameters:
  • model (Model) – A fitted model.

  • beta (float) – Controls tradeoff between mean and standard deviation in UCB.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

Monte-Carlo variants of the LogEI family of improvements-based acquisition functions, see [Ament2023logei] for details.

References

[Ament2023logei] (1,2,3,4,5,6,7,8,9)

S. Ament, S. Daulton, D. Eriksson, M. Balandat, and E. Bakshy. Unexpected Improvements to Expected Improvement for Bayesian Optimization. Advances in Neural Information Processing Systems 36, 2023.

class botorch.acquisition.logei.LogImprovementMCAcquisitionFunction(model, sampler=None, objective=None, posterior_transform=None, X_pending=None, constraints=None, eta=0.001, fat=True, tau_max=0.01)[source]

Bases: SampleReducingMCAcquisitionFunction

Abstract base class for Monte-Carlo-based batch LogEI acquisition functions.

Parameters:
  • model (Model) – A fitted model.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler. NOTE: For posteriors that do not support base samples, a sampler compatible with intended use case must be provided. See ForkedRNGSampler and StochasticSampler as examples.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape, m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are satisfied if constraint(samples) < 0.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. See the docs of compute_(log_)constraint_indicator for more details on this parameter.

  • fat (bool) – Toggles the logarithmic / linear asymptotic behavior of the smooth approximation to the ReLU.

  • tau_max (float) – Temperature parameter controlling the sharpness of the approximation to the max operator over the q candidate points.

class botorch.acquisition.logei.qLogExpectedImprovement(model, best_f, sampler=None, objective=None, posterior_transform=None, X_pending=None, constraints=None, eta=0.001, fat=True, tau_max=0.01, tau_relu=1e-06)[source]

Bases: LogImprovementMCAcquisitionFunction

MC-based batch Log Expected Improvement.

This computes qLogEI by (1) sampling the joint posterior over q points, (2) evaluating the smoothed log improvement over the current best for each sample, (3) smoothly maximizing over q, and (4) averaging over the samples in log space.

See [Ament2023logei] for details. Formally,

qLogEI(X) ~ log(qEI(X)) = log(E(max(max Y - best_f, 0))).

where Y ~ f(X), and X = (x_1,…,x_q), .

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> best_f = train_Y.max()[0]
>>> sampler = SobolQMCNormalSampler(1024)
>>> qLogEI = qLogExpectedImprovement(model, best_f, sampler)
>>> qei = qLogEI(test_X)

q-Log Expected Improvement.

Parameters:
  • model (Model) – A fitted model.

  • best_f (Union[float, Tensor]) – The best objective value observed so far (assumed noiseless). Can be a scalar, or a batch_shape-dim tensor. In case of a batched model, the tensor can specify different values for each element of the batch.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are satisfied if constraint(samples) < 0.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. See the docs of compute_(log_)smoothed_constraint_indicator for details.

  • fat (bool) – Toggles the logarithmic / linear asymptotic behavior of the smooth approximation to the ReLU.

  • tau_max (float) – Temperature parameter controlling the sharpness of the smooth approximations to max.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the smooth approximations to ReLU.

class botorch.acquisition.logei.qLogNoisyExpectedImprovement(model, X_baseline, sampler=None, objective=None, posterior_transform=None, X_pending=None, constraints=None, eta=0.001, fat=True, prune_baseline=False, cache_root=True, tau_max=0.01, tau_relu=1e-06, marginalize_dim=None)[source]

Bases: LogImprovementMCAcquisitionFunction, CachedCholeskyMCSamplerMixin

MC-based batch Log Noisy Expected Improvement.

This function does not assume a best_f is known (which would require noiseless observations). Instead, it uses samples from the joint posterior over the q test points and previously observed points. A smooth approximation to the canonical improvement over previously observed points is computed for each sample and the logarithm of the average is returned.

See [Ament2023logei] for details. Formally,

qLogNEI(X) ~ log(qNEI(X)) = Log E(max(max Y - max Y_baseline, 0)),

where (Y, Y_baseline) ~ f((X, X_baseline)), X = (x_1,…,x_q).

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> sampler = SobolQMCNormalSampler(1024)
>>> qLogNEI = qLogNoisyExpectedImprovement(model, train_X, sampler)
>>> acqval = qLogNEI(test_X)

q-Noisy Expected Improvement.

Parameters:
  • model (Model) – A fitted model.

  • X_baseline (Tensor) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. See MCAcquisitionFunction more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are satisfied if constraint(samples) < 0.

  • eta (Union[Tensor, float]) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. See the docs of compute_(log_)smoothed_constraint_indicator for details.

  • fat (bool) – Toggles the logarithmic / linear asymptotic behavior of the smooth approximation to the ReLU.

  • prune_baseline (bool) – If True, remove points in X_baseline that are highly unlikely to be the best point. This can significantly improve performance and is generally recommended. In order to customize pruning parameters, instead manually call botorch.acquisition.utils.prune_inferior_points on X_baseline before instantiating the acquisition function.

  • cache_root (bool) – A boolean indicating whether to cache the root decomposition over X_baseline and use low-rank updates.

  • tau_max (float) – Temperature parameter controlling the sharpness of the smooth approximations to max.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the smooth approximations to ReLU.

  • marginalize_dim (Optional[int]) – The dimension to marginalize over.

TODO: similar to qNEHVI, when we are using sequential greedy candidate selection, we could incorporate pending points X_baseline and compute the incremental q(Log)NEI from the new point. This would greatly increase efficiency for large batches.

compute_best_f(obj)[source]

Computes the best (feasible) noisy objective value.

Parameters:

obj (Tensor) – sample_shape x batch_shape x q-dim Tensor of objectives in forward.

Returns:

A sample_shape x batch_shape-dim Tensor of best feasible objectives.

Return type:

Tensor

botorch.acquisition.logei.check_tau(tau, name)[source]

Checks the validity of the tau arguments of the functions below, and returns tau if it is valid.

Parameters:
  • tau (FloatOrTensor)

  • name (str)

Return type:

FloatOrTensor

Multi-Objective Analytic Acquisition Functions

Analytic Acquisition Functions for Multi-objective Bayesian optimization.

References

[Yang2019] (1,2,3,4,5)

Yang, K., Emmerich, M., Deutz, A. et al. Efficient computation of expected hypervolume improvement using box decomposition algorithms. J Glob Optim 75, 3–34 (2019)

class botorch.acquisition.multi_objective.analytic.ExpectedHypervolumeImprovement(model, ref_point, partitioning, posterior_transform=None)[source]

Bases: MultiObjectiveAnalyticAcquisitionFunction

Expected Hypervolume Improvement supporting m>=2 outcomes.

This implements the computes EHVI using the algorithm from [Yang2019], but additionally computes gradients via auto-differentiation as proposed by [Daulton2020qehvi].

Note: this is currently inefficient in two ways due to the binary partitioning algorithm that we use for the box decomposition:

  • We have more boxes in our decomposition

  • If we used a box decomposition that used inf as the upper bound for

    the last dimension in all hypercells, then we could reduce the number of terms we need to compute from 2^m to 2^(m-1). [Yang2019] do this by using DKLV17 and LKF17 for the box decomposition.

TODO: Use DKLV17 and LKF17 for the box decomposition as in [Yang2019] for greater efficiency.

TODO: Add support for outcome constraints.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0]
>>> EHVI = ExpectedHypervolumeImprovement(model, ref_point, partitioning)
>>> ehvi = EHVI(test_X)
Parameters:
  • model (Model) – A fitted model.

  • ref_point (list[float]) – A list with m elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. This is a reference point for the outcome values (i.e., after applying posterior_transform if provided).

  • partitioning (NondominatedPartitioning) – A NondominatedPartitioning module that provides the non- dominated front and a partitioning of the non-dominated space in hyper- rectangles.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform.

psi(lower, upper, mu, sigma)[source]

Compute Psi function.

For each cell i and outcome k:

Psi(lower_{i,k}, upper_{i,k}, mu_k, sigma_k) = ( sigma_k * PDF((upper_{i,k} - mu_k) / sigma_k) + ( mu_k - lower_{i,k} ) * (1 - CDF(upper_{i,k} - mu_k) / sigma_k) )

See Equation 19 in [Yang2019] for more details.

Parameters:
  • lower (Tensor) – A num_cells x m-dim tensor of lower cell bounds

  • upper (Tensor) – A num_cells x m-dim tensor of upper cell bounds

  • mu (Tensor) – A batch_shape x 1 x m-dim tensor of means

  • sigma (Tensor) – A batch_shape x 1 x m-dim tensor of standard deviations (clamped).

Returns:

A batch_shape x num_cells x m-dim tensor of values.

Return type:

Tensor

nu(lower, upper, mu, sigma)[source]

Compute Nu function.

For each cell i and outcome k:

nu(lower_{i,k}, upper_{i,k}, mu_k, sigma_k) = ( upper_{i,k} - lower_{i,k} ) * (1 - CDF((upper_{i,k} - mu_k) / sigma_k))

See Equation 25 in [Yang2019] for more details.

Parameters:
  • lower (Tensor) – A num_cells x m-dim tensor of lower cell bounds

  • upper (Tensor) – A num_cells x m-dim tensor of upper cell bounds

  • mu (Tensor) – A batch_shape x 1 x m-dim tensor of means

  • sigma (Tensor) – A batch_shape x 1 x m-dim tensor of standard deviations (clamped).

Returns:

A batch_shape x num_cells x m-dim tensor of values.

Return type:

Tensor

forward(X)[source]

Takes in a batch_shape x 1 x d X Tensor of t-batches with 1 d-dim design point each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X.

Parameters:

X (Tensor)

Return type:

Tensor

Multi-Objective Hypervolume Knowledge Gradient Acquisition Functions

The hypervolume knowledge gradient acquisition function (HVKG).

References:

[Daulton2023hvkg] (1,2,3,4)

S. Daulton, M. Balandat, E. Bakshy. Hypervolume Knowledge Gradient: A Lookahead Approach for Multi-Objective Bayesian Optimization with Partial Information. Proceedings of the 40th International Conference on Machine Learning, 2023.

class botorch.acquisition.multi_objective.hypervolume_knowledge_gradient.qHypervolumeKnowledgeGradient(model, ref_point, num_fantasies=8, num_pareto=10, sampler=None, objective=None, inner_sampler=None, X_evaluation_mask=None, X_pending=None, X_pending_evaluation_mask=None, current_value=None, use_posterior_mean=True, cost_aware_utility=None)[source]

Bases: DecoupledAcquisitionFunction, MultiObjectiveMCAcquisitionFunction, OneShotAcquisitionFunction

Batch Hypervolume Knowledge Gradient using one-shot optimization.

The hypervolume knowledge gradient seeks to maximize the difference in hypervolume of the hypervolume-maximizing set of a fixed size after conditioning the unknown observation(s) that would be recevied if X where evalauted. See [Daulton2023hvkg] for details.

This computes the batch Hypervolume Knowledge Gradient using fantasies for the outer expectation and MC-sampling for the inner expectation.

In addition to the design variables, the input X also includes variables for the optimal designs for each of the fantasy models (Note this is N x N_pareto optimal designs). For a fixed number of fantasies, all points in X can be optimized in a “one-shot” fashion.

q-Hypervolume Knowledge Gradient.

Parameters:
  • model (Model) – A fitted model. Must support fantasizing.

  • ref_point (Tensor) – A m-dim tensor containing the reference point.

  • num_fantasies (int) – The number of fantasy points to use. More fantasy points result in a better approximation, at the expense of memory and wall time. Unused if sampler is specified.

  • num_pareto (int) – The number of pareto optimal designs to consider.

  • sampler (ListSampler | None) – The sampler used to sample fantasy observations. Optional if num_fantasies is specified. The optimization performance does not seem particularly sensitive to the number of fantasies. As the number of fantasies increases, the estimation of the expectation over fantasies becomes more accurate, but the one- shot optimization problem gets harder as there are more “fantasy” designs that need to be optimized.

  • objective (MCMultiOutputObjective | None) – The objective under which the samples are evaluated. If None, then the analytic posterior mean is used. Otherwise, the objective is MC-evaluated (using inner_sampler).

  • inner_sampler (MCSampler | None) – The sampler used for inner sampling. Ignored if the objective is None.

  • X_evaluation_mask (list[Tensor] | None) – A q x m-dim tensor of booleans indicating which objective(s) each of the q points should be evaluated on.

  • X_pending (Tensor | None) – A n’ x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • X_pending_evaluation_mask (Tensor | None) – A n’ x m-dim tensor of booleans indicating which objective(s) each of the n’ pending points are being evaluated on.

  • current_value (Tensor | None) – The current value, i.e. the expected best objective given the observed points D. If omitted, forward will not return the actual KG value, but the expected best objective given the data set D u X. If pending points are used, this should be the current value under the fantasy model conditioned on the pending points so that the incremental KG value from the new candidates (not pending points) is used.

  • use_posterior_mean (bool) – If true, optimize the hypervolume of the posterior mean, otherwise optimize the expected hypervolume. See [Daulton2023hvkg] for details.

  • cost_aware_utility (CostAwareUtility | None) – A CostAwareUtility specifying the cost function for evaluating the X on the objectives indicated by evaluation_mask.

property cost_sampler
forward(X)[source]

Evaluate qKnowledgeGradient on the candidate set X.

Parameters:

X (Tensor) –

A b x (q + num_fantasies) x d Tensor with b t-batches of q + num_fantasies design points each. We split this X tensor into two parts in the q dimension (dim=-2). The first q are the q-batch of design points and the last num_fantasies are the current solutions of the inner optimization problem.

X_fantasies = X[…, -num_fantasies:, :] X_fantasies.shape = b x num_fantasies x d

X_actual = X[…, :-num_fantasies, :] X_actual.shape = b x q x d

Returns:

A Tensor of shape b. For t-batch b, the q-KG value of the design

X_actual[b] is averaged across the fantasy models, where X_fantasies[b, i] is chosen as the final selection for the i-th fantasy model. NOTE: If current_value is not provided, then this is not the true KG value of X_actual[b], and X_fantasies[b, : ] must be maximized at fixed X_actual[b].

Return type:

Tensor

get_augmented_q_batch_size(q)[source]

Get augmented q batch size for one-shot optimization.

Parameters:

q (int) – The number of candidates to consider jointly.

Returns:

The augmented size for one-shot optimization (including variables parameterizing the fantasy solutions).

Return type:

int

extract_candidates(X_full)[source]

We only return X as the set of candidates post-optimization.

Parameters:

X_full (Tensor) – A b x (q + num_fantasies) x d-dim Tensor with b t-batches of q + num_fantasies design points each.

Returns:

A b x q x d-dim Tensor with b t-batches of q design points each.

Return type:

Tensor

class botorch.acquisition.multi_objective.hypervolume_knowledge_gradient.qMultiFidelityHypervolumeKnowledgeGradient(model, ref_point, target_fidelities, num_fantasies=8, num_pareto=10, sampler=None, objective=None, inner_sampler=None, X_pending=None, X_evaluation_mask=None, X_pending_evaluation_mask=None, current_value=None, cost_aware_utility=None, project=<function qMultiFidelityHypervolumeKnowledgeGradient.<lambda>>, valfunc_cls=None, valfunc_argfac=None, use_posterior_mean=True, **kwargs)[source]

Bases: qHypervolumeKnowledgeGradient

Batch Hypervolume Knowledge Gradient for multi-fidelity optimization.

See [Daulton2023hvkg] for details.

A version of qHypervolumeKnowledgeGradient that supports multi-fidelity optimization via a CostAwareUtility and the project and expand operators. If none of these are set, this acquisition function reduces to qHypervolumeKnowledgeGradient. Through valfunc_cls and valfunc_argfac, this can be changed into a custom multi-fidelity acquisition function.

Multi-Fidelity q-Knowledge Gradient (one-shot optimization).

Parameters:
  • model (Model) – A fitted model. Must support fantasizing.

  • ref_point (Tensor) – A m-dim tensor containing the reference point.

  • num_fantasies (int) – The number of fantasy points to use. More fantasy points result in a better approximation, at the expense of memory and wall time. Unused if sampler is specified.

  • num_pareto (int) – The number of pareto optimal designs to consider.

  • sampler (MCSampler | None) – The sampler used to sample fantasy observations. Optional if num_fantasies is specified.

  • objective (MCMultiOutputObjective | None) – The objective under which the samples are evaluated. If None, then the analytic posterior mean is used. Otherwise, the objective is MC-evaluated (using inner_sampler).

  • inner_sampler (MCSampler | None) – The sampler used for inner sampling. Ignored if the objective is None.

  • X_evaluation_mask (Tensor | None) – A q x m-dim tensor of booleans indicating which objective(s) each of the q points should be evaluated on.

  • X_pending (Tensor | None) – A n’ x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • X_pending_evaluation_mask (Tensor | None) – A n’ x m-dim tensor of booleans indicating which objective(s) each of the n’ pending points are being evaluated on.

  • current_value (Tensor | None) – The current value, i.e. the expected best objective given the observed points D. If omitted, forward will not return the actual KG value, but the expected best objective given the data set D u X. If pending points are used, this should be the current value under the fantasy model conditioned on the pending points so that the incremental KG value from the new candidates (not pending points) is used.

  • use_posterior_mean (bool) – A boolean indicating whether to use the to optimize the hypervolume of the posterior mean or whether to optimize the expected hypervolume. See [Daulton2023hvkg] for details.

  • cost_aware_utility (CostAwareUtility | None) – A CostAwareUtility specifying the cost function for evaluating the X on the objectives indicated by evaluation_mask.

  • project (Callable[[Tensor], Tensor]) – A callable mapping a batch_shape x q x d tensor of design points to a tensor with shape batch_shape x q_term x d projected to the desired target set (e.g. the target fidelities in case of multi-fidelity optimization). For the basic case, q_term = q.

  • valfunc_cls (type[AcquisitionFunction] | None) – An acquisition function class to be used as the terminal value function.

  • valfunc_argfac (Callable[[Model], dict[str, Any]] | None) – An argument factory, i.e. callable that maps a Model to a dictionary of kwargs for the terminal value function (e.g. best_f for ExpectedImprovement).

  • target_fidelities (dict[int, float])

  • kwargs (Any)

forward(X)[source]

Evaluate qMultiFidelityKnowledgeGradient on the candidate set X.

Parameters:

X (Tensor) –

A b x (q + num_fantasies) x d Tensor with b t-batches of q + num_fantasies design points each. We split this X tensor into two parts in the q dimension (dim=-2). The first q are the q-batch of design points and the last num_fantasies are the current solutions of the inner optimization problem.

