Spidr.BayesianOptimization.Acquisition

Acquisition functionality for Bayesian optimization. Acquisition functions quantify the
usefulness of points in a search domain, towards the task of finding a function optimum.

Definitions

record DataModel : (modelType : Type) -> ProbabilisticModel f t marginal modelType => Type

  A `DataModel` packages data with a model over that data.

Totality: total
Visibility: public export
Constructor: 

MkDataModel : modelType -> Dataset f t -> DataModel modelType

Projections:

.dataset : DataModel modelType -> Dataset f t

  The data the model is trained on

.model : DataModel modelType -> modelType

  A probabilistic model

.model : DataModel modelType -> modelType

  A probabilistic model

Totality: total
Visibility: public export

.dataset : DataModel modelType -> Dataset f t

  The data the model is trained on

Totality: total
Visibility: public export

0 Acquisition : (0 batchSize : Nat) -> {auto 0 _ : GT batchSize 0} -> (0 _ : Shape) -> Type

  An `Acquisition` function quantifies how useful it would be to query the objective at a given
  set of points, towards the goal of optimizing the objective.
  
  @batchSize The number of points in the feature domain that the `Acquisition` evaluates
    at once.
  @features The shape of the feature domain.

Totality: total
Visibility: public export

expectedImprovement : ProbabilisticModel features [1] Gaussian m => m -> Tensor [] F64 -> Acquisition 1 features

  Construct the acquisition function that estimates the absolute improvement in the best
  observation if we were to evaluate the objective at a given point.
  
  @model The model over the historic data.
  @best The current best observation.

Totality: total
Visibility: export

expectedImprovementByModel : {auto {conArg:7335} : ProbabilisticModel features [1] Gaussian modelType} -> ReaderT (DataModel modelType) Tag (Acquisition 1 features)

  Build an acquisition function that returns the absolute improvement, expected by the model, in
  the observation value at each point.

Totality: total
Visibility: export

probabilityOfFeasibility : Tensor [] F64 -> ClosedFormDistribution [1] dist => {auto {conArg:7516} : ProbabilisticModel features [1] dist modelType} -> ReaderT (DataModel modelType) Tag (Acquisition 1 features)

  Build an acquisition function that returns the probability that any given point will take a
  value less than the specified `limit`.

Totality: total
Visibility: export

negativeLowerConfidenceBound : (beta : Double) -> {auto 0 _ : beta >= 0.0 = True} -> {auto {conArg:7637} : ProbabilisticModel features [1] Gaussian modelType} -> ReaderT (DataModel modelType) Tag (Acquisition 1 features)

  Build an acquisition function that returns the negative of the lower confidence bound of the
  probabilistic model. The variance contribution is weighted by a factor `beta`.
  
  @beta The weighting given to the variance contribution.

Totality: total
Visibility: export

expectedConstrainedImprovement : Tensor [] F64 -> {auto {conArg:7802} : ProbabilisticModel features [1] Gaussian modelType} -> ReaderT (DataModel modelType) Tag (Acquisition 1 features -> Acquisition 1 features)

  Build the expected improvement acquisition function in the context of a constraint on the input
  domain, where points that do not satisfy the constraint do not offer an improvement. The
  complete acquisition function is built from a constraint acquisition function, which quantifies
  whether specified points in the input space satisfy the constraint.
  
  **NOTE** This function is not yet implemented.