Bayesian optimisation

Module that contains the different acquisition function classes for use in Bayesian optimisation. Both expected improvement and upper confidence bound are available

class topsearch.potentials.bayesian_optimisation.ExpectedImprovement(gaussian_process: GaussianProcess, zeta: float)

Description

Evaluate the expected improvement function to determine the utility in picking a particular point for a next experiment in BayesOpt. The function is constructed from a given Gaussian process that provides. Defined assuming maximisation of a dataset

gaussian_process

The gaussian process fit from which we build the acquisition function

Type:

object

zeta

Parameter in the acquisition functions that controls exploration vs exploitation

Type:

float

function(position: NDArray[Any, Any]) → float

Return the expected improvement at position

class topsearch.potentials.bayesian_optimisation.UpperConfidenceBound(gaussian_process: GaussianProcess, zeta: float)

Description

Evaluate the upper confidence bound, which gives the utility in picking a given point as the next experiment in Bayesian optimisation. The function is constructed from a given Gaussian process that provides both mean and variance. Defined assuming minimisation of a dataset

gaussian_process

The gaussian process fit from which we build the acquisition function

Type:

object

zeta

Parameter in the acquisition functions that controls exploration vs exploitation

Type:

float

function(position: NDArray[Any, Any]) → float

Return the value of the acquisition function