X_fantasies = X[…, -num_fantasies:, :] X_fantasies.shape = b x num_fantasies x d

X_actual = X[…, :-num_fantasies, :] X_actual.shape = b x q x d

In addition, X may be augmented with fidelity parameteres as part of thee d-dimension. Projecting fidelities to the target fidelity is handled by project.

Returns:

A Tensor of shape b. For t-batch b, the q-KG value of the design

X_actual[b] is averaged across the fantasy models, where X_fantasies[b, i] is chosen as the final selection for the i-th fantasy model. NOTE: If current_value is not provided, then this is not the true KG value of X_actual[b], and X_fantasies[b, : ] must be maximized at fixed X_actual[b].

Return type:

Tensor

Multi-Objective Monte-Carlo Acquisition Functions

Monte-Carlo Acquisition Functions for Multi-objective Bayesian optimization. In particular, this module contains implementations of 1) qEHVI [Daulton2020qehvi], and 2) qNEHVI [Daulton2021nehvi].

References

[Daulton2020qehvi] (1,2,3,4,5)

S. Daulton, M. Balandat, and E. Bakshy. Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective Bayesian Optimization. Advances in Neural Information Processing Systems 33, 2020.

[Daulton2021nehvi] (1,2,3)

S. Daulton, M. Balandat, and E. Bakshy. Parallel Bayesian Optimization of Multiple Noisy Objectives with Expected Hypervolume Improvement. Advances in Neural Information Processing Systems 34, 2021.

class botorch.acquisition.multi_objective.monte_carlo.qExpectedHypervolumeImprovement(model, ref_point, partitioning, sampler=None, objective=None, constraints=None, X_pending=None, eta=0.001, fat=False)[source]

Bases: MultiObjectiveMCAcquisitionFunction, SubsetIndexCachingMixin

q-Expected Hypervolume Improvement supporting m>=2 outcomes.

See [Daulton2020qehvi] for details.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0]
>>> qEHVI = qExpectedHypervolumeImprovement(model, ref_point, partitioning)
>>> qehvi = qEHVI(test_X)
Parameters:
  • model (Model) – A fitted model.

  • ref_point (Union[list[float], Tensor]) – A list or tensor with m elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. This is a reference point for the objective values (i.e. after applying`objective` to the samples).

  • partitioning (NondominatedPartitioning) – A NondominatedPartitioning module that provides the non- dominated front and a partitioning of the non-dominated space in hyper- rectangles. If constraints are present, this partitioning must only include feasible points.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMCMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. The acquisition function will compute expected feasible hypervolume.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

  • fat (bool) – A Boolean flag indicating whether to use the heavy-tailed approximation of the constraint indicator.

forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

class botorch.acquisition.multi_objective.monte_carlo.qNoisyExpectedHypervolumeImprovement(model, ref_point, X_baseline, sampler=None, objective=None, constraints=None, X_pending=None, eta=0.001, fat=False, prune_baseline=False, alpha=0.0, cache_pending=True, max_iep=0, incremental_nehvi=True, cache_root=True, marginalize_dim=None)[source]

Bases: NoisyExpectedHypervolumeMixin, qExpectedHypervolumeImprovement

q-Noisy Expected Hypervolume Improvement supporting m>=2 outcomes.

See [Daulton2021nehvi] for details.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0]
>>> qNEHVI = qNoisyExpectedHypervolumeImprovement(model, ref_point, train_X)
>>> qnehvi = qNEHVI(test_X)
Parameters:
  • model (Model) – A fitted model.

  • ref_point (Union[list[float], Tensor]) – A list or tensor with m elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. This is a reference point for the objective values (i.e. after applying objective to the samples).

  • X_baseline (Tensor) – A r x d-dim Tensor of r design points that have already been observed. These points are considered as potential approximate pareto-optimal design points.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler. Note: a pareto front is created for each mc sample, which can be computationally intensive for m > 2.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMCMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. The acquisition function will compute expected feasible hypervolume.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation, but have not yet been evaluated.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • fat (bool) – A Boolean flag indicating whether to use the heavy-tailed approximation of the constraint indicator.

  • prune_baseline (bool) – If True, remove points in X_baseline that are highly unlikely to be the pareto optimal and better than the reference point. This can significantly improve computation time and is generally recommended. In order to customize pruning parameters, instead manually call prune_inferior_points_multi_objective on X_baseline before instantiating the acquisition function.

  • alpha (float) – The hyperparameter controlling the approximate non-dominated partitioning. The default value of 0.0 means an exact partitioning is used. As the number of objectives m increases, consider increasing this parameter in order to limit computational complexity.

  • cache_pending (bool) – A boolean indicating whether to use cached box decompositions (CBD) for handling pending points. This is generally recommended.

  • max_iep (int) – The maximum number of pending points before the box decompositions will be recomputed.

  • incremental_nehvi (bool) – A boolean indicating whether to compute the incremental NEHVI from the i`th point where `i=1, …, q under sequential greedy optimization, or the full qNEHVI over q points.

  • cache_root (bool) – A boolean indicating whether to cache the root decomposition over X_baseline and use low-rank updates.

  • marginalize_dim (Optional[int]) – A batch dimension that should be marginalized. For example, this is useful when using a batched fully Bayesian model.

forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

Multi-objective variants of the LogEI family of acquisition functions, see [Ament2023logei] for details.

class botorch.acquisition.multi_objective.logei.qLogExpectedHypervolumeImprovement(model, ref_point, partitioning, sampler=None, objective=None, constraints=None, X_pending=None, eta=0.01, fat=True, tau_relu=1e-06, tau_max=0.01)[source]

Bases: MultiObjectiveMCAcquisitionFunction, SubsetIndexCachingMixin

Parallel Log Expected Hypervolume Improvement supporting m>=2 outcomes.

See [Ament2023logei] for details and the methodology behind the LogEI family of acquisition function. Line-by-line differences to the original differentiable expected hypervolume formulation of [Daulton2020qehvi] are described via inline comments in forward.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0]
>>> acq = qLogExpectedHypervolumeImprovement(model, ref_point, partitioning)
>>> value = acq(test_X)
Parameters:
  • model (Model) – A fitted model.

  • ref_point (Union[list[float], Tensor]) – A list or tensor with m elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. This is a reference point for the objective values (i.e. after applying`objective` to the samples).

  • partitioning (NondominatedPartitioning) – A NondominatedPartitioning module that provides the non- dominated front and a partitioning of the non-dominated space in hyper- rectangles. If constraints are present, this partitioning must only include feasible points.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. The acquisition function will compute expected feasible hypervolume.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

  • fat (bool) – Toggles the logarithmic / linear asymptotic behavior of the smooth approximation to the ReLU and the maximum.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the approximation to the ReLU over the q candidate points. For further details, see the comments above the definition of TAU_RELU.

  • tau_max (float) – Temperature parameter controlling the sharpness of the approximation to the max operator over the q candidate points. For further details, see the comments above the definition of TAU_MAX.

forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

class botorch.acquisition.multi_objective.logei.qLogNoisyExpectedHypervolumeImprovement(model, ref_point, X_baseline, sampler=None, objective=None, constraints=None, X_pending=None, eta=0.001, prune_baseline=False, alpha=0.0, cache_pending=True, max_iep=0, incremental_nehvi=True, cache_root=True, tau_relu=1e-06, tau_max=0.001, fat=True, marginalize_dim=None)[source]

Bases: NoisyExpectedHypervolumeMixin, qLogExpectedHypervolumeImprovement

q-Log Noisy Expected Hypervolume Improvement supporting m>=2 outcomes.

Based on the differentiable hypervolume formulation of [Daulton2021nehvi].

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0]
>>> qNEHVI = qNoisyExpectedHypervolumeImprovement(model, ref_point, train_X)
>>> qnehvi = qNEHVI(test_X)
Parameters:
  • model (Model) – A fitted model.

  • ref_point (Union[list[float], Tensor]) – A list or tensor with m elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. This is a reference point for the objective values (i.e. after applying objective to the samples).

  • X_baseline (Tensor) – A r x d-dim Tensor of r design points that have already been observed. These points are considered as potential approximate pareto-optimal design points.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler. Note: a pareto front is created for each mc sample, which can be computationally intensive for m > 2.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. The acquisition function will compute expected feasible hypervolume.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation, but have not yet been evaluated.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

  • prune_baseline (bool) – If True, remove points in X_baseline that are highly unlikely to be the pareto optimal and better than the reference point. This can significantly improve computation time and is generally recommended. In order to customize pruning parameters, instead manually call prune_inferior_points_multi_objective on X_baseline before instantiating the acquisition function.

  • alpha (float) – The hyperparameter controlling the approximate non-dominated partitioning. The default value of 0.0 means an exact partitioning is used. As the number of objectives m increases, consider increasing this parameter in order to limit computational complexity.

  • cache_pending (bool) – A boolean indicating whether to use cached box decompositions (CBD) for handling pending points. This is generally recommended.

  • max_iep (int) – The maximum number of pending points before the box decompositions will be recomputed.

  • incremental_nehvi (bool) – A boolean indicating whether to compute the incremental NEHVI from the i`th point where `i=1, …, q under sequential greedy optimization, or the full qNEHVI over q points.

  • cache_root (bool) – A boolean indicating whether to cache the root decomposition over X_baseline and use low-rank updates.

  • marginalize_dim (Optional[int]) – A batch dimension that should be marginalized.

  • tau_relu (float)

  • tau_max (float)

  • fat (bool)

forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

Multi-Objective Multi-Fidelity Acquisition Functions

Multi-Fidelity Acquisition Functions for Multi-objective Bayesian optimization.

References

[Irshad2021MOMF]

F. Irshad, S. Karsch, and A. Döpp. Expected hypervolume improvement for simultaneous multi-objective and multi-fidelity optimization. arXiv preprint arXiv:2112.13901, 2021.

class botorch.acquisition.multi_objective.multi_fidelity.MOMF(model, ref_point, partitioning, sampler=None, objective=None, constraints=None, eta=0.001, X_pending=None, cost_call=None)[source]

Bases: qExpectedHypervolumeImprovement

MOMF acquisition function supporting m>=2 outcomes. The model needs to have train_obj that has a fidelity objective appended to its end. In the following example we consider a 2-D output space but the ref_point is 3D because of fidelity objective.

See [Irshad2021MOMF] for details.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> ref_point = [0.0, 0.0, 0.0]
>>> cost_func = lambda X: 5 + X[..., -1]
>>> momf = MOMF(model, ref_point, partitioning, cost_func)
>>> momf_val = momf(test_X)
Parameters:
  • model (Model) – A fitted model. There are two default assumptions in the training data. train_X should have fidelity parameter s as the last dimension of the input and train_Y contains a trust objective as its last dimension.

  • ref_point (Union[list[float], Tensor]) – A list or tensor with m+1 elements representing the reference point (in the outcome space) w.r.t. to which compute the hypervolume. The ‘+1’ takes care of the trust objective appended to train_Y. This is a reference point for the objective values (i.e. after applying`objective` to the samples).

  • partitioning (NondominatedPartitioning) – A NondominatedPartitioning module that provides the non- dominated front and a partitioning of the non-dominated space in hyper- rectangles. If constraints are present, this partitioning must only include feasible points.

  • sampler (Optional[MCSampler]) – The sampler used to draw base samples. If not given, a sampler is generated using get_sampler.

  • objective (Optional[MCMultiOutputObjective]) – The MCMultiOutputObjective under which the samples are evaluated. Defaults to IdentityMCMultiOutputObjective().

  • constraints (Optional[list[Callable[[Tensor], Tensor]]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. The acquisition function will compute expected feasible hypervolume.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • cost_call (Optional[Callable[[Tensor], Tensor]]) – A callable cost function mapping a Tensor of dimension batch_shape x q x d to a cost Tensor of dimension batch_shape x q x m. Defaults to an AffineCostModel with C(s) = 1 + s.

  • eta (Union[Tensor, float]) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

forward(X)[source]

Takes in a batch_shape x q x d X Tensor of t-batches with q d-dim design points each, and returns a Tensor with shape batch_shape’, where batch_shape’ is the broadcasted batch shape of model and input X. Should utilize the result of set_X_pending as needed to account for pending function evaluations.

Parameters:

X (Tensor)

Return type:

Tensor

ParEGO: Multi-Objective Acquisition Function with Chebyshev Scalarization

class botorch.acquisition.multi_objective.parego.qLogNParEGO(model, X_baseline, scalarization_weights=None, sampler=None, objective=None, constraints=None, X_pending=None, eta=0.001, fat=True, prune_baseline=False, cache_root=True, tau_relu=1e-06, tau_max=0.01)[source]

Bases: qLogNoisyExpectedImprovement, MultiObjectiveMCAcquisitionFunction

q-LogNParEGO supporting m >= 2 outcomes. This acquisition function utilizes qLogNEI to compute the expected improvement over Chebyshev scalarization of the objectives.

This is adapted from qNParEGO proposed in [Daulton2020qehvi] to utilize log-improvement acquisition functions of [Ament2023logei]. See [Knowles2005] for the original ParEGO algorithm.

This implementation assumes maximization of all objectives. If any of the model outputs are to be minimized, either an objective should be used to negate the model outputs or the scalarization_weights should be provided with negative weights for the outputs to be minimized.

Args:
model: A fitted multi-output model, producing outputs for m objectives

and any number of outcome constraints. NOTE: The model posterior must have a mean attribute.

X_baseline: A batch_shape x r x d-dim Tensor of r design points

that have already been observed. These points are considered as the potential best design point.

scalarization_weights: A m-dim Tensor of weights to be used in the

Chebyshev scalarization. If omitted, samples from the unit simplex.

sampler: The sampler used to draw base samples. See MCAcquisitionFunction

more details.

objective: The MultiOutputMCAcquisitionObjective under which the samples are

evaluated before applying Chebyshev scalarization. Defaults to IdentityMultiOutputObjective().

constraints: A list of constraint callables which map a Tensor of posterior

samples of dimension sample_shape x batch-shape x q x m’-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are satisfied if constraint(samples) < 0.

X_pending: A batch_shape x q’ x d-dim Tensor of q’ design points

that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

eta: Temperature parameter(s) governing the smoothness of the sigmoid

approximation to the constraint indicators. See the docs of compute_(log_)smoothed_constraint_indicator for details.

fat: Toggles the logarithmic / linear asymptotic behavior of the smooth

approximation to the ReLU.

prune_baseline: If True, remove points in X_baseline that are

highly unlikely to be the best point. This can significantly improve performance and is generally recommended. In order to customize pruning parameters, instead manually call botorch.acquisition.utils.prune_inferior_points on X_baseline before instantiating the acquisition function.

cache_root: A boolean indicating whether to cache the root

decomposition over X_baseline and use low-rank updates.

tau_max: Temperature parameter controlling the sharpness of the smooth

approximations to max.

tau_relu: Temperature parameter controlling the sharpness of the smooth

approximations to ReLU.

Parameters:
  • model (Model)

  • X_baseline (Tensor)

  • scalarization_weights (Tensor | None)

  • sampler (MCSampler | None)

  • objective (MCMultiOutputObjective | None)

  • constraints (list[Callable[[Tensor], Tensor]] | None)

  • X_pending (Tensor | None)

  • eta (Tensor | float)

  • fat (bool)

  • prune_baseline (bool)

  • cache_root (bool)

  • tau_relu (float)

  • tau_max (float)

The One-Shot Knowledge Gradient

Batch Knowledge Gradient (KG) via one-shot optimization as introduced in [Balandat2020botorch]. For broader discussion of KG see also [Frazier2008knowledge] and [Wu2016parallelkg].

[Balandat2020botorch] (1,2)

M. Balandat, B. Karrer, D. R. Jiang, S. Daulton, B. Letham, A. G. Wilson, and E. Bakshy. BoTorch: A Framework for Efficient Monte-Carlo Bayesian Optimization. Advances in Neural Information Processing Systems 33, 2020.

[Frazier2008knowledge]

P. Frazier, W. Powell, and S. Dayanik. A Knowledge-Gradient policy for sequential information collection. SIAM Journal on Control and Optimization, 2008.

[Wu2016parallelkg]

J. Wu and P. Frazier. The parallel knowledge gradient method for batch bayesian optimization. NIPS 2016.

class botorch.acquisition.knowledge_gradient.qKnowledgeGradient(model, num_fantasies=64, sampler=None, objective=None, posterior_transform=None, inner_sampler=None, X_pending=None, current_value=None)[source]

Bases: MCAcquisitionFunction, OneShotAcquisitionFunction

Batch Knowledge Gradient using one-shot optimization.

This computes the batch Knowledge Gradient using fantasies for the outer expectation and either the model posterior mean or MC-sampling for the inner expectation.

In addition to the design variables, the input X also includes variables for the optimal designs for each of the fantasy models. For a fixed number of fantasies, all parts of X can be optimized in a “one-shot” fashion.

q-Knowledge Gradient (one-shot optimization).

Parameters:
  • model (Model) – A fitted model. Must support fantasizing.

  • num_fantasies (Optional[int]) – The number of fantasy points to use. More fantasy points result in a better approximation, at the expense of memory and wall time. Unused if sampler is specified.

  • sampler (Optional[MCSampler]) – The sampler used to sample fantasy observations. Optional if num_fantasies is specified.

  • objective (Optional[MCAcquisitionObjective]) – The objective under which the samples are evaluated. If None, then the analytic posterior mean is used. Otherwise, the objective is MC-evaluated (using inner_sampler).

  • posterior_transform (Optional[PosteriorTransform]) – An optional PosteriorTransform. If given, this transforms the posterior before evaluation. If objective is None, then the analytic posterior mean of the transformed posterior is used. If objective is given, the inner_sampler is used to draw samples from the transformed posterior, which are then evaluated under the objective.

  • inner_sampler (Optional[MCSampler]) – The sampler used for inner sampling. Ignored if the objective is None.

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • current_value (Optional[Tensor]) – The current value, i.e. the expected best objective given the observed points D. If omitted, forward will not return the actual KG value, but the expected best objective given the data set D u X.

forward(X)[source]

Evaluate qKnowledgeGradient on the candidate set X.

Parameters:

X (Tensor) –

A b x (q + num_fantasies) x d Tensor with b t-batches of q + num_fantasies design points each. We split this X tensor into two parts in the q dimension (dim=-2). The first q are the q-batch of design points and the last num_fantasies are the current solutions of the inner optimization problem.

X_fantasies = X[…, -num_fantasies:, :] X_fantasies.shape = b x num_fantasies x d

X_actual = X[…, :-num_fantasies, :] X_actual.shape = b x q x d

Returns:

A Tensor of shape b. For t-batch b, the q-KG value of the design

X_actual[b] is averaged across the fantasy models, where X_fantasies[b, i] is chosen as the final selection for the i-th fantasy model. NOTE: If current_value is not provided, then this is not the true KG value of X_actual[b], and X_fantasies[b, : ] must be maximized at fixed X_actual[b].

