ConditionedGaussianProcess

class ConditionedGaussianProcess(*args)

Conditioned Gaussian process.

ConditionedGaussianProcess(krigingResult, mesh)

Parameters:
krigingResultKrigingResult

Structure that contains all elements of kriging computations.

meshMesh

Mesh \cM over which the domain \cD is discretized.

Notes

ConditionedGaussianProcess helps to create Gaussian random fields, X: \Omega \times\cD \mapsto \Rset^d where \cD \in \Rset^n, with covariance function \cC: \cD \times \cD \mapsto \cM_{d \times d}(\Rset) ( \cC^{stat}: \cD \mapsto \cM_{d \times d}(\Rset) in the stationary case), conditionally to some observations.

Let X(\omega,x=x_1)=y_1,\cdots,X(\omega,x=x_n)=y_n be the observations of the Gaussian process. We assume the same Gaussian prior as in the KrigingAlgorithm:

Y(\vect{x}) = \Tr{\vect{f}(\vect{x})} \vect{\beta} + Z(\vect{x})

with \Tr{\vect{f}(\vect{x})} \vect{\beta} a general linear model, Z(\vect{x}) a zero-mean Gaussian process with a stationary autocorrelation function \cC^{stat}:

\mathbb{E}[Z(\vect{x}), Z(\vect{\tilde{x}})] = \sigma^2 \cC^{stat}_{\theta}(\vect{x} - \vect{\tilde{x}})

The ConditionedGaussianProcess generates realizations of the conditioned process. It focuses first on the KrigingAlgorithm to build such prior. Results are stored in a KrigingResult structure, which is given as input argument of the class. This last one, combined with the mesh argument, define both the prior Y(\cM) and the covariance evaluation on the mesh vertices \cC^{stat}_{\theta}(\cM) conditionally to the previous observations. It follows that the realizations are randomly generated from the Gaussian distribution \cN ( Y(\cM), \cC^{stat}_{\theta}(\cM) ).

In practice, we do not store the Gaussian distribution as we need only the random realization method. For that purpose, we use the Cholesky method : we compute the Cholesky factor \cL_{\theta}(\cM) of the covariance matrix \cC^{stat}_{\theta}(\cM) such as \cC^{stat}_{\theta}(\cM) = \cL_{\theta}(\cM) \Tr{\cL_{\theta}(\cM)}. It follows that the random realizations are obtained as following : realization = Y(\cM) + \cL_{\theta}(\cM) W with W a centered & reduced random Gaussian realization.

Examples

>>> import openturns as ot
>>> ot.RandomGenerator.SetSeed(0)
>>> # Kriging use case
>>> # Learning data
>>> levels = [8.0, 5.0]
>>> box = ot.Box(levels)
>>> inputSample = box.generate()
>>> # Scale each direction
>>> inputSample *= 10
>>> # Define model
>>> model = ot.SymbolicFunction(['x', 'y'], ['cos(0.5*x) + sin(y)'])
>>> outputSample = model(inputSample)
>>> # Definition of exponential model
>>> inputDimension = 2
>>> covarianceModel = ot.SquaredExponential([1.988, 0.924], [3.153])
>>> # Basis definition
>>> basis = ot.ConstantBasisFactory(inputDimension).build()
>>> # Kriring algorithm
>>> algo = ot.KrigingAlgorithm(inputSample, outputSample, covarianceModel, basis)
>>> algo.run()
>>> result = algo.getResult()
>>> vertices = [[1.0, 0.0], [2.0, 0.0], [2.0, 1.0], [1.0, 1.0], [1.5, 0.5]]
>>> simplices = [[0, 1, 4], [1, 2, 4], [2, 3, 4], [3, 0, 4]]
>>> mesh2D = ot.Mesh(vertices, simplices)
>>> process = ot.ConditionedGaussianProcess(result, mesh2D)

Methods

getClassName()

Accessor to the object's name.

getContinuousRealization()

Get a continuous realization.

getCovarianceModel()

Get the covariance model.

getDescription()

Get the description of the process.

getFuture(*args)

Prediction of the N future iterations of the process.

getInputDimension()

Get the dimension of the domain \cD.

getMarginal(indices)

Get the k^{th} marginal of the random process.

getMesh()

Get the mesh.

getName()

Accessor to the object's name.

getOutputDimension()

Get the dimension of the domain \cD.

getRealization()

Return a realization of the process.

getSample(size)

Get n realizations of the process.

getSamplingMethod()

Get the used method for getRealization.

getTimeGrid()

Get the time grid of observation of the process.

getTrend()

Get the trend function.

hasName()

Test if the object is named.

isComposite()

Test whether the process is composite or not.

isNormal()

Test whether the process is normal or not.

isStationary()

Test whether the process is stationary or not.

isTrendStationary()

Tell if the process is trend stationary or not.

setDescription(description)

Set the description of the process.

setMesh(mesh)

Set the mesh.

setName(name)

Accessor to the object's name.

setSamplingMethod(*args)

Set the used method for getRealization.

setTimeGrid(timeGrid)

Set the time grid of observation of the process.

