Covariance function for multi-output Gaussian Processes based on the linear model of coregionalization (LMC) More...

#include <LMC.h>

Inheritance diagram for LMC:

Public Types
using	HyperParameterMap = std::unordered_map< std::string, torch::Tensor >

typedef DataFileName	DataFileParameterType

Public Member Functions
	LMC (const InputParameters &parameters)

void	computeCovarianceMatrix (torch::Tensor &K, const torch::Tensor &x, const torch::Tensor &xp, const bool is_self_covariance) const override
	Generates the Covariance Matrix given two sets of points in the parameter space.

bool	computedKdhyper (torch::Tensor &dKdhp, const torch::Tensor &x, const std::string &hyper_param_name, unsigned int ind) const override
	Redirect dK/dhp for hyperparameter "hp".

void	loadHyperParamMap (const HyperParameterMap &map)
	Load some hyperparameters into the local map contained in this object.

void	buildHyperParamMap (HyperParameterMap &map) const
	Populates the input maps with the owned hyperparameters.

virtual bool	getTuningData (const std::string &name, unsigned int &size, Real &min, Real &max) const
	Get the default minimum and maximum and size of a hyperparameter.

void	dependentCovarianceTypes (std::map< UserObjectName, std::string > &name_type_map) const
	Populate a map with the names and types of the dependent covariance functions.

const std::vector< UserObjectName > &	dependentCovarianceNames () const
	Get the names of the dependent covariances.

virtual bool	isTunable (const std::string &name) const
	Check if a given parameter is tunable.

unsigned int	numOutputs () const
	Return the number of outputs assumed for this covariance function.

virtual bool	enabled () const

std::shared_ptr< MooseObject >	getSharedPtr ()

std::shared_ptr< const MooseObject >	getSharedPtr () const

bool	isKokkosObject () const

MooseApp &	getMooseApp () const

const std::string &	type () const

const std::string &	name () const

std::string	typeAndName () const

MooseObjectParameterName	uniqueParameterName (const std::string &parameter_name) const

MooseObjectName	uniqueName () const

const InputParameters &	parameters () const

const hit::Node *	getHitNode () const

bool	hasBase () const

const std::string &	getBase () const

const T &	getParam (const std::string &name) const

std::vector< std::pair< T1, T2 > >	getParam (const std::string &param1, const std::string &param2) const

const T *	queryParam (const std::string &name) const

const T &	getRenamedParam (const std::string &old_name, const std::string &new_name) const

T	getCheckedPointerParam (const std::string &name, const std::string &error_string="") const

bool	isParamValid (const std::string &name) const

bool	isParamSetByUser (const std::string &name) const

void	connectControllableParams (const std::string &parameter, const std::string &object_type, const std::string &object_name, const std::string &object_parameter) const

void	paramError (const std::string &param, Args... args) const

void	paramWarning (const std::string &param, Args... args) const

void	paramWarning (const std::string &param, Args... args) const

void	paramInfo (const std::string &param, Args... args) const

std::string	messagePrefix (const bool hit_prefix=true) const

std::string	errorPrefix (const std::string &) const

void	mooseError (Args &&... args) const

void	mooseDocumentedError (const std::string &repo_name, const unsigned int issue_num, Args &&... args) const

void	mooseErrorNonPrefixed (Args &&... args) const

void	mooseWarning (Args &&... args) const

void	mooseWarning (Args &&... args) const

void	mooseWarningNonPrefixed (Args &&... args) const

void	mooseWarningNonPrefixed (Args &&... args) const

void	mooseDeprecated (Args &&... args) const

void	mooseDeprecated (Args &&... args) const

void	mooseDeprecatedNoTrace (Args &&... args) const

void	mooseInfo (Args &&... args) const

void	callMooseError (std::string msg, const bool with_prefix, const hit::Node *node=nullptr, const bool show_trace=true) const

std::string	getDataFileName (const std::string &param) const

std::string	getDataFileNameByName (const std::string &relative_path) const

std::string	getDataFilePath (const std::string &relative_path) const

const Parallel::Communicator &	comm () const

processor_id_type	n_processors () const

processor_id_type	processor_id () const

Static Public Member Functions
static InputParameters	validParams ()

static bool	isScalarHyperParameter (const torch::Tensor &tensor)
	Return true if a hyperparameter tensor stores one scalar value.