Return type:

Tensor

evaluate(X, bounds, **kwargs)[source]

Evaluate qKnowledgeGradient on the candidate set X_actual by solving the inner optimization problem.

Parameters:
  • X (Tensor) – A b x q x d Tensor with b t-batches of q design points each. Unlike forward(), this does not include solutions of the inner optimization problem.

  • bounds (Tensor) – A 2 x d tensor of lower and upper bounds for each column of the solutions to the inner problem.

  • kwargs (Any) – Additional keyword arguments. This includes the options for optimization of the inner problem, i.e. num_restarts, raw_samples, an options dictionary to be passed on to the optimization helpers, and a scipy_options dictionary to be passed to scipy.minimize.

Returns:

A Tensor of shape b. For t-batch b, the q-KG value of the design

X[b] is averaged across the fantasy models. NOTE: If current_value is not provided, then this is not the true KG value of X[b].

Return type:

Tensor

get_augmented_q_batch_size(q)[source]

Get augmented q batch size for one-shot optimization.

Parameters:

q (int) – The number of candidates to consider jointly.

Returns:

The augmented size for one-shot optimization (including variables parameterizing the fantasy solutions).

Return type:

int

extract_candidates(X_full)[source]

We only return X as the set of candidates post-optimization.

Parameters:

X_full (Tensor) – A b x (q + num_fantasies) x d-dim Tensor with b t-batches of q + num_fantasies design points each.

Returns:

A b x q x d-dim Tensor with b t-batches of q design points each.

Return type:

Tensor

class botorch.acquisition.knowledge_gradient.qMultiFidelityKnowledgeGradient(model, num_fantasies=64, sampler=None, objective=None, posterior_transform=None, inner_sampler=None, X_pending=None, current_value=None, cost_aware_utility=None, project=<function qMultiFidelityKnowledgeGradient.<lambda>>, expand=<function qMultiFidelityKnowledgeGradient.<lambda>>, valfunc_cls=None, valfunc_argfac=None)[source]

Bases: qKnowledgeGradient

Batch Knowledge Gradient for multi-fidelity optimization.

A version of qKnowledgeGradient that supports multi-fidelity optimization via a CostAwareUtility and the project and expand operators. If none of these are set, this acquisition function reduces to qKnowledgeGradient. Through valfunc_cls and valfunc_argfac, this can be changed into a custom multi-fidelity acquisition function (it is only KG if the terminal value is computed using a posterior mean).

Multi-Fidelity q-Knowledge Gradient (one-shot optimization).

Parameters:
  • model (Model) – A fitted model. Must support fantasizing.

  • num_fantasies (Optional[int]) – The number of fantasy points to use. More fantasy points result in a better approximation, at the expense of memory and wall time. Unused if sampler is specified.

  • sampler (Optional[MCSampler]) – The sampler used to sample fantasy observations. Optional if num_fantasies is specified.

  • objective (Optional[MCAcquisitionObjective]) – The objective under which the samples are evaluated. If None, then the analytic posterior mean is used. Otherwise, the objective is MC-evaluated (using inner_sampler).

  • posterior_transform (Optional[PosteriorTransform]) – An optional PosteriorTransform. If given, this transforms the posterior before evaluation. If objective is None, then the analytic posterior mean of the transformed posterior is used. If objective is given, the inner_sampler is used to draw samples from the transformed posterior, which are then evaluated under the objective.

  • inner_sampler (Optional[MCSampler]) – The sampler used for inner sampling. Ignored if the objective is None.

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • current_value (Optional[Tensor]) – The current value, i.e. the expected best objective given the observed points D. If omitted, forward will not return the actual KG value, but the expected best objective given the data set D u X.

  • cost_aware_utility (Optional[CostAwareUtility]) – A CostAwareUtility computing the cost-transformed utility from a candidate set and samples of increases in utility.

  • project (Callable[[Tensor], Tensor]) – A callable mapping a batch_shape x q x d tensor of design points to a tensor with shape batch_shape x q_term x d projected to the desired target set (e.g. the target fidelities in case of multi-fidelity optimization). For the basic case, q_term = q.

  • expand (Callable[[Tensor], Tensor]) – A callable mapping a batch_shape x q x d input tensor to a batch_shape x (q + q_e)’ x d-dim output tensor, where the q_e additional points in each q-batch correspond to additional (“trace”) observations.

  • valfunc_cls (Optional[type[AcquisitionFunction]]) – An acquisition function class to be used as the terminal value function.

  • valfunc_argfac (Optional[Callable[[Model], dict[str, Any]]]) – An argument factory, i.e. callable that maps a Model to a dictionary of kwargs for the terminal value function (e.g. best_f for ExpectedImprovement).

property cost_sampler
forward(X)[source]

Evaluate qMultiFidelityKnowledgeGradient on the candidate set X.

Parameters:

X (Tensor) –

A b x (q + num_fantasies) x d Tensor with b t-batches of q + num_fantasies design points each. We split this X tensor into two parts in the q dimension (dim=-2). The first q are the q-batch of design points and the last num_fantasies are the current solutions of the inner optimization problem.

X_fantasies = X[…, -num_fantasies:, :] X_fantasies.shape = b x num_fantasies x d

X_actual = X[…, :-num_fantasies, :] X_actual.shape = b x q x d

In addition, X may be augmented with fidelity parameters as part of thee d-dimension. Projecting fidelities to the target fidelity is handled by project.

Returns:

A Tensor of shape b. For t-batch b, the q-KG value of the design

X_actual[b] is averaged across the fantasy models, where X_fantasies[b, i] is chosen as the final selection for the i-th fantasy model. NOTE: If current_value is not provided, then this is not the true KG value of X_actual[b], and X_fantasies[b, : ] must be maximized at fixed X_actual[b].

Return type:

Tensor

class botorch.acquisition.knowledge_gradient.ProjectedAcquisitionFunction(base_value_function, project)[source]

Bases: AcquisitionFunction

Defines a wrapper around an AcquisitionFunction that incorporates the project operator. Typically used to handle value functions in look-ahead methods.

Parameters:
  • base_value_function (AcquisitionFunction) – The wrapped AcquisitionFunction.

  • project (Callable[[Tensor], Tensor]) – A callable mapping a batch_shape x q x d tensor of design points to a tensor with shape batch_shape x q_term x d projected to the desired target set (e.g. the target fidelities in case of multi-fidelity optimization). For the basic case, q_term = q.

forward(X)[source]

Evaluate the acquisition function on the candidate set X.

Parameters:

X (Tensor) – A (b) x q x d-dim Tensor of (b) t-batches with q d-dim design points each.

Returns:

A (b)-dim Tensor of acquisition function values at the given design points X.

Return type:

Tensor

Multi-Step Lookahead Acquisition Functions

A general implementation of multi-step look-ahead acquisition function with configurable value functions. See [Jiang2020multistep].

[Jiang2020multistep]

S. Jiang, D. R. Jiang, M. Balandat, B. Karrer, J. Gardner, and R. Garnett. Efficient Nonmyopic Bayesian Optimization via One-Shot Multi-Step Trees. In Advances in Neural Information Processing Systems 33, 2020.

class botorch.acquisition.multi_step_lookahead.qMultiStepLookahead(model, batch_sizes, num_fantasies=None, samplers=None, valfunc_cls=None, valfunc_argfacs=None, objective=None, posterior_transform=None, inner_mc_samples=None, X_pending=None, collapse_fantasy_base_samples=True)[source]

Bases: MCAcquisitionFunction, OneShotAcquisitionFunction

MC-based batch Multi-Step Look-Ahead (one-shot optimization).

q-Multi-Step Look-Ahead (one-shot optimization).

Performs a k-step lookahead by means of repeated fantasizing.

Allows to specify the stage value functions by passing the respective class objects via the valfunc_cls list. Optionally, valfunc_argfacs takes a list of callables that generate additional kwargs for these constructors. By default, valfunc_cls will be chosen as [None, …, None, PosteriorMean], which corresponds to the (parallel) multi-step KnowledgeGradient. If, in addition, k=1 and q_1 = 1, this reduces to the classic Knowledge Gradient.

WARNING: The complexity of evaluating this function is exponential in the number of lookahead steps!

Parameters:
  • model (Model) – A fitted model.

  • batch_sizes (list[int]) – A list [q_1, …, q_k] containing the batch sizes for the k look-ahead steps.

  • num_fantasies (Optional[list[int]]) – A list [f_1, …, f_k] containing the number of fantasy points to use for the k look-ahead steps.

  • samplers (Optional[list[MCSampler]]) – A list of MCSampler objects to be used for sampling fantasies in each stage.

  • valfunc_cls (Optional[list[Optional[type[AcquisitionFunction]]]]) – A list of k + 1 acquisition function classes to be used as the (stage + terminal) value functions. Each element (except for the last one) can be None, in which case a zero stage value is assumed for the respective stage. If None, this defaults to [None, …, None, PosteriorMean]

  • valfunc_argfacs (Optional[list[Optional[TAcqfArgConstructor]]]) – A list of k + 1 “argument factories”, i.e. callables that map a Model and input tensor X to a dictionary of kwargs for the respective stage value function constructor (e.g. best_f for ExpectedImprovement). If None, only the standard (model, sampler and objective) kwargs will be used.

  • objective (Optional[MCAcquisitionObjective]) – The objective under which the output is evaluated. If None, use the model output (requires a single-output model or a posterior transform). Otherwise the objective is MC-evaluated (using inner_sampler).

  • posterior_transform (Optional[PosteriorTransform]) – An optional PosteriorTransform. If given, this transforms the posterior before evaluation. If objective is None, then the output of the transformed posterior is used. If objective is given, the inner_sampler is used to draw samples from the transformed posterior, which are then evaluated under the objective.

  • inner_mc_samples (Optional[list[int]]) – A list [n_0, …, n_k] containing the number of MC samples to be used for evaluating the stage value function. Ignored if the objective is None.

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • collapse_fantasy_base_samples (bool) – If True, collapse_batch_dims of the Samplers will be applied on fantasy batch dimensions as well, meaning that base samples are the same in all subtrees starting from the same level.

forward(X)[source]

Evaluate qMultiStepLookahead on the candidate set X.

Parameters:

X (Tensor) – A batch_shape x q’ x d-dim Tensor with q’ design points for each batch, where q’ = q_0 + f_1 q_1 + f_2 f_1 q_2 + …. Here q_i is the number of candidates jointly considered in look-ahead step i, and f_i is respective number of fantasies.

Returns:

The acquisition value for each batch as a tensor of shape batch_shape.

Return type:

Tensor

get_augmented_q_batch_size(q)[source]

Get augmented q batch size for one-shot optimization.

Parameters:

q (int) – The number of candidates to consider jointly.

Returns:

The augmented size for one-shot optimization (including variables parameterizing the fantasy solutions): q_0 + f_1 q_1 + f_2 f_1 q_2 + …

Return type:

int

get_split_shapes(X)[source]

Get the split shapes from X.

Parameters:

X (Tensor) – A batch_shape x q_aug x d-dim tensor including fantasy points.

Returns:

A 3-tuple (batch_shape, shapes, sizes), where shape[i] = f_i x …. x f_1 x batch_shape x q_i x d and size[i] = f_i * … f_1 * q_i.

Return type:

tuple[Size, list[Size], list[int]]

get_multi_step_tree_input_representation(X)[source]

Get the multi-step tree representation of X.

Parameters:

X (Tensor) – A batch_shape x q’ x d-dim Tensor with q’ design points for each batch, where q’ = q_0 + f_1 q_1 + f_2 f_1 q_2 + …. Here q_i is the number of candidates jointly considered in look-ahead step i, and f_i is respective number of fantasies.

Returns:

A list [X_j, …, X_k] of tensors, where X_i has shape f_i x …. x f_1 x batch_shape x q_i x d.

Return type:

list[Tensor]

extract_candidates(X_full)[source]

We only return X as the set of candidates post-optimization.

Parameters:

X_full (Tensor) – A batch_shape x q’ x d-dim Tensor with q’ design points for each batch, where q’ = q + f_1 q_1 + f_2 f_1 q_2 + ….

Returns:

A batch_shape x q x d-dim Tensor with q design points for each batch.

Return type:

Tensor

get_induced_fantasy_model(X)[source]

Fantasy model induced by X.

Parameters:

X (Tensor) – A batch_shape x q’ x d-dim Tensor with q’ design points for each batch, where q’ = q_0 + f_1 q_1 + f_2 f_1 q_2 + …. Here q_i is the number of candidates jointly considered in look-ahead step i, and f_i is respective number of fantasies.

Returns:

The fantasy model induced by X.

Return type:

Model

botorch.acquisition.multi_step_lookahead.warmstart_multistep(acq_function, bounds, num_restarts, raw_samples, full_optimizer)[source]

Warm-start initialization for multi-step look-ahead acquisition functions.

For now uses the same q’ as in full_optimizer. TODO: allow different q.

Parameters:
  • acq_function (qMultiStepLookahead) – A qMultiStepLookahead acquisition function.

  • bounds (Tensor) – A 2 x d tensor of lower and upper bounds for each column of features.

  • num_restarts (int) – The number of starting points for multistart acquisition function optimization.

  • raw_samples (int) – The number of raw samples to consider in the initialization heuristic.

  • full_optimizer (Tensor) – The full tree of optimizers of the previous iteration of shape batch_shape x q’ x d. Typically obtained by passing return_best_only=False and return_full_tree=True into optimize_acqf.

Returns:

A num_restarts x q’ x d tensor for initial points for optimization.

Return type:

Tensor

This is a very simple initialization heuristic. TODO: Use the observed values to identify the fantasy sub-tree that is closest to the observed value.

botorch.acquisition.multi_step_lookahead.make_best_f(model, X)[source]

Extract the best observed training input from the model.

Parameters:
  • model (Model)

  • X (Tensor)

Return type:

dict[str, Any]

Active Learning Acquisition Functions

Active learning acquisition functions.

[Seo2014activedata]

S. Seo, M. Wallat, T. Graepel, and K. Obermayer. Gaussian process regression: Active data selection and test point rejection. IJCNN 2000.

[Chen2014seqexpdesign]

X. Chen and Q. Zhou. Sequential experimental designs for stochastic kriging. Winter Simulation Conference 2014.

[Binois2017repexp]

M. Binois, J. Huang, R. B. Gramacy, and M. Ludkovski. Replication or exploration? Sequential design for stochastic simulation experiments. ArXiv 2017.

class botorch.acquisition.active_learning.qNegIntegratedPosteriorVariance(model, mc_points, sampler=None, posterior_transform=None, X_pending=None)[source]

Bases: AcquisitionFunction

Batch Integrated Negative Posterior Variance for Active Learning.

This acquisition function quantifies the (negative) integrated posterior variance (excluding observation noise, computed using MC integration) of the model. In that, it is a proxy for global model uncertainty, and thus purely focused on “exploration”, rather the “exploitation” of many of the classic Bayesian Optimization acquisition functions.

See [Seo2014activedata], [Chen2014seqexpdesign], and [Binois2017repexp].

q-Integrated Negative Posterior Variance.

Parameters:
  • model (Model) – A fitted model.

  • mc_points (Tensor) – A batch_shape x N x d tensor of points to use for MC-integrating the posterior variance. Usually, these are qMC samples on the whole design space, but biased sampling directly allows weighted integration of the posterior variance.

  • sampler (Optional[MCSampler]) – The sampler used for drawing fantasy samples. In the basic setting of a standard GP (default) this is a dummy, since the variance of the model after conditioning does not actually depend on the sampled values.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • X_pending (Optional[Tensor]) – A n’ x d-dim Tensor of n’ design points that have points that have been submitted for function evaluation but have not yet been evaluated.

forward(X)[source]

Evaluate the acquisition function on the candidate set X.

Parameters:

X (Tensor) – A (b) x q x d-dim Tensor of (b) t-batches with q d-dim design points each.

Returns:

A (b)-dim Tensor of acquisition function values at the given design points X.

Return type:

Tensor

class botorch.acquisition.active_learning.PairwiseMCPosteriorVariance(model, objective, sampler=None)[source]

Bases: MCAcquisitionFunction

Variance of difference for Active Learning

Given a model and an objective, calculate the posterior sample variance of the objective on the difference of pairs of points. See more implementation details in forward. This acquisition function is typically used with a pairwise model (e.g., PairwiseGP) and a likelihood/link function on the pair difference (e.g., logistic or probit) for pure exploration

Pairwise Monte Carlo Posterior Variance

Parameters:
  • model (Model) – A fitted model.

  • objective (MCAcquisitionObjective) – An MCAcquisitionObjective representing the link function (e.g., logistic or probit.) applied on the difference of (usually 1-d) two samples. Can be implemented via GenericMCObjective.

  • sampler (Optional[MCSampler]) – The sampler used for drawing MC samples.

forward(X)[source]

Evaluate PairwiseMCPosteriorVariance on the candidate set X.

Parameters:

X (Tensor) – A batch_size x q x d-dim Tensor. q should be a multiple of 2.

Returns:

Tensor of shape batch_size x q representing the posterior variance of link function at X that active learning hopes to maximize

Return type:

Tensor

Bayesian Active Learning Acquisition Functions

Acquisition functions for Bayesian active learning. This includes: BALD [Houlsby2011bald] and its batch version [kirsch2019batchbald].

References

[kirsch2019batchbald] (1,2)

Andreas Kirsch, Joost van Amersfoort, Yarin Gal. BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning. In Proceedings of the Annual Conference on Neural Information Processing Systems (NeurIPS), 2019.

botorch.acquisition.bayesian_active_learning.check_negative_info_gain(info_gain)[source]

Check if the (expected) information gain is negative, raise a warning if so.

Parameters:

info_gain (Tensor)

Return type:

None

class botorch.acquisition.bayesian_active_learning.FullyBayesianAcquisitionFunction(model)[source]

Bases: AcquisitionFunction

Base class for acquisition functions which require a Fully Bayesian model treatment.

Parameters:

model (Model) – A fully bayesian single-outcome model.

class botorch.acquisition.bayesian_active_learning.qBayesianActiveLearningByDisagreement(model, sampler=None, posterior_transform=None, X_pending=None)[source]

Bases: FullyBayesianAcquisitionFunction, MCSamplerMixin

Batch implementation [kirsch2019batchbald] of BALD [Houlsby2011bald], which maximizes the mutual information between the next observation and the hyperparameters of the model. Computed by Monte Carlo integration.