__init__(*args)
getClassName()

Accessor to the object’s name.

Returns:
class_namestr

The object class name (object.__class__.__name__).

getContinuousRealization()

Get a continuous realization.

Returns:
realizationFunction

According to the process, the continuous realizations are built:

  • either using a dedicated functional model if it exists: e.g. a functional basis process.

  • or using an interpolation from a discrete realization of the process on \cM: in dimension d=1, a linear interpolation and in dimension d \geq 2, a piecewise constant function (the value at a given position is equal to the value at the nearest vertex of the mesh of the process).

getCovarianceModel()

Get the covariance model.

Returns:
covarianceModelCovarianceModel

Temporal covariance model C.

getDescription()

Get the description of the process.

Returns:
descriptionDescription

Description of the process.

getFuture(*args)

Prediction of the N future iterations of the process.

Parameters:
stepNumberint, N \geq 0

Number of future steps.

sizeint, size \geq 0, optional

Number of futures needed. Default is 1.

Returns:
predictionProcessSample or TimeSeries

N future iterations of the process. If size = 1, prediction is a TimeSeries. Otherwise, it is a ProcessSample.

getInputDimension()

Get the dimension of the domain \cD.

Returns:
nint

Dimension of the domain \cD: n.

getMarginal(indices)

Get the k^{th} marginal of the random process.

Parameters:
kint or list of ints 0 \leq k < d

Index of the marginal(s) needed.

Returns:
marginalsProcess

Process defined with marginal(s) of the random process.

getMesh()

Get the mesh.

Returns:
meshMesh

Mesh over which the domain \cD is discretized.

getName()

Accessor to the object’s name.

Returns:
namestr

The name of the object.

getOutputDimension()

Get the dimension of the domain \cD.

Returns:
dint

Dimension of the domain \cD.

getRealization()

Return a realization of the process.

Returns:
realizationField

A realization of the process.

Examples

>>> import openturns as ot
>>> ot.RandomGenerator.SetSeed(0)
>>> # Kriging use case
>>> # Learning data
>>> levels = [8.0, 5.0]
>>> box = ot.Box(levels)
>>> inputSample = box.generate()
>>> # Scale each direction
>>> inputSample *= 10
>>> # Define model
>>> model = ot.SymbolicFunction(['x', 'y'], ['cos(0.5*x) + sin(y)'])
>>> outputSample = model(inputSample)
>>> # Definition of exponential model
>>> inputDimension = 2
>>> covarianceModel = ot.SquaredExponential(inputDimension *[0.95])
>>> # Basis definition
>>> basis = ot.ConstantBasisFactory(inputDimension).build()
>>> # Kriring algorithm
>>> algo = ot.KrigingAlgorithm(inputSample, outputSample, covarianceModel, basis)
>>> algo.run()
>>> result = algo.getResult()
>>> vertices = [[1.0, 0.0], [2.0, 0.0], [2.0, 1.0],[1.0, 1.0], [1.5, 0.5]]
>>> simplices = [[0, 1, 4], [1, 2, 4], [2, 3, 4], [3, 0, 4]]
>>> mesh2D = ot.Mesh(vertices, simplices)
>>> process = ot.ConditionedGaussianProcess(result, mesh2D)
>>> # Get a realization of the process
>>> realization = process.getRealization()
getSample(size)

Get n realizations of the process.

Parameters:
nint, n \geq 0

Number of realizations of the process needed.

Returns:
processSampleProcessSample

n realizations of the random process. A process sample is a collection of fields which share the same mesh \cM \in \Rset^n.

getSamplingMethod()

Get the used method for getRealization.

Returns:
samplingMethodint

Used method for sampling.

Notes

Available parameters are :

  • 0 (GaussianProcess.CHOLESKY) : Cholesky factor sampling (default method)

  • 1 (GaussianProcess.HMAT) : H-Matrix method (if H-Mat available)

  • 2 (GaussianProcess.GALLIGAOGIBBS) : Gibbs method (in dimension 1 only)

getTimeGrid()

Get the time grid of observation of the process.