static bool	isVectorHyperParameter (const torch::Tensor &tensor)
	Return true if a hyperparameter tensor stores a vector of values.

static void	callMooseError (MooseApp const app, const InputParameters &params, std::string msg, const bool with_prefix, const hit::Node node, const bool show_trace=true)

Public Attributes
	usingCombinedWarningSolutionWarnings

const ConsoleStream	_console

Static Public Attributes
static const std::string	type_param

static const std::string	name_param

static const std::string	unique_name_param

static const std::string	app_param

static const std::string	moose_base_param

static const std::string	kokkos_object_param

Protected Member Functions
void	computeBMatrix (torch::Tensor &Bmat, const unsigned int exp_i) const
	Computes the covariance matrix for the outputs (using the latent coefficients) We use a $B = \sum_i a_i a_i^T + diag(lambda_i)$ expansion here where $a_i$ and $lambda_i$ are vectors.

void	computeAGradient (torch::Tensor &grad, const unsigned int exp_i, const unsigned int index) const
	Computes the gradient of $B$ with respect to the entries in $a_i$ in the following expression: $B = \sum_i a_i a_i^T + diag(lambda_i)$.

void	computeLambdaGradient (torch::Tensor &grad, const unsigned int exp_i, const unsigned int index) const
	Computes the gradient of $B$ with respect to the entries in $lambda_i$ in the following expression: $B = \sum_i a_i a_i^T + diag(lambda_i)$.

torch::Tensor &	addRealHyperParameter (const std::string &name, const Real value, const bool is_tunable)
	Register a scalar hyperparameter to this covariance function.

torch::Tensor &	addVectorRealHyperParameter (const std::string &name, const std::vector< Real > &value, const bool is_tunable)
	Register a vector hyperparameter to this covariance function.

void	flagInvalidSolutionInternal (const InvalidSolutionID invalid_solution_id) const

InvalidSolutionID	registerInvalidSolutionInternal (const std::string &message, const bool warning) const

CovarianceFunctionBase *	getCovarianceFunctionByName (const UserObjectName &name) const
	Lookup a CovarianceFunction object by name and return pointer.

Protected Attributes
const unsigned int	_num_expansion_terms
	The number of expansion terms in the output ovariance matrix.

HyperParameterMap	_hyperparameters
	Map of hyperparameters stored as rank-0 or rank-1 tensors.

std::unordered_set< std::string >	_tunable_hp
	list of tunable hyper-parameters

const unsigned int	_num_outputs
	The number of outputs this covariance function is used to describe.

const std::vector< UserObjectName >	_dependent_covariance_names
	The names of the dependent covariance functions.

std::vector< std::string >	_dependent_covariance_types
	The types of the dependent covariance functions.

std::vector< CovarianceFunctionBase * >	_covariance_functions
	Vector of pointers to the dependent covariance functions.

const bool &	_enabled

MooseApp &	_app

Factory &	_factory

ActionFactory &	_action_factory

const std::string &	_type

const std::string &	_name

const InputParameters &	_pars

const Parallel::Communicator &	_communicator

Static Private Member Functions
static const hit::Node *	getHitNode (const InputParameters &params)

static std::string	messagePrefix (const InputParameters &params, const bool hit_prefix)

Private Attributes
std::vector< const torch::Tensor * >	_a_coeffs
	The vectors in the $B = \sum_i a_i a_i^T + diag(lambda_i)$ expansion.

std::vector< const torch::Tensor * >	_lambdas

const ParallelParamObject &	_parent

const MooseBase &	_si_moose_base

const FEProblemBase *	_si_problem

FEProblemBase &	_covar_feproblem
	Reference to FEProblemBase instance.

Detailed Description

Covariance function for multi-output Gaussian Processes based on the linear model of coregionalization (LMC)

Definition at line 19 of file LMC.h.

Member Typedef Documentation

◆ HyperParameterMap

using CovarianceFunctionBase::HyperParameterMap = std::unordered_map<std::string, torch::Tensor>

inherited

Definition at line 23 of file CovarianceFunctionBase.h.

Constructor & Destructor Documentation

◆ LMC()

LMC::LMC ( const InputParameters & parameters )

Definition at line 30 of file LMC.C.