Parameters:
  • model (Union[ModelListGP, SaasFullyBayesianSingleTaskGP]) – A fully bayesian model (SaasFullyBayesianSingleTaskGP).

  • sampler (Optional[MCSampler]) – The sampler used for drawing samples to approximate the entropy of the Gaussian Mixture posterior.

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

  • X_pending (Optional[Tensor]) – A batch_shape x m x d-dim Tensor of m design points

forward(X)[source]

Evaluate qBayesianActiveLearningByDisagreement on the candidate set X. A monte carlo-estimated information gain is computed over a Gaussian Mixture marginal posterior, and the Gaussian conditional posterior to obtain the qBayesianActiveLearningByDisagreement on the candidate set X.

Parameters:

X (Tensor) – batch_shape x q x D-dim Tensor of input points.

Returns:

A batch_shape x num_models-dim Tensor of BALD values.

Return type:

Tensor

Preference Acquisition Functions

Preference acquisition functions. This includes: Analytical EUBO acquisition function as introduced in [Lin2022preference] and its MC-based generalization qEUBO as proposed in [Astudillo2023qeubo].

[Astudillo2023qeubo] (1,2)

Astudillo, R., Lin, Z.J., Bakshy, E. and Frazier, P.I. qEUBO: A Decision-Theoretic Acquisition Function for Preferential Bayesian Optimization. International Conference on Artificial Intelligence and Statistics (AISTATS), 2023.

[Lin2022preference] (1,2,3)

Lin, Z.J., Astudillo, R., Frazier, P.I. and Bakshy, E. Preference Exploration for Efficient Bayesian Optimization with Multiple Outcomes. International Conference on Artificial Intelligence and Statistics (AISTATS), 2022.

[Houlsby2011bald] (1,2,3)

Houlsby, N., Huszár, F., Ghahramani, Z. and Lengyel, M. Bayesian Active Learning for Gaussian Process Classification. NIPS Workshop on Bayesian optimization, experimental design and bandits: Theory and applications, 2011.

class botorch.acquisition.preference.AnalyticExpectedUtilityOfBestOption(pref_model, outcome_model=None, previous_winner=None)[source]

Bases: AnalyticAcquisitionFunction

Analytic Preferential Expected Utility of Best Options, i.e., Analytical EUBO

Analytic implementation of Expected Utility of the Best Option under the Laplace model (assumes a PairwiseGP is used as the preference model) as proposed in [Lin2022preference].

Parameters:
  • pref_model (Model) – The preference model that maps the outcomes (i.e., Y) to scalar-valued utility.

  • outcome_model (Optional[DeterministicModel]) – A deterministic model that maps parameters (i.e., X) to outcomes (i.e., Y). The outcome model f defines the search space of Y = f(X). If model is None, we are directly calculating EUBO on the parameter space. When used with OneSamplePosteriorDrawModel, we are obtaining EUBO-zeta as described in [Lin2022preference].

  • previous_winner (Optional[Tensor]) – Tensor representing the previous winner in the Y space.

forward(X)[source]

Evaluate analytical EUBO on the candidate set X.

Parameters:

X (Tensor) – A batch_shape x q x d-dim Tensor, where q = 2 if previous_winner is not None, and q = 1 otherwise.

Returns:

The acquisition value for each batch as a tensor of shape batch_shape.

Return type:

Tensor

class botorch.acquisition.preference.qExpectedUtilityOfBestOption(pref_model, outcome_model=None, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Bases: MCAcquisitionFunction

MC-based Expected Utility of Best Option (qEUBO)

This computes qEUBO by (1) sampling the joint posterior over q points (2) evaluating the maximum objective value accross the q points (3) averaging over the samples

qEUBO(X) = E[max Y], Y ~ f(X), where X = (x_1,…,x_q)

MC-based Expected Utility of Best Option (qEUBO) as proposed in [Astudillo2023qeubo].

Parameters:
  • pref_model (Model) – The preference model that maps the outcomes (i.e., Y) to scalar-valued utility.

  • outcome_model (Optional[DeterministicModel]) –

    A deterministic model that maps parameters (i.e., X) to

    outcomes (i.e., Y). The outcome model f defines the search space of Y = f(X). If model is None, we are directly calculating qEUBO on the parameter space.

    sampler: The sampler used to draw base samples. See MCAcquisitionFunction

    more details.

  • objective (Optional[MCAcquisitionObjective]) – The MCAcquisitionObjective under which the samples are evaluated. Defaults to IdentityMCObjective().

  • posterior_transform (Optional[PosteriorTransform]) – A PosteriorTransform (optional).

  • X_pending (Optional[Tensor]) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call. Copied and set to have no gradient.

  • sampler (Optional[MCSampler])

forward(X)[source]

Evaluate qEUBO on the candidate set X.

Parameters:

X (Tensor) – A batch_shape x q x d-dim Tensor of t-batches with q d-dim design points each.

Returns:

A batch_shape’-dim Tensor of qEUBO values at the given design points X, where batch_shape’ is the broadcasted batch shape of model and input X.

Return type:

Tensor

class botorch.acquisition.preference.PairwiseBayesianActiveLearningByDisagreement(pref_model, outcome_model=None, num_samples=1024, std_noise=0.0)[source]

Bases: MCAcquisitionFunction

MC Bayesian Active Learning by Disagreement

Monte Carlo implementation of Bayesian Active Learning by Disagreement (BALD) proposed in [Houlsby2011bald].

Parameters:
  • pref_model (Model) – The preference model that maps the outcomes (i.e., Y) to scalar-valued utility.

  • outcome_model (Optional[DeterministicModel]) – A deterministic model that maps parameters (i.e., X) to outcomes (i.e., Y). The outcome model f defines the search space of Y = f(X). If model is None, we are directly calculating BALD on the parameter space.

  • num_samples (Optional[int]) – number of samples to approximate the conditional_entropy.

  • std_noise (Optional[float]) – Additional observational noise to include. Defaults to 0.

forward(X)[source]

Evaluate MC BALD on the candidate set X.

Parameters:

X (Tensor) – A batch_shape x 2 x d-dim Tensor of t-batches with q=2 d-dim design points each.

Returns:

A batch_shape’-dim Tensor of MC BALD values at the given design points pair X, where batch_shape’ is the broadcasted batch shape of model and input X.

Return type:

Tensor

Objectives and Cost-Aware Utilities

Objectives

Objective Modules to be used with acquisition functions.

class botorch.acquisition.objective.PosteriorTransform(*args, **kwargs)[source]

Bases: Module, ABC

Abstract base class for objectives that transform the posterior.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

abstract evaluate(Y)[source]

Evaluate the transform on a set of outcomes.

Parameters:

Y (Tensor) – A batch_shape x q x m-dim tensor of outcomes.

Returns:

A batch_shape x q’ [x m’]-dim tensor of transformed outcomes.

Return type:

Tensor

abstract forward(posterior)[source]

Compute the transformed posterior.

Parameters:

posterior – The posterior to be transformed.

Returns:

The transformed posterior object.

Return type:

Posterior

class botorch.acquisition.objective.ScalarizedPosteriorTransform(weights, offset=0.0)[source]

Bases: PosteriorTransform

An affine posterior transform for scalarizing multi-output posteriors.

For a Gaussian posterior at a single point (q=1) with mean mu and covariance matrix Sigma, this yields a single-output posterior with mean weights^T * mu and variance weights^T Sigma w.

Example

Example for a model with two outcomes:

>>> weights = torch.tensor([0.5, 0.25])
>>> posterior_transform = ScalarizedPosteriorTransform(weights)
>>> EI = ExpectedImprovement(
... model, best_f=0.1, posterior_transform=posterior_transform
... )
Parameters:
  • weights (Tensor) – A one-dimensional tensor with m elements representing the linear weights on the outputs.

  • offset (float) – An offset to be added to posterior mean.

scalarize: bool = True
evaluate(Y)[source]

Evaluate the transform on a set of outcomes.

Parameters:

Y (Tensor) – A batch_shape x q x m-dim tensor of outcomes.

Returns:

A batch_shape x q-dim tensor of transformed outcomes.

Return type:

Tensor

forward(posterior)[source]

Compute the posterior of the affine transformation.

Parameters:

posterior (Union[GPyTorchPosterior, PosteriorList]) – A posterior with the same number of outputs as the elements in self.weights.

Returns:

A single-output posterior.

Return type:

GPyTorchPosterior

class botorch.acquisition.objective.ExpectationPosteriorTransform(n_w, weights=None)[source]

Bases: PosteriorTransform

Transform the batch x (q * n_w) x m posterior into a batch x q x m posterior of the expectation. The expectation is calculated over each consecutive n_w block of points in the posterior.

This is intended for use with InputPerturbation or AppendFeatures for optimizing the expectation over n_w points. This should not be used when there are constraints present, since this does not take into account the feasibility of the objectives.

Note: This is different than ScalarizedPosteriorTransform in that this operates over the q-batch dimension.

A posterior transform calculating the expectation over the q-batch dimension.

Parameters:
  • n_w (int) – The number of points in the q-batch of the posterior to compute the expectation over. This corresponds to the size of the feature_set of AppendFeatures or the size of the perturbation_set of InputPerturbation.

  • weights (Optional[Tensor]) – An optional n_w x m-dim tensor of weights. Can be used to compute a weighted expectation. Weights are normalized before use.

evaluate(Y)[source]

Evaluate the expectation of a set of outcomes.

Parameters:

Y (Tensor) – A batch_shape x (q * n_w) x m-dim tensor of outcomes.

Returns:

A batch_shape x q x m-dim tensor of expectation outcomes.

Return type:

Tensor

forward(posterior)[source]

Compute the posterior of the expectation.

Parameters:

posterior (GPyTorchPosterior) – An m-outcome joint posterior over q * n_w points.

Returns:

An m-outcome joint posterior over q expectations.

Return type:

GPyTorchPosterior

class botorch.acquisition.objective.MCAcquisitionObjective(*args, **kwargs)[source]

Bases: Module, ABC

Abstract base class for MC-based objectives.

Parameters:
  • _verify_output_shape – If True and X is given, check that the q-batch shape of the objectives agrees with that of X.

  • _is_mo – A boolean denoting whether the objectives are multi-output.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

abstract forward(samples, X=None)[source]

Evaluate the objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim Tensor of objective values (assuming maximization).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcome = mc_obj(samples)
class botorch.acquisition.objective.IdentityMCObjective(*args, **kwargs)[source]

Bases: MCAcquisitionObjective

Trivial objective extracting the last dimension.

Example

>>> identity_objective = IdentityMCObjective()
>>> samples = sampler(posterior)
>>> objective = identity_objective(samples)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(samples, X=None)[source]

Evaluate the objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim Tensor of objective values (assuming maximization).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcome = mc_obj(samples)
class botorch.acquisition.objective.LinearMCObjective(weights)[source]

Bases: MCAcquisitionObjective

Linear objective constructed from a weight tensor.

For input samples and mc_obj = LinearMCObjective(weights), this produces mc_obj(samples) = sum_{i} weights[i] * samples[…, i]

Example

Example for a model with two outcomes:

>>> weights = torch.tensor([0.75, 0.25])
>>> linear_objective = LinearMCObjective(weights)
>>> samples = sampler(posterior)
>>> objective = linear_objective(samples)
Parameters:

weights (Tensor) – A one-dimensional tensor with m elements representing the linear weights on the outputs.

forward(samples, X=None)[source]

Evaluate the linear objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim tensor of objective values.

Return type:

Tensor

class botorch.acquisition.objective.GenericMCObjective(objective)[source]

Bases: MCAcquisitionObjective

Objective generated from a generic callable.

Allows to construct arbitrary MC-objective functions from a generic callable. In order to be able to use gradient-based acquisition function optimization it should be possible to backpropagate through the callable.

Example

>>> generic_objective = GenericMCObjective(
        lambda Y, X: torch.sqrt(Y).sum(dim=-1),
    )
>>> samples = sampler(posterior)
>>> objective = generic_objective(samples)
Parameters:

objective (Callable[[Tensor, Optional[Tensor]], Tensor]) – A callable f(samples, X) mapping a sample_shape x batch-shape x q x m-dim Tensor samples and an optional batch-shape x q x d-dim Tensor X to a sample_shape x batch-shape x q-dim Tensor of objective values.

forward(samples, X=None)[source]

Evaluate the objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim Tensor of objective values.

Return type:

Tensor

class botorch.acquisition.objective.ConstrainedMCObjective(objective, constraints, infeasible_cost=0.0, eta=0.001)[source]

Bases: GenericMCObjective

Feasibility-weighted objective.

An Objective allowing to maximize some scalable objective on the model outputs subject to a number of constraints. Constraint feasibilty is approximated by a sigmoid function.

mc_acq(X) = ( (objective(X) + infeasible_cost) * prod_i (1 - sigmoid(constraint_i(X))) ) - infeasible_cost

See botorch.utils.objective.apply_constraints for details on the constraint handling.

Example

>>> bound = 0.0
>>> objective = lambda Y: Y[..., 0]
>>> # apply non-negativity constraint on f(x)[1]
>>> constraint = lambda Y: bound - Y[..., 1]
>>> constrained_objective = ConstrainedMCObjective(objective, [constraint])
>>> samples = sampler(posterior)
>>> objective = constrained_objective(samples)

TODO: Deprecate this as default way to handle constraints with MC acquisition functions once we have data on how well SampleReducingMCAcquisitionFunction works.

Parameters:
  • objective (Callable[[Tensor, Optional[Tensor]], Tensor]) – A callable f(samples, X) mapping a sample_shape x batch-shape x q x m-dim Tensor samples and an optional batch-shape x q x d-dim Tensor X to a sample_shape x batch-shape x q-dim Tensor of objective values.

  • constraints (list[Callable[[Tensor], Tensor]]) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility.

  • infeasible_cost (Union[Tensor, float]) – The cost of a design if all associated samples are infeasible.

  • eta (Union[Tensor, float]) – The temperature parameter of the sigmoid function approximating the constraint. Can be either a float or a 1-dim tensor. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value.

forward(samples, X=None)[source]

Evaluate the feasibility-weighted objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim Tensor of objective values weighted by feasibility (assuming maximization).

Return type:

Tensor

class botorch.acquisition.objective.LearnedObjective(pref_model, sample_shape=None, seed=None)[source]

Bases: MCAcquisitionObjective

Learned preference objective constructed from a preference model.

For input samples, it samples each individual sample again from the latent preference posterior distribution using pref_model and return the posterior mean.

Example

>>> train_X = torch.rand(2, 2)
>>> train_comps = torch.LongTensor([[0, 1]])
>>> pref_model = PairwiseGP(train_X, train_comps)
>>> learned_pref_obj = LearnedObjective(pref_model)
>>> samples = sampler(posterior)
>>> objective = learned_pref_obj(samples)
Parameters:
  • pref_model (Model) – A BoTorch model, which models the latent preference/utility function. Given an input tensor of size sample_size x batch_shape x q x d, its posterior method should return a Posterior object with single outcome representing the utility values of the input.

  • sample_shape (Optional[torch.Size]) – Determines the number of preference-model samples drawn per outcome-model sample when the LearnedObjective is called. Note that this is an additional layer of sampling relative to what is needed when evaluating most MC acquisition functions in order to account for uncertainty in the preference model. If None, it will default to torch.Size([16]), so that 16 samples will be drawn from the preference model at each outcome sample. This number is relatively high because sampling from the preference model is general cheap relative to generating the outcome model posterior.

  • seed (Optional[int])

forward(samples, X=None)[source]

Sample each element of samples.

Parameters:
  • samples (Tensor) – A sample_size x batch_shape x q x d-dim Tensors of samples from a model posterior.

  • X (Tensor | None)

Returns:

A (sample_size * num_samples) x batch_shape x q-dim Tensor of objective values sampled from utility posterior using pref_model.

Return type:

Tensor

Multi-Objective Objectives

class botorch.acquisition.multi_objective.objective.MCMultiOutputObjective(*args, **kwargs)[source]

Bases: MCAcquisitionObjective

Abstract base class for MC multi-output objectives.

Parameters:

_is_mo – A boolean denoting whether the objectives are multi-output.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

abstract forward(samples, X=None)[source]

Evaluate the multi-output objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim Tensors of inputs.

Returns:

A sample_shape x batch_shape x q x m’-dim Tensor of objective values with m’ the output dimension. This assumes maximization in each output dimension).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcomes = multi_obj(samples)
class botorch.acquisition.multi_objective.objective.GenericMCMultiOutputObjective(objective)[source]

Bases: GenericMCObjective, MCMultiOutputObjective

Multi-output objective generated from a generic callable.

Allows to construct arbitrary MC-objective functions from a generic callable. In order to be able to use gradient-based acquisition function optimization it should be possible to backpropagate through the callable.

Parameters:

objective (Callable[[Tensor, Optional[Tensor]], Tensor]) – A callable f(samples, X) mapping a sample_shape x batch-shape x q x m-dim Tensor samples and an optional batch-shape x q x d-dim Tensor X to a sample_shape x batch-shape x q-dim Tensor of objective values.

class botorch.acquisition.multi_objective.objective.IdentityMCMultiOutputObjective(outcomes=None, num_outcomes=None)[source]

Bases: MCMultiOutputObjective

Trivial objective that returns the unaltered samples.

Example

>>> identity_objective = IdentityMCMultiOutputObjective()
>>> samples = sampler(posterior)
>>> objective = identity_objective(samples)

Initialize Objective.

Parameters:
  • outcomes (Optional[list[int]]) – A list of the m’ indices that the weights should be applied to.

  • num_outcomes (Optional[int]) – The total number of outcomes m

forward(samples, X=None)[source]

Evaluate the multi-output objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim Tensors of inputs.

Returns:

A sample_shape x batch_shape x q x m’-dim Tensor of objective values with m’ the output dimension. This assumes maximization in each output dimension).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcomes = multi_obj(samples)
class botorch.acquisition.multi_objective.objective.WeightedMCMultiOutputObjective(weights, outcomes=None, num_outcomes=None)[source]

Bases: IdentityMCMultiOutputObjective

Objective that reweights samples by given weights vector.

Example

>>> weights = torch.tensor([1.0, -1.0])
>>> weighted_objective = WeightedMCMultiOutputObjective(weights)
>>> samples = sampler(posterior)
>>> objective = weighted_objective(samples)

Initialize Objective.

Parameters:
  • weights (Tensor) – m’-dim tensor of outcome weights.

  • outcomes (Optional[list[int]]) – A list of the m’ indices that the weights should be applied to.

  • num_outcomes (Optional[int]) – the total number of outcomes m

forward(samples, X=None)[source]

Evaluate the multi-output objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim Tensors of inputs.

