piglot.optimisers.botorch package

Submodules

piglot.optimisers.botorch.bayes module

Main optimiser classes for using BoTorch with piglot

class BayesianBoTorch(objective: Objective, n_initial: int | None = None, n_test: int = 0, acquisition: str | None = None, beta: float = 1.0, noisy: float = False, q: int = 1, seed: int = 1, load_file: str | None = None, export: str | None = None, device: str | None = None, reference_point: List[float] | None = None, nadir_scale: float = 0.1, skip_initial: bool = False, pca_variance: float | None = None, num_restarts: int | None = None, raw_samples: int | None = None, mc_samples: int | None = None, batch_size: int | None = None, num_fantasies: int | None = None, sequential: bool = False)

Bases: Optimiser

Driver for optimisation using BoTorch.

default_acquisition(composite: bool, multi_objective: bool, noisy: bool, q: int) → str

Return the default acquisition function for the given optimisation problem.

Parameters

compositebool, optional

Whether the optimisation problem is a composition.

multi_objectivebool, optional

Whether the optimisation problem is multi-objective.

noisybool, optional

Whether the optimisation problem is noisy.

qint, optional

Number of candidates to generate.

Returns

str

Name of the default acquisition function.

fit_mll_pytorch_loop(mll: ExactMarginalLogLikelihood, n_iters: int = 100) → None

Fit a GP model using a PyTorch optimisation loop.

Parameters

mllExactMarginalLogLikelihood

Marginal log-likelihood to optimise.

n_itersint, optional

Number of iterations to optimise for, by default 100

get_default_torch_device() → str

Utility to return the default PyTorch device (pre Pytorch v2.3).

Returns

str

Name of the default PyTorch device.

piglot.optimisers.botorch.dataset module

Dataset classes for optimising with BoTorch.

class BayesDataset(n_dim: int, n_outputs: int, export: str | None = None, dtype: dtype = torch.float64, std_tol: float = 1e-06, pca_variance: float | None = None, device: str = 'cpu')

Bases: object

Dataset class for multi-outcome data.

classmethod load(filename: str) → T

Load data from a given input file.

Parameters

filenamestr

Path to the file to read from.

Returns

BayesDataset

Dataset loaded from the file.

min() → Tuple[ndarray, ndarray]

Return the minimum objective value of the dataset.

Returns

Tuple[np.ndarray, np.ndarray]

Parameters and objective value for the minimum point.

numel_latent_space() → int

Return the number of components of the latent space.

Returns

int

Number of components of the latent space.

push(params: ndarray, results: ndarray, covariance: ndarray, objective: float | None) → None

Add a point to the dataset.

Parameters

paramsnp.ndarray

Parameter values for this observation.

resultObjectiveResult

Result for this observation.

save(filename: str) → None

Save all dataset data to a file.

Parameters

filenamestr

Output file path.

to(device: str) → BayesDataset

Move the dataset to a given device.

Parameters

devicestr

Device to move the dataset to.

Returns

BayesDataset

The dataset in the new device.

transform_outcomes(values: Tensor | None = None, covariances: Tensor | None = None, diagonalise: bool = True) → Tuple[Tensor, Tensor]

Transform outcomes to the latent standardised space.

Parameters

valuestorch.Tensor

Values to transform.

covariancestorch.Tensor

Variances to transform.

diagonalisebool

Whether to diagonalise the covariance matrix (default: True).

Returns

Tuple[torch.Tensor, torch.Tensor]

Transformed values and variances.

untransform_outcomes(values: Tensor) → Tensor

Transform outcomes back to the original space.

Parameters

valuestorch.Tensor

Values to transform.

Returns

torch.Tensor

Transformed values.

update_stats() → None

Update the statistics of the dataset.

class PCA(variance: float)

Bases: object

Principal Component Analysis transformation.

fit(data: Tensor, covariances: Tensor) → None

Fit the PCA transformation to the given data.

Parameters

datatorch.Tensor

Data to fit the PCA transformation to.

covariancestorch.Tensor

Covariances of the data to fit the PCA transformation to.

to(device: str) → PCA

Move the PCA to a given device.

Parameters

devicestr

Device to move the PCA to.

Returns

PCA

The PCA in the new device.

transform(values: torch.Tensor, covariances: torch.Tensor | None = None) → torch.Tensor | Tuple[torch.Tensor, torch.Tensor]

Transform data to the latent space.

Parameters

valuestorch.Tensor

Values to transform.

covariancestorch.Tensor | None

Variances to transform, if any.

Returns

torch.Tensor | Tuple[torch.Tensor, torch.Tensor]

Transformed values and covariances (if any).

untransform(data: Tensor) → Tensor

Transform data back to the original space.

Parameters

datatorch.Tensor

Data to untransform.

Returns

torch.Tensor

Untransformed data.

class Standardiser(std_tol: float = 1e-06)

Bases: object

Standardisation transformation.

fit(data: Tensor) → None

Fit the standardisation transformation to the given data.

Parameters

datatorch.Tensor

Data to fit the standardisation transformation to.

to(device: str) → Standardiser

Move the standardiser to a given device.

Parameters

devicestr

Device to move the standardiser to.

Returns

Standardiser

The standardiser in the new device.

transform(values: torch.Tensor, covariances: torch.Tensor | None = None) → torch.Tensor | Tuple[torch.Tensor, torch.Tensor]

Standardise data.

Parameters

valuestorch.Tensor

Values to transform.

covariancestorch.Tensor | None

Variances to transform, if any.

Returns

torch.Tensor | Tuple[torch.Tensor, torch.Tensor]

Transformed values and covariances (if any).

untransform(data: Tensor) → Tensor

Unstandardise data.

Parameters

datatorch.Tensor

Data to unstandardise.

Returns

torch.Tensor

Unstandardised data.

Module contents