Returns:
timeGridRegularGrid

Time grid of a process when the mesh associated to the process can be interpreted as a RegularGrid. We check if the vertices of the mesh are scalar and are regularly spaced in \Rset but we don’t check if the connectivity of the mesh is conform to the one of a regular grid (without any hole and composed of ordered instants).

getTrend()

Get the trend function.

Returns:
trendTrendTransform

Trend function.

hasName()

Test if the object is named.

Returns:
hasNamebool

True if the name is not empty.

isComposite()

Test whether the process is composite or not.

Returns:
isCompositebool

True if the process is composite (built upon a function and a process).

isNormal()

Test whether the process is normal or not.

Returns:
isNormalbool

True if the process is normal.

Notes

A stochastic process is normal if all its finite dimensional joint distributions are normal, which means that for all k \in \Nset and I_k \in \Nset^*, with cardI_k=k, there is \vect{m}_1, \dots, \vect{m}_k \in \Rset^d and \mat{C}_{1,\dots,k}\in\mathcal{M}_{kd,kd}(\Rset) such that:

\Expect{\exp\left\{i\Tr{\vect{X}}_{I_k} \vect{U}_{k}  \right\}} =
\exp{\left\{i\Tr{\vect{U}}_{k}\vect{M}_{k}-\frac{1}{2}\Tr{\vect{U}}_{k}\mat{C}_{1,\dots,k}\vect{U}_{k}\right\}}

where \Tr{\vect{X}}_{I_k} = (\Tr{X}_{\vect{t}_1}, \hdots, \Tr{X}_{\vect{t}_k}), \\Tr{vect{U}}_{k} = (\Tr{\vect{u}}_{1}, \hdots, \Tr{\vect{u}}_{k}) and \Tr{\vect{M}}_{k} = (\Tr{\vect{m}}_{1}, \hdots, \Tr{\vect{m}}_{k}) and \mat{C}_{1,\dots,k} is the symmetric matrix:

\mat{C}_{1,\dots,k} = \left(
\begin{array}{cccc}
  C(\vect{t}_1, \vect{t}_1) &C(\vect{t}_1, \vect{t}_2) & \hdots & C(\vect{t}_1, \vect{t}_{k}) \\
  \hdots & C(\vect{t}_2, \vect{t}_2)  & \hdots & C(\vect{t}_2, \vect{t}_{k}) \\
  \hdots & \hdots & \hdots & \hdots \\
  \hdots & \hdots & \hdots & C(\vect{t}_{k}, \vect{t}_{k})
\end{array}
\right)

A Gaussian process is entirely defined by its mean function m and its covariance function C (or correlation function R).

isStationary()

Test whether the process is stationary or not.

Returns:
isStationarybool

True if the process is stationary.

Notes

A process X is stationary if its distribution is invariant by translation: \forall k \in \Nset, \forall (\vect{t}_1, \dots, \vect{t}_k) \in \cD, \forall \vect{h}\in \Rset^n, we have:

(X_{\vect{t}_1}, \dots, X_{\vect{t}_k})
\stackrel{\mathcal{D}}{=} (X_{\vect{t}_1+\vect{h}}, \dots, X_{\vect{t}_k+\vect{h}})

isTrendStationary()

Tell if the process is trend stationary or not.

Returns:
isTrendStationarybool

True if the process is trend stationary.

setDescription(description)

Set the description of the process.

Parameters:
descriptionsequence of str

Description of the process.

setMesh(mesh)

Set the mesh.

Parameters:
meshMesh

Mesh over which the domain \cD is discretized.

setName(name)

Accessor to the object’s name.

Parameters:
namestr

The name of the object.

setSamplingMethod(*args)

Set the used method for getRealization.

Available parameters are :

  • 0 (GaussianProcess.CHOLESKY) : Cholesky factor sampling (default method)

  • 1 (GaussianProcess.HMAT) : H-Matrix method (if H-Mat available)

  • 2 (GaussianProcess.GALLIGAOGIBBS) : Gibbs method (in dimension 1 only)

Parameters:
samplingMethodint

Fix a method for sampling.

setTimeGrid(timeGrid)

Set the time grid of observation of the process.

Returns:
timeGridRegularGrid

Time grid of observation of the process when the mesh associated to the process can be interpreted as a RegularGrid. We check if the vertices of the mesh are scalar and are regularly spaced in \Rset but we don’t check if the connectivity of the mesh is conform to the one of a regular grid (without any hole and composed of ordered instants).

Examples using the class

Kriging : generate trajectories from a metamodel

Kriging : generate trajectories from a metamodel