Returns:

A sample_shape x batch_shape x q x m’-dim Tensor of objective values with m’ the output dimension. This assumes maximization in each output dimension).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcomes = multi_obj(samples)
class botorch.acquisition.multi_objective.objective.FeasibilityWeightedMCMultiOutputObjective(model, X_baseline, constraint_idcs, objective=None)[source]

Bases: MCMultiOutputObjective

Construct a feasibility-weighted objective.

This applies feasibility weighting before calculating the objective value. Defaults to identity if no constraints or objective is present.

NOTE: By passing in a single-output MCAcquisitionObjective as the objective, this can be used as a single-output MCAcquisitionObjective as well.

Parameters:
  • model (Model) – A fitted Model.

  • X_baseline (Tensor) – An n x d-dim tensor of points already observed.

  • constraint_idcs (list[int]) – The outcome indices of the constraints. Constraints are handled by weighting the samples according to a sigmoid approximation of feasibility. A positive constraint outcome implies feasibility.

  • objective (Optional[MCMultiOutputObjective]) – An optional objective to apply after feasibility-weighting the samples.

forward(samples, X=None)[source]

Evaluate the multi-output objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim Tensors of inputs.

Returns:

A sample_shape x batch_shape x q x m’-dim Tensor of objective values with m’ the output dimension. This assumes maximization in each output dimension).

Return type:

Tensor

This method is usually not called directly, but via the objectives.

Example

>>> # `__call__` method:
>>> samples = sampler(posterior)
>>> outcomes = multi_obj(samples)

Cost-Aware Utility

Cost functions for cost-aware acquisition functions, e.g. multi-fidelity KG. To be used in a context where there is an objective/cost tradeoff.

class botorch.acquisition.cost_aware.CostAwareUtility(*args, **kwargs)[source]

Bases: Module, ABC

Abstract base class for cost-aware utilities.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

abstract forward(X, deltas, sampler=None)[source]

Evaluate the cost-aware utility on the candidates and improvements.

Parameters:
  • X (Tensor) – A batch_shape x q x d-dim Tensor of with q d-dim design points each for each t-batch.

  • deltas (Tensor) – A num_fantasies x batch_shape-dim Tensor of num_fantasy samples from the marginal improvement in utility over the current state at X for each t-batch.

  • sampler (MCSampler | None) – A sampler used for sampling from the posterior of the cost model. Some subclasses ignore this argument.

Returns:

A num_fantasies x batch_shape-dim Tensor of cost-transformed utilities.

Return type:

Tensor

class botorch.acquisition.cost_aware.GenericCostAwareUtility(cost)[source]

Bases: CostAwareUtility

Generic cost-aware utility wrapping a callable.

Generic cost-aware utility wrapping a callable.

Parameters:

cost (Callable[[Tensor, Tensor], Tensor]) – A callable mapping a batch_shape x q x d’-dim candidate set to a batch_shape-dim tensor of costs

forward(X, deltas, sampler=None)[source]

Evaluate the cost function on the candidates and improvements.

Parameters:
  • X (Tensor) – A batch_shape x q x d’-dim Tensor of with q d-dim design points for each t-batch.

  • deltas (Tensor) – A num_fantasies x batch_shape-dim Tensor of num_fantasy samples from the marginal improvement in utility over the current state at X for each t-batch.

  • sampler (MCSampler | None) – Ignored.

Returns:

A num_fantasies x batch_shape-dim Tensor of cost-weighted utilities.

Return type:

Tensor

class botorch.acquisition.cost_aware.InverseCostWeightedUtility(cost_model, use_mean=True, cost_objective=None, min_cost=0.01)[source]

Bases: CostAwareUtility

A cost-aware utility using inverse cost weighting based on a model.

Computes the cost-aware utility by inverse-weighting samples U = (u_1, …, u_N) of the increase in utility. If use_mean=True, this uses the posterior mean mean_cost of the cost model, i.e. weighted utility = mean(U) / mean_cost. If use_mean=False, it uses samples C = (c_1, …, c_N) from the posterior of the cost model and performs the inverse weighting on the sample level: weighted utility = mean(u_1 / c_1, …, u_N / c_N).

Where values in (u_1, …, u_N) are negative, or for mean(U) < 0, the weighted utility is instead calculated via scaling by the cost, i.e. if use_mean=True: weighted_utility = mean(U) * mean_cost and if use_mean=False: weighted utility = mean(u_1 * c_1, u_2 / c_2, u_3 * c_3, …, u_N / c_N), depending on whether (u_* >= 0), as with u_2 and u_N in this case, or (u_* < 0) as with u_1 and u_3.

The cost is additive across multiple elements of a q-batch.

Cost-aware utility that weights increase in utility by inverse cost. For negative increases in utility, the utility is instead scaled by the cost. See the class description for more information.

Parameters:
  • cost_model (Union[DeterministicModel, GPyTorchModel]) – A model of the cost of evaluating a candidate set X, where X are the same features as in the model for the acquisition function this is to be used with. If no cost_objective is specified, the outputs are required to be non-negative.

  • use_mean (bool) – If True, use the posterior mean, otherwise use posterior samples from the cost model.

  • cost_objective (Optional[MCAcquisitionObjective]) – If specified, transform the posterior mean / the posterior samples from the cost model. This can be used e.g. to un-transform predictions/samples of a cost model fit on the log-transformed cost (often done to ensure non-negativity). If the cost model is multi-output, then by default this will sum the cost across outputs.

  • min_cost (float) – A value used to clamp the cost samples so that they are not too close to zero, which may cause numerical issues.

Returns:

The inverse-cost-weighted utility.

forward(X, deltas, sampler=None, X_evaluation_mask=None)[source]

Evaluate the cost function on the candidates and improvements. Note that negative values of deltas are instead scaled by the cost, and not inverse-weighted. See the class description for more information.

Parameters:
  • X (Tensor) – A batch_shape x q x d-dim Tensor of with q d-dim design points each for each t-batch.

  • deltas (Tensor) – A num_fantasies x batch_shape-dim Tensor of num_fantasy samples from the marginal improvement in utility over the current state at X for each t-batch.

  • sampler (MCSampler | None) – A sampler used for sampling from the posterior of the cost model (required if use_mean=False, ignored if use_mean=True).

  • X_evaluation_mask (Tensor | None) – A q x m-dim boolean tensor indicating which outcomes should be evaluated for each design in the batch.

Returns:

A num_fantasies x batch_shape-dim Tensor of cost-weighted utilities.

Return type:

Tensor

Risk Measures

Risk Measures implemented as Monte-Carlo objectives, based on Bayesian optimization of risk measures as introduced in [Cakmak2020risk]. For a broader discussion of Monte-Carlo methods for VaR and CVaR risk measures, see also [Hong2014review].

[Cakmak2020risk]

S. Cakmak, R. Astudillo, P. Frazier, and E. Zhou. Bayesian Optimization of Risk Measures. Advances in Neural Information Processing Systems 33, 2020.

[Hong2014review]

L. J. Hong, Z. Hu, and G. Liu. Monte carlo methods for value-at-risk and conditional value-at-risk: a review. ACM Transactions on Modeling and Computer Simulation, 2014.

class botorch.acquisition.risk_measures.RiskMeasureMCObjective(n_w, preprocessing_function=None)[source]

Bases: MCAcquisitionObjective, ABC

Objective transforming the posterior samples to samples of a risk measure.

The risk measure is calculated over joint q-batch samples from the posterior. If the q-batch includes samples corresponding to multiple inputs, it is assumed that first n_w samples correspond to first input, second n_w samples correspond to second input etc.

The risk measures are commonly defined for minimization by considering the upper tail of the distribution, i.e., treating larger values as being undesirable. BoTorch by default assumes a maximization objective, so the default behavior here is to calculate the risk measures w.r.t. the lower tail of the distribution. This can be changed by passing a preprocessing function with weights=torch.tensor([-1.0]).

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

abstract forward(samples, X=None)[source]

Calculate the risk measure corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q-dim tensor of risk measure samples.

Return type:

Tensor

class botorch.acquisition.risk_measures.CVaR(alpha, n_w, preprocessing_function=None)[source]

Bases: RiskMeasureMCObjective

The Conditional Value-at-Risk risk measure.

The Conditional Value-at-Risk measures the expectation of the worst outcomes (small rewards or large losses) with a total probability of 1 - alpha. It is commonly defined as the conditional expectation of the reward function, with the condition that the reward is smaller than the corresponding Value-at-Risk (also defined below).

Note: Due to the use of a discrete w_set of samples, the VaR and CVaR

calculated here are (possibly biased) Monte-Carlo approximations of the true risk measures.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • alpha (float) – The risk level, float in (0.0, 1.0].

  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the CVaR corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q-dim tensor of CVaR samples.

Return type:

Tensor

class botorch.acquisition.risk_measures.VaR(alpha, n_w, preprocessing_function=None)[source]

Bases: CVaR

The Value-at-Risk risk measure.

Value-at-Risk measures the smallest possible reward (or largest possible loss) after excluding the worst outcomes with a total probability of 1 - alpha. It is commonly used in financial risk management, and it corresponds to the 1 - alpha quantile of a given random variable.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • alpha (float) – The risk level, float in (0.0, 1.0].

  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the VaR corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q-dim tensor of VaR samples.

Return type:

Tensor

class botorch.acquisition.risk_measures.WorstCase(n_w, preprocessing_function=None)[source]

Bases: RiskMeasureMCObjective

The worst-case risk measure.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the worst-case measure corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q-dim tensor of worst-case samples.

Return type:

Tensor

class botorch.acquisition.risk_measures.Expectation(n_w, preprocessing_function=None)[source]

Bases: RiskMeasureMCObjective

The expectation risk measure.

For unconstrained problems, we recommend using the ExpectationPosteriorTransform instead. ExpectationPosteriorTransform directly transforms the posterior distribution over q * n_w to a posterior of q expectations, significantly reducing the cost of posterior sampling as a result.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the expectation corresponding to the given samples. This calculates the expectation / mean / average of each n_w samples across the q-batch dimension. If self.weights is given, the samples are scalarized across the output dimension before taking the expectation.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q-dim tensor of expectation samples.

Return type:

Tensor

Thompson Sampling

class botorch.acquisition.thompson_sampling.PathwiseThompsonSampling(model, posterior_transform=None)[source]

Bases: AcquisitionFunction

Single-outcome Thompson Sampling packaged as an (analytic) acquisition function. Querying the acquisition function gives the summed values of one or more draws from a pathwise drawn posterior sample, and thus it maximization yields one (or multiple) Thompson sample(s).

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> TS = PathwiseThompsonSampling(model)

Single-outcome TS.

Parameters:
  • model (Model) – A fitted GP model.

  • posterior_transform (PosteriorTransform | None) – A PosteriorTransform. If using a multi-output model, a PosteriorTransform that transforms the multi-output posterior into a single-output posterior is required.

redraw()[source]
Return type:

None

forward(X)[source]

Evaluate the pathwise posterior sample draws on the candidate set X.

Parameters:

X (Tensor) – A (b1 x … bk) x 1 x d-dim batched tensor of d-dim design points.

Returns:

A (b1 x … bk) x [num_models for fully bayesian]-dim tensor of evaluations on the posterior sample draws.

Return type:

Tensor

Multi-Output Risk Measures

Multi-output extensions of the risk measures, implemented as Monte-Carlo objectives. Except for MVaR, the risk measures are computed over each output dimension independently. In contrast, MVaR is computed using the joint distribution of the outputs, and provides more accurate risk estimates.

References

[Prekopa2012MVaR] (1,2,3,4)

A. Prekopa. Multivariate value at risk and related topics. Annals of Operations Research, 2012.

[Cousin2013MVaR] (1,2)

A. Cousin and E. Di Bernardino. On multivariate extensions of Value-at-Risk. Journal of Multivariate Analysis, 2013.

[Daulton2022MARS] (1,2,3)

S. Daulton, S, Cakmak, M. Balandat, M. Osborne, E. Zhou, and E. Bakshy. Robust multi-objective Bayesian optimization under input noise. Proceedings of the 39th International Conference on Machine Learning, 2022.

class botorch.acquisition.multi_objective.multi_output_risk_measures.MultiOutputRiskMeasureMCObjective(n_w, preprocessing_function=None)[source]

Bases: RiskMeasureMCObjective, MCMultiOutputObjective, ABC

Objective transforming the multi-output posterior samples to samples of a multi-output risk measure.

The risk measure is calculated over joint q-batch samples from the posterior. If the q-batch includes samples corresponding to multiple inputs, it is assumed that first n_w samples correspond to first input, second n_w samples correspond to second input, etc.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to remove non-objective outcomes or to align all outcomes for maximization. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch x m’-dim tensor.

abstract forward(samples, X=None)[source]

Calculate the risk measure corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of risk measure samples.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.MultiOutputExpectation(n_w, preprocessing_function=None)[source]

Bases: MultiOutputRiskMeasureMCObjective

A multi-output MC expectation risk measure.

For unconstrained problems, we recommend using the ExpectationPosteriorTransform instead. ExpectationPosteriorTransform directly transforms the posterior distribution over q * n_w to a posterior of q expectations, significantly reducing the cost of posterior sampling as a result.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to remove non-objective outcomes or to align all outcomes for maximization. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch x m’-dim tensor.

forward(samples, X=None)[source]

Calculate the expectation of the given samples. Expectation is calculated over each n_w samples in the q-batch dimension.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of expectation samples.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.IndependentCVaR(alpha, n_w, preprocessing_function=None)[source]

Bases: CVaR, MultiOutputRiskMeasureMCObjective

The multi-output Conditional Value-at-Risk risk measure that operates on each output dimension independently. Since this does not consider the joint distribution of the outputs (i.e., that the outputs were evaluated on same perturbed input and are not independent), the risk estimates provided by IndependentCVaR in general are more optimistic than the definition of CVaR would suggest.

The Conditional Value-at-Risk measures the expectation of the worst outcomes (small rewards or large losses) with a total probability of 1 - alpha. It is commonly defined as the conditional expectation of the reward function, with the condition that the reward is smaller than the corresponding Value-at-Risk (also defined below).

NOTE: Due to the use of a discrete w_set of samples, the VaR and CVaR calculated here are (possibly biased) Monte-Carlo approximations of the true risk measures.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • alpha (float) – The risk level, float in (0.0, 1.0].

  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the CVaR corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of CVaR samples.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.IndependentVaR(alpha, n_w, preprocessing_function=None)[source]

Bases: IndependentCVaR

The multi-output Value-at-Risk risk measure that operates on each output dimension independently. For the same reasons as IndependentCVaR, the risk estimates provided by this are in general more optimistic than the definition of VaR would suggest.

Value-at-Risk measures the smallest possible reward (or largest possible loss) after excluding the worst outcomes with a total probability of 1 - alpha. It is commonly used in financial risk management, and it corresponds to the 1 - alpha quantile of a given random variable.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • alpha (float) – The risk level, float in (0.0, 1.0].

  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to scalarize multi-output samples before calculating the risk measure. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch-dim tensor.

forward(samples, X=None)[source]

Calculate the VaR corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of VaR samples.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.MultiOutputWorstCase(n_w, preprocessing_function=None)[source]

Bases: MultiOutputRiskMeasureMCObjective

The multi-output worst-case risk measure.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to remove non-objective outcomes or to align all outcomes for maximization. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch x m’-dim tensor.

forward(samples, X=None)[source]

Calculate the worst-case measure corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of worst-case samples.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.MVaR(n_w, alpha, expectation=False, preprocessing_function=None, *, pad_to_n_w=False, filter_dominated=True, use_counting=False)[source]

Bases: MultiOutputRiskMeasureMCObjective

The multivariate Value-at-Risk as introduced in [Prekopa2012MVaR].

MVaR is defined as the non-dominated set of points in the extended domain of the random variable that have multivariate CDF greater than or equal to alpha. Note that MVaR is set valued and the size of the set depends on the particular realizations of the random variable. [Cousin2013MVaR] instead propose to use the expectation of the set-valued MVaR as the multivariate VaR. We support this alternative with an expectation flag.

This supports approximate gradients as discussed in [Daulton2022MARS].

The multivariate Value-at-Risk.

Parameters:
  • n_w (int) – The size of the w_set to calculate the risk measure over.

  • alpha (float) – The risk level of MVaR, float in (0.0, 1.0]. Each MVaR value dominates alpha fraction of all observations.

  • expectation (bool) – If True, returns the expectation of the MVaR set as is done in [Cousin2013MVaR]. Otherwise, it returns the union of all values in the MVaR set. Default: False.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to remove non-objective outcomes or to align all outcomes for maximization. For constrained optimization, this should also apply feasibility-weighting to samples. Given a batch x m-dim tensor of samples, this should return a batch x m’-dim tensor.

  • pad_to_n_w (bool) – If True, instead of padding up to k’, which is the size of the largest MVaR set across all batches, we pad the MVaR set up to n_w. This produces a return tensor of known size, however, it may in general be much larger than the alternative. See forward for more details on the return shape. NOTE: this is only relevant if expectation=False.

  • filter_dominated (bool) – If True, returns the non-dominated subset of alpha level points (this is MVaR as defined by [Prekopa2012MVaR]). Disabling this will make it faster, and may be preferable if the dominated points will be filtered out later, e.g., while calculating the hypervolume. Disabling this is not recommended if expectation=True.

  • use_counting (bool) – If True, uses get_mvar_set_via_counting for finding the MVaR set. This is method is less memory intensive than the vectorized implementation, which is beneficial when n_w is quite large.

get_mvar_set_via_counting(Y)[source]

Find MVaR set based on the definition in [Prekopa2012MVaR].

This first calculates the CDF for each point on the extended domain of the random variable (the grid defined by the given samples), then takes the values with CDF equal to (rounded if necessary) alpha. The non-dominated subset of these form the MVaR set.

This implementation processes each batch of Y in a for loop using a counting based implementation. It requires less memory than the vectorized implementation and should be used with large (>128) n_w values.

Parameters:

Y (Tensor) – A batch x n_w x m-dim tensor of outcomes.

Returns:

A batch length list of k x m-dim tensor of MVaR values, where k depends on the corresponding batch inputs. Note that MVaR values in general are not in-sample points.

Return type:

list[Tensor]

get_mvar_set_vectorized(Y)[source]

Find MVaR set based on the definition in [Prekopa2012MVaR].

This first calculates the CDF for each point on the extended domain of the random variable (the grid defined by the given samples), then takes the values with CDF equal to (rounded if necessary) alpha. The non-dominated subset of these form the MVaR set.

This implementation uses computes the CDF of each point using highly vectorized operations. As such, it may use large amounts of memory, particularly when the batch size and/or n_w are large. It is typically faster than the alternative implementation when computing MVaR of a large batch of points with small to moderate (<128 for m=2, <64 for m=3) n_w.

Parameters:

Y (Tensor) – A batch x n_w x m-dim tensor of observations.

Returns:

A batch length list of k x m-dim tensor of MVaR values, where k depends on the corresponding batch inputs. Note that MVaR values in general are not in-sample points.

Return type:

list[Tensor]

make_differentiable(prepared_samples, mvars)[source]

An experimental approach for obtaining the gradient of the MVaR via component-wise mapping to original samples. See [Daulton2022MARS].

Parameters:
  • prepared_samples (Tensor) – A (sample_shape * batch_shape * q) x n_w x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • mvars (Tensor) – A (sample_shape * batch_shape * q) x k x m-dim tensor of padded MVaR values.

Returns:

The same mvars with entries mapped to inputs to produce gradients.

Return type:

Tensor

forward(samples, X=None)[source]

Calculate the MVaR corresponding to the given samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x (q * n_w) x m-dim tensor of posterior samples. The q-batches should be ordered so that each n_w block of samples correspond to the same input.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Ignored.

Returns:

A sample_shape x batch_shape x q x m’-dim tensor of MVaR values, if self.expectation=True. Otherwise, this returns a sample_shape x batch_shape x (q * k’) x m’-dim tensor, where k’ is the maximum k across all batches that is returned by get_mvar_set_…. Each (q * k’) x m’ corresponds to the k MVaR values for each q batch of n_w inputs, padded up to k’ by repeating the last element. If self.pad_to_n_w, we set k’ = self.n_w, producing a deterministic return shape.

Return type:

Tensor

class botorch.acquisition.multi_objective.multi_output_risk_measures.MARS(alpha, n_w, chebyshev_weights, baseline_Y=None, ref_point=None, preprocessing_function=None)[source]

Bases: VaR, MultiOutputRiskMeasureMCObjective

MVaR Approximation based on Random Scalarizations as introduced in [Daulton2022MARS].

This approximates MVaR via VaR of Chebyshev scalarizations, where each scalarization corresponds to a point in the MVaR set. As implemented, this uses one set of scalarization weights to approximate a single MVaR value. Note that due to the normalization within the Chebyshev scalarization, the output of this risk measure may not be on the same scale as its inputs.

Transform the posterior samples to samples of a risk measure.

Parameters:
  • alpha (float) – The risk level, float in (0.0, 1.0].

  • n_w (int) – The size of the perturbation set to calculate the risk measure over.

  • chebyshev_weights (Tensor | list[float]) – The weights to use in the Chebyshev scalarization. The Chebyshev scalarization is applied before computing VaR. The weights must be non-negative. See preprocessing_function to support minimization objectives.

  • baseline_Y (Tensor | None) – An n’ x d-dim tensor of baseline outcomes to use in determining the normalization bounds for Chebyshev scalarization. It is recommended to set this via set_baseline_Y helper.

  • ref_point (Tensor | list[float] | None) – An optional MVaR reference point to use in determining the normalization bounds for Chebyshev scalarization.

  • preprocessing_function (Callable[[Tensor], Tensor] | None) – A preprocessing function to apply to the samples before computing the risk measure. This can be used to remove non-objective outcomes or to align all outcomes for maximization. For constrained optimization, this should also apply feasibility-weighting to samples.

set_baseline_Y(model, X_baseline, Y_samples=None)[source]

Set the baseline_Y based on the MVaR predictions of the model for X_baseline.

Parameters:
  • model (Model | None) – The model being used for MARS optimization. Must have a compatible InputPerturbation transform attached. Ignored if Y_samples is given.

  • X_baseline (Tensor | None) – An n x d-dim tensor of previously evaluated points. Ignored if Y_samples is given.

  • Y_samples (Tensor | None) – An optional (n * n_w) x d-dim tensor of predictions. If given, instead of sampling from the model, these are used.

Return type:

None

property chebyshev_weights: Tensor

The weights used in Chebyshev scalarization.

property baseline_Y: Tensor | None

Baseline outcomes used in determining the normalization bounds.

property chebyshev_objective: Callable[[Tensor, Tensor | None], Tensor]

The objective for applying the Chebyshev scalarization.

Utilities

Fixed Feature Acquisition Function

A wrapper around AcquisitionFunctions to fix certain features for optimization. This is useful e.g. for performing contextual optimization.

botorch.acquisition.fixed_feature.get_dtype_of_sequence(values)[source]

Return torch.float32 if everything is single-precision and torch.float64 otherwise.

Numbers (non-tensors) are double-precision.

Parameters:

values (Sequence[Tensor | float])

Return type:

dtype

botorch.acquisition.fixed_feature.get_device_of_sequence(values)[source]

CPU if everything is on the CPU; Cuda otherwise.

Numbers (non-tensors) are considered to be on the CPU.

Parameters:

values (Sequence[Tensor | float])

Return type:

dtype

class botorch.acquisition.fixed_feature.FixedFeatureAcquisitionFunction(acq_function, d, columns, values)[source]

Bases: AcquisitionFunction

A wrapper around AcquisitionFunctions to fix a subset of features.

Example

>>> model = SingleTaskGP(train_X, train_Y)  # d = 5
>>> qEI = qExpectedImprovement(model, best_f=0.0)
>>> columns = [2, 4]
>>> values = X[..., columns]
>>> qEI_FF = FixedFeatureAcquisitionFunction(qEI, 5, columns, values)
>>> qei = qEI_FF(test_X)  # d' = 3

Derived Acquisition Function by fixing a subset of input features.

Parameters:
  • acq_function (AcquisitionFunction) – The base acquisition function, operating on input tensors X_full of feature dimension d.

  • d (int) – The feature dimension expected by acq_function.

  • columns (list[int]) – d_f < d indices of columns in X_full that are to be fixed to the provided values.

  • values (Union[Tensor, Sequence[Union[Tensor, float]]]) – The values to which to fix the columns in columns. Either a full batch_shape x q x d_f tensor of values (if values are different for each of the q input points), or an array-like of values that is broadcastable to the input across t-batch and q-batch dimensions, e.g. a list of length d_f if values are the same across all t and q-batch dimensions, or a combination of Tensor`s and numbers which can be broadcasted to form a tensor with trailing dimension size of `d_f.

forward(X)[source]

Evaluate base acquisition function under the fixed features.

Parameters:

X (Tensor) – Input tensor of feature dimension d’ < d such that d’ + d_f = d.

Returns:

Base acquisition function evaluated on tensor X_full constructed by adding values in the appropriate places (see _construct_X_full).

property X_pending

Return the X_pending of the base acquisition function.

Constructors for Acquisition Function Input Arguments

A registry of helpers for generating inputs to acquisition function constructors programmatically from a consistent input format.

botorch.acquisition.input_constructors.get_acqf_input_constructor(acqf_cls)[source]

Get acquisition function input constructor from registry.

Parameters:

acqf_cls (type[AcquisitionFunction]) – The AcquisitionFunction class (not instance) for which to retrieve the input constructor.

Returns:

The input constructor associated with acqf_cls.

Return type:

Callable[[…], dict[str, Any]]

botorch.acquisition.input_constructors.allow_only_specific_variable_kwargs(f)[source]

Decorator for allowing a function to accept keyword arguments that are not explicitly listed in the function signature, but only specific ones.

This decorator is applied in acqf_input_constructor so that all constructors obtained with acqf_input_constructor allow keyword arguments such as training_data and objective, even if they do not appear in the signature of f. Any other keyword arguments will raise an error.

Parameters:

f (Callable[[...], T])

Return type:

Callable[[…], T]

botorch.acquisition.input_constructors.acqf_input_constructor(*acqf_cls)[source]

Decorator for registering acquisition function input constructors.

Parameters:

acqf_cls (type[AcquisitionFunction]) – The AcquisitionFunction classes (not instances) for which to register the input constructor.

Return type:

Callable[[…], AcquisitionFunction]

botorch.acquisition.input_constructors.construct_inputs_posterior_mean(model, posterior_transform=None)[source]

Construct kwargs for PosteriorMean acquisition function.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Model | PosteriorTransform | None]

botorch.acquisition.input_constructors.construct_inputs_best_f(model, training_data, posterior_transform=None, best_f=None, maximize=True)[source]

Construct kwargs for the acquisition functions requiring best_f.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model. Used to determine default value for best_f.

  • best_f (float | Tensor | None) – Threshold above (or below) which improvement is defined.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • maximize (bool) – If True, consider the problem a maximization problem.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_ucb(model, posterior_transform=None, beta=0.2, maximize=True)[source]

Construct kwargs for UpperConfidenceBound.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • beta (float | Tensor) – Either a scalar or a one-dim tensor with b elements (batch mode) representing the trade-off parameter between mean and covariance

  • maximize (bool) – If True, consider the problem a maximization problem.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_noisy_ei(model, training_data, num_fantasies=20, maximize=True)[source]

Construct kwargs for NoisyExpectedImprovement.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • num_fantasies (int) – The number of fantasies to generate. The higher this number the more accurate the model (at the expense of model complexity and performance).

  • maximize (bool) – If True, consider the problem a maximization problem.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qSimpleRegret(model, objective=None, posterior_transform=None, X_pending=None, sampler=None, constraints=None, X_baseline=None)[source]

Construct kwargs for qSimpleRegret.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A batch_shape, m x d-dim Tensor of m design points that have points that have been submitted for function evaluation but have not yet been evaluated.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • X_baseline (Tensor | None) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point. If omitted, checks that all training_data have the same input features and take the first X.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qEI(model, training_data, objective=None, posterior_transform=None, X_pending=None, sampler=None, best_f=None, constraints=None, eta=0.001)[source]

Construct kwargs for the qExpectedImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • best_f (float | Tensor | None) – Threshold above (or below) which improvement is defined.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qLogEI(model, training_data, objective=None, posterior_transform=None, X_pending=None, sampler=None, best_f=None, constraints=None, eta=0.001, fat=True, tau_max=0.01, tau_relu=1e-06)[source]

Construct kwargs for the qExpectedImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • best_f (float | Tensor | None) – Threshold above (or below) which improvement is defined.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • fat (bool) – Toggles the logarithmic / linear asymptotic behavior of the smooth approximation to the ReLU.

  • tau_max (float) – Temperature parameter controlling the sharpness of the smooth approximations to max.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the smooth approximations to ReLU.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qNEI(model, training_data, objective=None, posterior_transform=None, X_pending=None, sampler=None, X_baseline=None, prune_baseline=True, cache_root=True, constraints=None, eta=0.001)[source]

Construct kwargs for the qNoisyExpectedImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • X_baseline (Tensor | None) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point. If omitted, checks that all training_data have the same input features and take the first X.

  • prune_baseline (bool | None) – If True, remove points in X_baseline that are highly unlikely to be the best point. This can significantly improve performance and is generally recommended.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • cache_root (bool | None)

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qLogNEI(model, training_data, objective=None, posterior_transform=None, X_pending=None, sampler=None, X_baseline=None, prune_baseline=True, cache_root=True, constraints=None, eta=0.001, fat=True, tau_max=0.01, tau_relu=1e-06)[source]

Construct kwargs for the qLogNoisyExpectedImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • X_baseline (Tensor | None) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point. If omitted, checks that all training_data have the same input features and take the first X.

  • prune_baseline (bool | None) – If True, remove points in X_baseline that are highly unlikely to be the best point. This can significantly improve performance and is generally recommended.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • fat (bool) – Toggles the use of the fat-tailed non-linearities to smoothly approximate the constraints indicator function.

  • tau_max (float) – Temperature parameter controlling the sharpness of the smooth approximations to max.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the smooth approximations to ReLU.

  • cache_root (bool | None)

Returns:

A dict mapping kwarg names of the constructor to values.

botorch.acquisition.input_constructors.construct_inputs_qPI(model, training_data, objective=None, posterior_transform=None, X_pending=None, sampler=None, tau=0.001, best_f=None, constraints=None, eta=0.001)[source]

Construct kwargs for the qProbabilityOfImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • tau (float) – The temperature parameter used in the sigmoid approximation of the step function. Smaller values yield more accurate approximations of the function, but result in gradients estimates with higher variance.

  • best_f (float | Tensor | None) – The best objective value observed so far (assumed noiseless). Can be a batch_shape-shaped tensor, which in case of a batched model specifies potentially different values for each element of the batch.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qUCB(model, objective=None, posterior_transform=None, X_pending=None, sampler=None, X_baseline=None, constraints=None, beta=0.2)[source]

Construct kwargs for the qUpperConfidenceBound constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • objective (MCAcquisitionObjective | None) – The objective to be used in the acquisition function.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • X_baseline (Tensor | None) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are used to compute with infeasible cost when there are constraints.

  • constraints (List[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • beta (float) – Controls tradeoff between mean and standard deviation in UCB.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_EHVI(model, training_data, objective_thresholds, posterior_transform=None, constraints=None, alpha=None, Y_pmean=None)[source]

Construct kwargs for ExpectedHypervolumeImprovement constructor.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qEHVI(model, training_data, objective_thresholds, objective=None, constraints=None, alpha=None, sampler=None, X_pending=None, eta=0.001, mc_samples=128, qmc=True)[source]

Construct kwargs for qExpectedHypervolumeImprovement and qLogExpectedHypervolumeImprovement.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qNEHVI(model, training_data, objective_thresholds, objective=None, X_baseline=None, constraints=None, alpha=None, sampler=None, X_pending=None, eta=0.001, fat=False, mc_samples=128, qmc=True, prune_baseline=True, cache_pending=True, max_iep=0, incremental_nehvi=True, cache_root=True)[source]

Construct kwargs for qNoisyExpectedHypervolumeImprovement’s constructor.

Parameters:
  • model (Model)

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset])

  • objective_thresholds (Tensor)

  • objective (MCMultiOutputObjective | None)

  • X_baseline (Tensor | None)

  • constraints (list[Callable[[Tensor], Tensor]] | None)

  • alpha (float | None)

  • sampler (MCSampler | None)

  • X_pending (Tensor | None)

  • eta (float)

  • fat (bool)

  • mc_samples (int)

  • qmc (bool)

  • prune_baseline (bool)

  • cache_pending (bool)

  • max_iep (int)

  • incremental_nehvi (bool)

  • cache_root (bool)

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qLogNEHVI(model, training_data, objective_thresholds, objective=None, X_baseline=None, constraints=None, alpha=None, sampler=None, X_pending=None, eta=0.001, fat=True, mc_samples=128, qmc=True, prune_baseline=True, cache_pending=True, max_iep=0, incremental_nehvi=True, cache_root=True, tau_relu=1e-06, tau_max=0.01)[source]

Construct kwargs for qLogNoisyExpectedHypervolumeImprovement’s constructor.”

Parameters:
  • model (Model)

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset])

  • objective_thresholds (Tensor)

  • objective (MCMultiOutputObjective | None)

  • X_baseline (Tensor | None)

  • constraints (list[Callable[[Tensor], Tensor]] | None)

  • alpha (float | None)

  • sampler (MCSampler | None)

  • X_pending (Tensor | None)

  • eta (float)

  • fat (bool)

  • mc_samples (int)

  • qmc (bool)

  • prune_baseline (bool)

  • cache_pending (bool)

  • max_iep (int)

  • incremental_nehvi (bool)

  • cache_root (bool)

  • tau_relu (float)

  • tau_max (float)

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qLogNParEGO(model, training_data, scalarization_weights=None, objective=None, X_pending=None, sampler=None, X_baseline=None, prune_baseline=True, cache_root=True, constraints=None, eta=0.001, fat=True, tau_max=0.01, tau_relu=1e-06)[source]

Construct kwargs for the qLogNoisyExpectedImprovement constructor.

Parameters:
  • model (Model) – The model to be used in the acquisition function.

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Dataset(s) used to train the model.

  • scalarization_weights (Tensor | None) – A m-dim Tensor of weights to be used in the Chebyshev scalarization. If omitted, samples from the unit simplex.

  • objective (MCMultiOutputObjective | None) – The MultiOutputMCAcquisitionObjective under which the samples are evaluated before applying Chebyshev scalarization. Defaults to IdentityMultiOutputObjective().

  • X_pending (Tensor | None) – A m x d-dim Tensor of m design points that have been submitted for function evaluation but have not yet been evaluated. Concatenated into X upon forward call.

  • sampler (MCSampler | None) – The sampler used to draw base samples. If omitted, uses the acquisition functions’s default sampler.

  • X_baseline (Tensor | None) – A batch_shape x r x d-dim Tensor of r design points that have already been observed. These points are considered as the potential best design point. If omitted, checks that all training_data have the same input features and take the first X.

  • prune_baseline (bool | None) – If True, remove points in X_baseline that are highly unlikely to be the best point. This can significantly improve performance and is generally recommended.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are considered satisfied if the output is less than zero.

  • eta (Tensor | float) – Temperature parameter(s) governing the smoothness of the sigmoid approximation to the constraint indicators. For more details, on this parameter, see the docs of compute_smoothed_feasibility_indicator.

  • fat (bool) – Toggles the use of the fat-tailed non-linearities to smoothly approximate the constraints indicator function.

  • tau_max (float) – Temperature parameter controlling the sharpness of the smooth approximations to max.

  • tau_relu (float) – Temperature parameter controlling the sharpness of the smooth approximations to ReLU.

  • cache_root (bool | None)

Returns:

A dict mapping kwarg names of the constructor to values.

botorch.acquisition.input_constructors.construct_inputs_qMES(model, training_data, bounds, posterior_transform=None, candidate_size=1000, maximize=True)[source]

Construct kwargs for qMaxValueEntropy constructor.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_mf_base(target_fidelities, fidelity_weights=None, cost_intercept=1.0, num_trace_observations=0)[source]

Construct kwargs for a multifidelity acquisition function’s constructor.

Parameters:
  • target_fidelities (dict[int, int | float])

  • fidelity_weights (dict[int, float] | None)

  • cost_intercept (float)

  • num_trace_observations (int)

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qKG(model, training_data, bounds, objective=None, posterior_transform=None, num_fantasies=64, with_current_value=False, **optimize_objective_kwargs)[source]

Construct kwargs for qKnowledgeGradient constructor.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qHVKG(model, training_data, bounds, objective_thresholds, objective=None, posterior_transform=None, num_fantasies=8, num_pareto=10, **optimize_objective_kwargs)[source]

Construct kwargs for qKnowledgeGradient constructor.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qMFKG(model, training_data, bounds, target_fidelities, objective=None, posterior_transform=None, fidelity_weights=None, cost_intercept=1.0, num_trace_observations=0, num_fantasies=64, **optimize_objective_kwargs)[source]

Construct kwargs for qMultiFidelityKnowledgeGradient constructor.

Parameters:
Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qMFHVKG(model, training_data, bounds, target_fidelities, objective_thresholds, objective=None, posterior_transform=None, fidelity_weights=None, cost_intercept=1.0, num_trace_observations=0, num_fantasies=8, num_pareto=10, **optimize_objective_kwargs)[source]

Construct kwargs for qMultiFidelityHypervolumeKnowledgeGradient constructor.

Parameters:
  • model (Model)

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset])

  • bounds (list[tuple[float, float]])

  • target_fidelities (dict[int, int | float])

  • objective_thresholds (Tensor)

  • objective (MCMultiOutputObjective | None)

  • posterior_transform (PosteriorTransform | None)

  • fidelity_weights (dict[int, float] | None)

  • cost_intercept (float)

  • num_trace_observations (int)

  • num_fantasies (int)

  • num_pareto (int)

  • optimize_objective_kwargs (None | MCAcquisitionObjective | PosteriorTransform | tuple[Tensor, Tensor] | dict[int, float] | bool | int | dict[str, Any] | Callable[[Tensor], Tensor] | Tensor)

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qMFMES(model, training_data, bounds, target_fidelities, num_fantasies=64, fidelity_weights=None, cost_intercept=1.0, num_trace_observations=0, candidate_size=1000, maximize=True)[source]

Construct kwargs for qMultiFidelityMaxValueEntropy constructor.

Parameters:
  • model (Model)

  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset])

  • bounds (list[tuple[float, float]])

  • target_fidelities (dict[int, int | float])

  • num_fantasies (int)

  • fidelity_weights (dict[int, float] | None)

  • cost_intercept (float)

  • num_trace_observations (int)

  • candidate_size (int)

  • maximize (bool)

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_analytic_eubo(model, pref_model=None, previous_winner=None, sample_multiplier=1.0, objective=None, posterior_transform=None)[source]

Construct kwargs for the AnalyticExpectedUtilityOfBestOption constructor.

model is the primary model defined over the parameter space. It can be the outcome model in BOPE or the preference model in PBO. pref_model is the model defined over the outcome/metric space, which is typically the preference model in BOPE.

If both model and pref_model exist, we are performing Bayesian Optimization with Preference Exploration (BOPE). When only pref_model is None, we are performing preferential BO (PBO).

Parameters:
  • model (Model) – The outcome model to be used in the acquisition function in BOPE when pref_model exists; otherwise, model is the preference model and we are doing Preferential BO

  • pref_model (Model | None) – The preference model to be used in preference exploration as in BOPE; if None, we are doing PBO and model is the preference model.

  • previous_winner (Tensor | None) – The previous winner of the best option.

  • sample_multiplier (float | None) – The scale factor for the single-sample model.

  • objective (LearnedObjective | None) – Ignored. This argument is allowed to be passed then ignored because of the way that EUBO is typically used in a BOPE loop.

  • posterior_transform (PosteriorTransform | None) – Ignored. This argument is allowed to be passed then ignored because of the way that EUBO is typically used in a BOPE loop.

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.construct_inputs_qeubo(model, pref_model=None, sample_multiplier=1.0, sampler=None, objective=None, posterior_transform=None, X_pending=None)[source]

Construct kwargs for the qExpectedUtilityOfBestOption (qEUBO) constructor.

model is the primary model defined over the parameter space. It can be the outcomde model in BOPE or the preference model in PBO. pref_model is the model defined over the outcome/metric space, which is typically the preference model in BOPE.

If both model and pref_model exist, we are performing Bayesian Optimization with Preference Exploration (BOPE). When only pref_model is None, we are performing preferential BO (PBO).

Parameters:
  • model (Model) – The outcome model to be used in the acquisition function in BOPE when pref_model exists; otherwise, model is the preference model and we are doing Preferential BO

  • pref_model (Model | None) – The preference model to be used in preference exploration as in BOPE; if None, we are doing PBO and model is the preference model.

  • sample_multiplier (float | None) – The scale factor for the single-sample model.

  • sampler (MCSampler | None)

  • objective (MCAcquisitionObjective | None)

  • posterior_transform (PosteriorTransform | None)

  • X_pending (Tensor | None)

Returns:

A dict mapping kwarg names of the constructor to values.

Return type:

dict[str, Any]

botorch.acquisition.input_constructors.get_best_f_analytic(training_data, posterior_transform=None)[source]
Parameters:
Return type:

Tensor

botorch.acquisition.input_constructors.get_best_f_mc(training_data, objective=None, posterior_transform=None, constraints=None, model=None)[source]

Computes the maximum value of the objective over the training data.

Parameters:
  • training_data (SupervisedDataset | dict[Hashable, SupervisedDataset]) – Has fields Y, which is evaluated by objective, and X, which is used as X_baseline. Y is of shape batch_shape x q x m.

  • objective (MCAcquisitionObjective | None) – The objective under which to evaluate the training data. If omitted, uses IdentityMCObjective.

  • posterior_transform (PosteriorTransform | None) – An optional PosteriorTransform to apply to Y before computing the objective.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – For assessing feasibility.

  • model (Model | None) – Used by compute_best_feasible_objective when there are no feasible observations.

Returns:

A Tensor of shape batch_shape.

Return type:

Tensor

botorch.acquisition.input_constructors.optimize_objective(model, bounds, q, acq_function=None, objective=None, posterior_transform=None, linear_constraints=None, fixed_features=None, qmc=True, mc_samples=512, seed_inner=None, optimizer_options=None, post_processing_func=None, batch_initial_conditions=None, sequential=False)[source]

Optimize an objective under the given model.

Parameters:
  • model (Model) – The model to be used in the objective.

  • bounds (Tensor) – A 2 x d tensor of lower and upper bounds for each column of X.

  • q (int) – The cardinality of input sets on which the objective is to be evaluated.

  • objective (MCAcquisitionObjective | None) – The objective to optimize.

  • posterior_transform (PosteriorTransform | None) – The posterior transform to be used in the acquisition function.

  • linear_constraints (tuple[Tensor, Tensor] | None) – A tuple of (A, b). Given k linear constraints on a d-dimensional space, A is k x d and b is k x 1 such that A x <= b. (Not used by single task models).

  • fixed_features (dict[int, float] | None) – A dictionary of feature assignments {feature_index: value} to hold fixed during generation.

  • qmc (bool) – Toggle for enabling (qmc=1) or disabling (qmc=0) use of Quasi Monte Carlo.

  • mc_samples (int) – Integer number of samples used to estimate Monte Carlo objectives.

  • seed_inner (int | None) – Integer seed used to initialize the sampler passed to MCObjective.

  • optimizer_options (dict[str, Any] | None) – Table used to lookup keyword arguments for the optimizer.

  • post_processing_func (Callable[[Tensor], Tensor] | None) – A function that post-processes an optimization result appropriately (i.e. according to round-trip transformations).

  • batch_initial_conditions (Tensor | None) – A Tensor of initial values for the optimizer.

  • sequential (bool) – If False, uses joint optimization, otherwise uses sequential optimization.

  • acq_function (AcquisitionFunction | None)

Returns:

A tuple containing the best input locations and corresponding objective values.

Return type:

tuple[Tensor, Tensor]

botorch.acquisition.input_constructors.construct_inputs_qJES(model, bounds, num_optima=64, maximize=True, condition_noiseless=True, X_pending=None, estimation_type='LB', num_samples=64)[source]
Parameters:
  • model (Model)

  • bounds (list[tuple[float, float]])

  • num_optima (int)

  • maximize (bool)

  • condition_noiseless (bool)

  • X_pending (Tensor | None)

  • estimation_type (str)

  • num_samples (int)

botorch.acquisition.input_constructors.construct_inputs_BALD(model, X_pending=None, sampler=None, posterior_transform=None)[source]
Parameters:
botorch.acquisition.input_constructors.construct_inputs_NIPV(model, bounds, num_mc_points=128, X_pending=None, posterior_transform=None)[source]

Construct inputs for qNegIntegratedPosteriorVariance.

Parameters:
  • model (Model)

  • bounds (list[tuple[float, float]])

  • num_mc_points (int)

  • X_pending (Tensor | None)

  • posterior_transform (PosteriorTransform | None)

Return type:

dict[str, Any]

Penalized Acquisition Function Wrapper

Modules to add regularization to acquisition functions.

class botorch.acquisition.penalized.L2Penalty(init_point)[source]

Bases: Module

L2 penalty class to be added to any arbitrary acquisition function to construct a PenalizedAcquisitionFunction.

Initializing L2 regularization.

Parameters:

init_point (Tensor) – The “1 x dim” reference point against which we want to regularize.

forward(X)[source]
Parameters:

X (Tensor) – A “batch_shape x q x dim” representing the points to be evaluated.

Returns:

A tensor of size “batch_shape” representing the acqfn for each q-batch.

Return type:

Tensor

class botorch.acquisition.penalized.L1Penalty(init_point)[source]

Bases: Module

L1 penalty class to be added to any arbitrary acquisition function to construct a PenalizedAcquisitionFunction.

Initializing L1 regularization.

Parameters:

init_point (Tensor) – The “1 x dim” reference point against which we want to regularize.

forward(X)[source]
Parameters:

X (Tensor) – A “batch_shape x q x dim” representing the points to be evaluated.

Returns:

A tensor of size “batch_shape” representing the acqfn for each q-batch.

Return type:

Tensor

class botorch.acquisition.penalized.GaussianPenalty(init_point, sigma)[source]

Bases: Module

Gaussian penalty class to be added to any arbitrary acquisition function to construct a PenalizedAcquisitionFunction.

Initializing Gaussian regularization.

Parameters:
  • init_point (Tensor) – The “1 x dim” reference point against which we want to regularize.

  • sigma (float) – The parameter used in gaussian function.

forward(X)[source]
Parameters:

X (Tensor) – A “batch_shape x q x dim” representing the points to be evaluated.

Returns:

A tensor of size “batch_shape” representing the acqfn for each q-batch.

Return type:

Tensor

class botorch.acquisition.penalized.GroupLassoPenalty(init_point, groups)[source]

Bases: Module

Group lasso penalty class to be added to any arbitrary acquisition function to construct a PenalizedAcquisitionFunction.

Initializing Group-Lasso regularization.

Parameters:
  • init_point (Tensor) – The “1 x dim” reference point against which we want to regularize.

  • groups (list[list[int]]) – Groups of indices used in group lasso.

forward(X)[source]

X should be batch_shape x 1 x dim tensor. Evaluation for q-batch is not implemented yet.

Parameters:

X (Tensor)

Return type:

Tensor

botorch.acquisition.penalized.narrow_gaussian(X, a)[source]
Parameters:
  • X (Tensor)

  • a (Tensor)

Return type:

Tensor

botorch.acquisition.penalized.nnz_approx(X, target_point, a)[source]

Differentiable relaxation of ||X - target_point||_0

Parameters:
  • X (Tensor) – An n x d tensor of inputs.

  • target_point (Tensor) – A tensor of size n corresponding to the target point.

  • a (Tensor) – A scalar tensor that controls the differentiable relaxation.

Return type:

Tensor

class botorch.acquisition.penalized.L0Approximation(target_point, a=1.0, **tkwargs)[source]

Bases: Module

Differentiable relaxation of the L0 norm using a Gaussian basis function.

Initializing L0 penalty with differentiable relaxation.

Parameters:
  • target_point (Tensor) – A tensor corresponding to the target point.

  • a (float) – A hyperparameter that controls the differentiable relaxation.

  • tkwargs (Any)

class botorch.acquisition.penalized.L0PenaltyApprox(target_point, a=1.0, **tkwargs)[source]

Bases: L0Approximation

Differentiable relaxation of the L0 norm to be added to any arbitrary acquisition function to construct a PenalizedAcquisitionFunction.

Initializing L0 penalty with differentiable relaxation.

Parameters:
  • target_point (Tensor) – A tensor corresponding to the target point.

  • a (float) – A hyperparameter that controls the differentiable relaxation.

  • tkwargs (Any)

class botorch.acquisition.penalized.PenalizedAcquisitionFunction(raw_acqf, penalty_func, regularization_parameter)[source]

Bases: AcquisitionFunction

Single-outcome acquisition function regularized by the given penalty.

The usage is similar to:

raw_acqf = NoisyExpectedImprovement(…) penalty = GroupLassoPenalty(…) acqf = PenalizedAcquisitionFunction(raw_acqf, penalty)

Initializing Group-Lasso regularization.

Parameters:
  • raw_acqf (AcquisitionFunction) – The raw acquisition function that is going to be regularized.

  • penalty_func (torch.nn.Module) – The regularization function.

  • regularization_parameter (float) – Regularization parameter used in optimization.

forward(X)[source]

Evaluate the acquisition function on the candidate set X.

Parameters:

X (Tensor) – A (b) x q x d-dim Tensor of (b) t-batches with q d-dim design points each.

Returns:

A (b)-dim Tensor of acquisition function values at the given design points X.

Return type:

Tensor

property X_pending: Tensor | None
set_X_pending(X_pending=None)[source]

Informs the acquisition function about pending design points.

Parameters:

X_pending (Tensor | None) – n x d Tensor with n d-dim design points that have been submitted for evaluation but have not yet been evaluated.

Return type:

None

botorch.acquisition.penalized.group_lasso_regularizer(X, groups)[source]

Computes the group lasso regularization function for the given point.

Parameters:
  • X (Tensor) – A bxd tensor representing the points to evaluate the regularization at.

  • groups (list[list[int]]) – List of indices of different groups.

Returns:

Computed group lasso norm of at the given points.

Return type:

Tensor

class botorch.acquisition.penalized.L1PenaltyObjective(init_point)[source]

Bases: Module

L1 penalty objective class. An instance of this class can be added to any arbitrary objective to construct a PenalizedMCObjective.

Initializing L1 penalty objective.

Parameters:

init_point (Tensor) – The “1 x dim” reference point against which we want to regularize.

forward(X)[source]
Parameters:

X (Tensor) – A “batch_shape x q x dim” representing the points to be evaluated.

Returns:

A “1 x batch_shape x q” tensor representing the penalty for each point. The first dimension corresponds to the dimension of MC samples.

Return type:

Tensor

class botorch.acquisition.penalized.PenalizedMCObjective(objective, penalty_objective, regularization_parameter, expand_dim=None)[source]

Bases: GenericMCObjective

Penalized MC objective.

Allows to construct a penalized MC-objective by adding a penalty term to the original objective.

mc_acq(X) = objective(X) + penalty_objective(X)

Note: PenalizedMCObjective allows adding penalty at the MCObjective level, different from the AcquisitionFunction level in PenalizedAcquisitionFunction.

Example

>>> regularization_parameter = 0.01
>>> init_point = torch.zeros(3) # assume data dim is 3
>>> objective = lambda Y, X: torch.sqrt(Y).sum(dim=-1)
>>> l1_penalty_objective = L1PenaltyObjective(init_point=init_point)
>>> l1_penalized_objective = PenalizedMCObjective(
        objective, l1_penalty_objective, regularization_parameter
    )
>>> samples = sampler(posterior)
        objective, l1_penalty_objective, regularization_parameter

Penalized MC objective.

Parameters:
  • objective (Callable[[Tensor, Optional[Tensor]], Tensor]) – A callable f(samples, X) mapping a sample_shape x batch-shape x q x m-dim Tensor samples and an optional batch-shape x q x d-dim Tensor X to a sample_shape x batch-shape x q-dim Tensor of objective values.

  • penalty_objective (torch.nn.Module) – A torch.nn.Module f(X) that takes in a batch-shape x q x d-dim Tensor X and outputs a 1 x batch-shape x q-dim Tensor of penalty objective values.

  • regularization_parameter (float) – weight of the penalty (regularization) term

  • expand_dim (Optional[int]) – dim to expand penalty_objective to match with objective when fully bayesian model is used. If None, no expansion is performed.

forward(samples, X=None)[source]

Evaluate the penalized objective on the samples.

Parameters:
  • samples (Tensor) – A sample_shape x batch_shape x q x m-dim Tensors of samples from a model posterior.

  • X (Tensor | None) – A batch_shape x q x d-dim tensor of inputs. Relevant only if the objective depends on the inputs explicitly.

Returns:

A sample_shape x batch_shape x q-dim Tensor of objective values with penalty added for each point.

Return type:

Tensor

class botorch.acquisition.penalized.L0PenaltyApproxObjective(target_point, a=1.0, **tkwargs)[source]

Bases: L0Approximation

Differentiable relaxation of the L0 norm penalty objective class. An instance of this class can be added to any arbitrary objective to construct a PenalizedMCObjective.

Initializing L0 penalty with differentiable relaxation.

Parameters:
  • target_point (Tensor) – A tensor corresponding to the target point.

  • a (float) – A hyperparameter that controls the differentiable relaxation.

  • tkwargs (Any)

Prior-Guided Acquisition Function Wrapper

Prior-Guided Acquisition Functions

References

[Hvarfner2022] (1,2)

C. Hvarfner, D. Stoll, A. Souza, M. Lindauer, F. Hutter, L. Nardi. PiBO: Augmenting Acquisition Functions with User Beliefs for Bayesian Optimization. ICLR 2022.

class botorch.acquisition.prior_guided.PriorGuidedAcquisitionFunction(acq_function, prior_module, log=False, prior_exponent=1.0, X_pending=None)[source]

Bases: AcquisitionFunction

Class for weighting acquisition functions by a prior distribution.

Supports MC and batch acquisition functions via SampleReducingAcquisitionFunction.

See [Hvarfner2022] for details.

Initialize the prior-guided acquisition function.

Parameters:
  • acq_function (AcquisitionFunction) – The base acquisition function.

  • prior_module (Module) – A Module that computes the probability (or log probability) for the provided inputs. prior_module.forward should take a batch_shape x q-dim tensor of inputs and return a batch_shape x q-dim tensor of probabilities.

  • log (bool) – A boolean that should be true if the acquisition function emits a log-transformed value and the prior module emits a log probability.

  • prior_exponent (float) – The exponent applied to the prior. This can be used for example to decay the effect the prior over time as in [Hvarfner2022].

  • X_pending (Optional[Tensor]) – n x d Tensor with n d-dim design points that have been submitted for evaluation but have not yet been evaluated. Note: X_pending should be provided as an argument to or set on PriorGuidedAcquisitionFunction, but not set on the underlying acquisition function.

forward(X)[source]

Compute the acquisition function weighted by the prior.

Parameters:

X (Tensor)

Return type:

Tensor

Proximal Acquisition Function Wrapper

A wrapper around AcquisitionFunctions to add proximal weighting of the acquisition function.

class botorch.acquisition.proximal.ProximalAcquisitionFunction(acq_function, proximal_weights, transformed_weighting=True, beta=None)[source]

Bases: AcquisitionFunction

A wrapper around AcquisitionFunctions to add proximal weighting of the acquisition function. The acquisition function is weighted via a squared exponential centered at the last training point, with varying lengthscales corresponding to proximal_weights. Can only be used with acquisition functions based on single batch models. Acquisition functions must be positive or beta must be specified to apply a SoftPlus transform before proximal weighting.

Small values of proximal_weights corresponds to strong biasing towards recently observed points, which smoothes optimization with a small potential decrese in convergence rate.

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> EI = ExpectedImprovement(model, best_f=0.0)
>>> proximal_weights = torch.ones(d)
>>> EI_proximal = ProximalAcquisitionFunction(EI, proximal_weights)
>>> eip = EI_proximal(test_X)

Derived Acquisition Function weighted by proximity to recently observed point.

Parameters:
  • acq_function (AcquisitionFunction) – The base acquisition function, operating on input tensors of feature dimension d.

  • proximal_weights (Tensor) – A d dim tensor used to bias locality along each axis.

  • transformed_weighting (Optional[bool]) – If True, the proximal weights are applied in the transformed input space given by acq_function.model.input_transform (if available), otherwise proximal weights are applied in real input space.

  • beta (Optional[float]) – If not None, apply a softplus transform to the base acquisition function, allows negative base acquisition function values.

forward(X)[source]

Evaluate base acquisition function with proximal weighting.

Parameters:

X (Tensor) – Input tensor of feature dimension d .

Returns:

Base acquisition function evaluated on tensor X multiplied by proximal weighting.

Return type:

Tensor

Factory Functions for Acquisition Functions

Utilities for acquisition functions.

botorch.acquisition.factory.get_acquisition_function(acquisition_function_name, model, objective, X_observed, posterior_transform=None, X_pending=None, constraints=None, eta=0.001, mc_samples=512, seed=None, *, tau=0.001, prune_baseline=True, marginalize_dim=None, cache_root=True, beta=None, ref_point=None, Y=None, alpha=0.0)[source]

Convenience function for initializing botorch acquisition functions.

Parameters:
  • acquisition_function_name (str) – Name of the acquisition function.

  • model (Model) – A fitted model.

  • objective (MCAcquisitionObjective) – A MCAcquisitionObjective.

  • X_observed (Tensor) – A m1 x d-dim Tensor of m1 design points that have already been observed.

  • posterior_transform (PosteriorTransform | None) – A PosteriorTransform (optional).

  • X_pending (Tensor | None) – A m2 x d-dim Tensor of m2 design points whose evaluation is pending.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility. Used for all acquisition functions except qSR and qUCB.

  • eta (float | Tensor | None) – The temperature parameter for the sigmoid function used for the differentiable approximation of the constraints. In case of a float the same eta is used for every constraint in constraints. In case of a tensor the length of the tensor must match the number of provided constraints. The i-th constraint is then estimated with the i-th eta value. Used for all acquisition functions except qSR and qUCB.

  • mc_samples (int) – The number of samples to use for (q)MC evaluation of the acquisition function.

  • seed (int | None) – If provided, perform deterministic optimization (i.e. the function to optimize is fixed and not stochastic).

  • tau (float)

  • prune_baseline (bool)

  • marginalize_dim (int | None)

  • cache_root (bool)

  • beta (float | None)

  • ref_point (Tensor | list[float] | None)

  • Y (Tensor | None)

  • alpha (float)

Returns:

The requested acquisition function.

Return type:

MCAcquisitionFunction

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> obj = LinearMCObjective(weights=torch.tensor([1.0, 2.0]))
>>> acqf = get_acquisition_function("qEI", model, obj, train_X)

General Utilities for Acquisition Functions

Utilities for acquisition functions.

botorch.acquisition.utils.get_acquisition_function(*args, **kwargs)[source]
Return type:

None

botorch.acquisition.utils.repeat_to_match_aug_dim(target_tensor, reference_tensor)[source]

Repeat target_tensor until it has the same first dimension as reference_tensor This works regardless of the batch shapes and q. This is useful as we sometimes modify sample shapes such as in LearnedObjective.

Parameters:
  • target_tensor (Tensor) – A sample_size x batch_shape x q x m-dim Tensor

  • reference_tensor (Tensor) – A (augmented_sample * sample_size) x batch_shape x q-dim Tensor. augmented_sample could be 1.

Returns:

The content of target_tensor potentially repeated so that its first dimension matches that of reference_tensor. The shape will be (augmented_sample * sample_size) x batch_shape x q x m.

Return type:

Tensor

Examples

>>> import torch
>>> target_tensor = torch.arange(3).repeat(2, 1).T
>>> target_tensor
tensor([[0, 0],
        [1, 1],
        [2, 2]])
>>> repeat_to_match_aug_dim(target_tensor, torch.zeros(6))
tensor([[0, 0],
        [1, 1],
        [2, 2],
        [0, 0],
        [1, 1],
        [2, 2]])
botorch.acquisition.utils.compute_best_feasible_objective(samples, obj, constraints, model=None, objective=None, posterior_transform=None, X_baseline=None, infeasible_obj=None)[source]

Computes the largest obj value that is feasible under the constraints. If constraints is None, returns the best unconstrained objective value.

When no feasible observations exist and infeasible_obj is not None, returns infeasible_obj (potentially reshaped). When no feasible observations exist and infeasible_obj is None, uses model, objective, posterior_transform, and X_baseline to infer and return an infeasible_obj M s.t. M < min_x f(x).

Parameters:
  • samples (Tensor) – (sample_shape) x batch_shape x q x m-dim posterior samples.

  • obj (Tensor) – A (sample_shape) x batch_shape x q-dim Tensor of MC objective values.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map posterior samples to a scalar. The associated constraint is considered satisfied if this scalar is less than zero.

  • model (Model | None) – A Model, only required when there are no feasible observations.

  • objective (MCAcquisitionObjective | None) – An MCAcquisitionObjective, only optionally used when there are no feasible observations.

  • posterior_transform (PosteriorTransform | None) – A PosteriorTransform, only optionally used when there are no feasible observations.

  • X_baseline (Tensor | None) – A batch_shape x d-dim Tensor of baseline points, only required when there are no feasible observations.

  • infeasible_obj (Tensor | None) – A Tensor to be returned when no feasible points exist.

Returns:

A (sample_shape) x batch_shape-dim Tensor of best feasible objectives.

Return type:

Tensor

botorch.acquisition.utils.get_infeasible_cost(X, model, objective=None, posterior_transform=None)[source]

Get infeasible cost for a model and objective.

For each outcome, computes an infeasible cost M such that -M < min_x f(x) almost always, so that feasible points are preferred.

Parameters:
  • X (Tensor) – A n x d Tensor of n design points to use in evaluating the minimum. These points should cover the design space well. The more points the better the estimate, at the expense of added computation.

  • model (Model) – A fitted botorch model with m outcomes.

  • objective (Callable[[Tensor, Tensor | None], Tensor] | None) – The objective with which to evaluate the model output.

  • posterior_transform (PosteriorTransform | None) – A PosteriorTransform (optional).

Returns:

An m-dim tensor of infeasible cost values.

Return type:

Tensor

Example

>>> model = SingleTaskGP(train_X, train_Y)
>>> objective = lambda Y: Y[..., -1] ** 2
>>> M = get_infeasible_cost(train_X, model, obj)
botorch.acquisition.utils.prune_inferior_points(model, X, objective=None, posterior_transform=None, constraints=None, num_samples=2048, max_frac=1.0, sampler=None, marginalize_dim=None)[source]

Prune points from an input tensor that are unlikely to be the best point.

Given a model, an objective, and an input tensor X, this function returns the subset of points in X that have some probability of being the best point under the objective. This function uses sampling to estimate the probabilities, the higher the number of points n in X the higher the number of samples num_samples should be to obtain accurate estimates.

Parameters:
  • model (Model) – A fitted model. Batched models are currently not supported.

  • X (Tensor) – An input tensor of shape n x d. Batched inputs are currently not supported.

  • objective (MCAcquisitionObjective | None) – The objective under which to evaluate the posterior.

  • posterior_transform (PosteriorTransform | None) – A PosteriorTransform (optional).

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of constraint callables which map a Tensor of posterior samples of dimension sample_shape x batch-shape x q x m-dim to a sample_shape x batch-shape x q-dim Tensor. The associated constraints are satisfied if constraint(samples) < 0.

  • num_samples (int) – The number of samples used to compute empirical probabilities of being the best point.

  • max_frac (float) – The maximum fraction of points to retain. Must satisfy 0 < max_frac <= 1. Ensures that the number of elements in the returned tensor does not exceed ceil(max_frac * n).

  • sampler (MCSampler | None) – If provided, will use this customized sampler instead of automatically constructing one with num_samples.

  • marginalize_dim (int | None) – A batch dimension that should be marginalized. For example, this is useful when using a batched fully Bayesian model.

Returns:

A n’ x d with subset of points in X, where

n’ = min(N_nz, ceil(max_frac * n))

with N_nz the number of points in X that have non-zero (empirical, under num_samples samples) probability of being the best point.

Return type:

Tensor

botorch.acquisition.utils.project_to_target_fidelity(X, target_fidelities=None, d=None)[source]

Project X onto the target set of fidelities.

This function assumes that the set of feasible fidelities is a box, so projecting here just means setting each fidelity parameter to its target value. If X does not contain the fidelity dimensions, this will insert them and set them to their target values.

Parameters:
  • X (Tensor) – A batch_shape x q x (d or d-d_f)-dim Tensor of with q d or d-d_f-dim design points for each t-batch, where d_f is the number of fidelity dimensions. If the argument d is not provided, X must include the fidelity dimensions and have a trailing`X` must include the fidelity dimensions and have a trailing

  • target_fidelities (dict[int, float] | None) – A dictionary mapping a subset of columns of X (the fidelity parameters) to their respective target fidelity value. If omitted, assumes that the last column of X is the fidelity parameter with a target value of 1.0.

  • d (int | None) – The total dimension d.

Returns:

A batch_shape x q x d-dim Tensor X_proj with fidelity parameters

projected to the provided fidelity values.

Return type:

Tensor

botorch.acquisition.utils.expand_trace_observations(X, fidelity_dims=None, num_trace_obs=0)[source]

Expand X with trace observations.

Expand a tensor of inputs with “trace observations” that are obtained during the evaluation of the candidate set. This is used in multi-fidelity optimization. It can be though of as augmenting the q-batch with additional points that are the expected trace observations.

Let f_i be the i-th fidelity parameter. Then this functions assumes that for each element of the q-batch, besides the fidelity f_i, we will observe additonal fidelities f_i1, …, f_iK, where K = num_trace_obs, during evaluation of the candidate set X. Specifically, this function assumes that f_ij = (K-j) / (num_trace_obs + 1) * f_i for all i. That is, the expansion is performed in parallel for all fidelities (it does not expand out all possible combinations).

Parameters:
  • X (Tensor) – A batch_shape x q x d-dim Tensor of with q d-dim design points (incl. the fidelity parameters) for each t-batch.

  • fidelity_dims (list[int] | None) – The indices of the fidelity parameters. If omitted, assumes that the last column of X contains the fidelity parameters.

  • num_trace_obs (int) – The number of trace observations to use.

Returns:

A batch_shape x (q + num_trace_obs x q) x d Tensor X_expanded that

expands X with trace observations.

Return type:

Tensor

botorch.acquisition.utils.project_to_sample_points(X, sample_points)[source]

Augment X with sample points at which to take weighted average.

Parameters:
  • X (Tensor) – A batch_shape x 1 x d-dim Tensor of with one d`-dim design points for each t-batch.

  • sample_points (Tensor) – p x d’-dim Tensor (d’ < d) of d’-dim sample points at which to compute the expectation. The d’-dims refer to the trailing columns of X.

Returns:

A batch_shape x p x d Tensor where the q-batch includes the p sample points.

Return type:

Tensor

botorch.acquisition.utils.get_optimal_samples(model, bounds, num_optima, raw_samples=1024, num_restarts=20, maximize=True)[source]

Draws sample paths from the posterior and maximizes the samples using GD.

Parameters:
  • model (Model) – The model from which samples are drawn.

  • bounds (Tensor) – (Tensor): Bounds of the search space. If the model inputs are normalized, the bounds should be normalized as well.

  • num_optima (int) – The number of paths to be drawn and optimized.

  • raw_samples (int, optional) – The number of candidates randomly sample. Defaults to 1024.

  • num_restarts (int, optional) – The number of candidates to do gradient-based optimization on. Defaults to 20.

  • maximize (bool) – Whether to maximize or minimize the samples.

Returns:

The optimal input locations and corresponding outputs, x* and f*.

Return type:

Tuple[Tensor, Tensor]

Multi-Objective Utilities for Acquisition Functions

Utilities for multi-objective acquisition functions.

botorch.acquisition.multi_objective.utils.get_default_partitioning_alpha(num_objectives)[source]

Determines an approximation level based on the number of objectives.

If alpha is 0, FastNondominatedPartitioning should be used. Otherwise, an approximate NondominatedPartitioning should be used with approximation level alpha.

Parameters:

num_objectives (int) – the number of objectives.

Returns:

The approximation level alpha.

Return type:

float

botorch.acquisition.multi_objective.utils.prune_inferior_points_multi_objective(model, X, ref_point, objective=None, constraints=None, num_samples=2048, max_frac=1.0, marginalize_dim=None)[source]

Prune points from an input tensor that are unlikely to be pareto optimal.

Given a model, an objective, and an input tensor X, this function returns the subset of points in X that have some probability of being pareto optimal, better than the reference point, and feasible. This function uses sampling to estimate the probabilities, the higher the number of points n in X the higher the number of samples num_samples should be to obtain accurate estimates.

Parameters:
  • model (Model) – A fitted model. Batched models are currently not supported.

  • X (Tensor) – An input tensor of shape n x d. Batched inputs are currently not supported.

  • ref_point (Tensor) – The reference point.

  • objective (MCMultiOutputObjective | None) – The objective under which to evaluate the posterior.

  • constraints (list[Callable[[Tensor], Tensor]] | None) – A list of callables, each mapping a Tensor of dimension sample_shape x batch-shape x q x m to a Tensor of dimension sample_shape x batch-shape x q, where negative values imply feasibility.

  • num_samples (int) – The number of samples used to compute empirical probabilities of being the best point.

  • max_frac (float) – The maximum fraction of points to retain. Must satisfy 0 < max_frac <= 1. Ensures that the number of elements in the returned tensor does not exceed ceil(max_frac * n).

  • marginalize_dim (int | None) – A batch dimension that should be marginalized. For example, this is useful when using a batched fully Bayesian model.

Returns:

A n’ x d with subset of points in X, where

n’ = min(N_nz, ceil(max_frac * n))

with N_nz the number of points in X that have non-zero (empirical, under num_samples samples) probability of being pareto optimal.

Return type:

Tensor

botorch.acquisition.multi_objective.utils.compute_sample_box_decomposition(pareto_fronts, partitioning=<class 'botorch.utils.multi_objective.box_decompositions.dominated.DominatedPartitioning'>, maximize=True, num_constraints=0)[source]

Computes the box decomposition associated with some sampled optimal objectives. This also supports the single-objective and constrained optimization setting. An objective y is feasible if y <= 0.

To take advantage of batch computations, we pad the hypercell bounds with a 2 x (M + K)-dim Tensor of zeros [0, 0].

Parameters:
  • pareto_fronts (Tensor) – A num_pareto_samples x num_pareto_points x M dim Tensor containing the sampled optimal set of objectives.

  • partitioning (BoxDecomposition) – A BoxDecomposition module that is used to obtain the hyper-rectangle bounds for integration. In the unconstrained case, this gives the partition of the dominated space. In the constrained case, this gives the partition of the feasible dominated space union the infeasible space.

  • maximize (bool) – If true, the box-decomposition is computed assuming maximization.

  • num_constraints (int | None) – The number of constraints K.

Returns:

A num_pareto_samples x 2 x J x (M + K)-dim Tensor containing the bounds for the hyper-rectangles. The number J is the smallest number of boxes needed to partition all the Pareto samples.

Return type:

Tensor

botorch.acquisition.multi_objective.utils.random_search_optimizer(model, bounds, num_points, maximize, pop_size=1024, max_tries=10)[source]

Optimize a function via random search.

Parameters:
  • model (GenericDeterministicModel) – The model.

  • bounds (Tensor) – A 2 x d-dim Tensor containing the input bounds.

  • num_points (int) – The number of optimal points to be outputted.

  • maximize (bool) – If true, we consider a maximization problem.

  • pop_size (int) – The number of function evaluations per try.

  • max_tries (int) – The maximum number of tries.

Returns:

A two-element tuple containing

  • A num_points x d-dim Tensor containing the collection of optimal inputs.

  • A num_points x M-dim Tensor containing the collection of optimal

    objectives.

Return type:

tuple[Tensor, Tensor]

botorch.acquisition.multi_objective.utils.sample_optimal_points(model, bounds, num_samples, num_points, optimizer=<function random_search_optimizer>, maximize=True, optimizer_kwargs=None)[source]

Compute a collection of optimal inputs and outputs from samples of a Gaussian Process (GP).

Steps: (1) The samples are generated using random Fourier features (RFFs). (2) The samples are optimized sequentially using an optimizer.

TODO: We can generalize the GP sampling step to accommodate for other sampling

strategies rather than restricting to RFFs e.g. decoupled sampling.

TODO: Currently this defaults to random search optimization, might want to

explore some other alternatives.

Parameters:
  • model (Model) – The model. This does not support models which include fantasy observations.

  • bounds (Tensor) – A 2 x d-dim Tensor containing the input bounds.

  • num_samples (int) – The number of GP samples.

  • num_points (int) – The number of optimal points to be outputted.

  • optimizer (Callable[[GenericDeterministicModel, Tensor, int, bool, Any], tuple[Tensor, Tensor]]) – A callable that solves the deterministic optimization problem.

  • maximize (bool) – If true, we consider a maximization problem.

  • optimizer_kwargs (dict[str, Any] | None) – The additional arguments for the optimizer.

Returns:

A two-element tuple containing

  • A num_samples x num_points x d-dim Tensor containing the collection of

    optimal inputs.

  • A num_samples x num_points x M-dim Tensor containing the collection of

    optimal objectives.

Return type:

tuple[Tensor, Tensor]