  : CovarianceFunctionBase(parameters),
    _num_expansion_terms(getParam<unsigned int>("num_latent_funcs"))
{
  // We use a random number generator to obtain the initial guess for the
  // hyperparams
  MooseRandom generator_latent;
  generator_latent.seed(0, 1980);
 
  // First add and initialize the a A coefficients in the (aa^T+lambda*I) matrix
  for (const auto exp_i : make_range(_num_expansion_terms))
  {
    const std::string a_coeff_name = "acoeff_" + std::to_string(exp_i);
    std::vector<Real> acoeff_values(_num_outputs);
    for (const auto out_i : make_range(_num_outputs))
      acoeff_values[out_i] = 3.0 * generator_latent.rand(0) + 1.0;
    auto & acoeff_vector = addVectorRealHyperParameter(a_coeff_name, acoeff_values, true);
    _a_coeffs.push_back(&acoeff_vector);
  }
 
  // Then add and initialize the lambda coefficients in the (aa^T+lambda*I) matrix
  for (const auto exp_i : make_range(_num_expansion_terms))
  {
    const std::string lambda_name = "lambda_" + std::to_string(exp_i);
    std::vector<Real> lambda_values(_num_outputs);
    for (const auto out_i : make_range(_num_outputs))
      lambda_values[out_i] = 3.0 * generator_latent.rand(0) + 1.0;
    auto & lambda_vector = addVectorRealHyperParameter(lambda_name, lambda_values, true);
    _lambdas.push_back(&lambda_vector);
  }
}

Member Function Documentation

◆ addRealHyperParameter()

torch::Tensor & CovarianceFunctionBase::addRealHyperParameter	(	const std::string &	name,
		const Real	value,
		const bool	is_tunable
	)

protectedinherited

Register a scalar hyperparameter to this covariance function.

Parameters

name	The name of the parameter
value	The initial value of the parameter
is_tunable	If the parameter is tunable during optimization

Definition at line 95 of file CovarianceFunctionBase.C.

{
  const auto prefixed_name = _name + ":" + name;
  return insertHyperParameter(
      _hyperparameters, _tunable_hp, prefixed_name, makeScalarHyperParameter(value), is_tunable);
}

◆ addVectorRealHyperParameter()

torch::Tensor & CovarianceFunctionBase::addVectorRealHyperParameter	(	const std::string &	name,
		const std::vector< Real > &	value,
		const bool	is_tunable
	)

protectedinherited

Register a vector hyperparameter to this covariance function.

Parameters

name	The name of the parameter
value	The initial value of the parameter
is_tunable	If the parameter is tunable during optimization

Definition at line 105 of file CovarianceFunctionBase.C.

{
  const auto prefixed_name = _name + ":" + name;
  return insertHyperParameter(
      _hyperparameters, _tunable_hp, prefixed_name, makeVectorHyperParameter(value), is_tunable);
}

Referenced by LMC().

◆ buildHyperParamMap()

void CovarianceFunctionBase::buildHyperParamMap ( HyperParameterMap & map ) const

inherited

Populates the input maps with the owned hyperparameters.

Parameters

map	Map of hyperparameters that should be populated

Definition at line 153 of file CovarianceFunctionBase.C.

{
  // First, add the hyperparameters of the dependent covariance functions
  for (const auto dependent_covar : _covariance_functions)
    dependent_covar->buildHyperParamMap(map);
 
  // At the end we just append the hyperparameters this object owns
  for (const auto & iter : _hyperparameters)
    if (!isScalarHyperParameter(iter.second) && !isVectorHyperParameter(iter.second))
      mooseError("Unsupported hyperparameter rank ", iter.second.dim(), " for ", iter.first, ".");
    else
      map[iter.first] = iter.second.clone();
}

Referenced by StochasticTools::GaussianProcess::setupCovarianceMatrix().

◆ computeAGradient()

void LMC::computeAGradient	(	torch::Tensor &	grad,
		const unsigned int	exp_i,
		const unsigned int	index
	)		const

protected

Computes the gradient of $B$ with respect to the entries in $a_i$ in the following expression: $B = \sum_i a_i a_i^T + diag(lambda_i)$.

Parameters

grad	The gradient matrix that should be populated
exp_i	The index of the expansion of B
index	The index within the vector $a_i$

Definition at line 168 of file LMC.C.

{
  const auto & a_coeffs = *_a_coeffs[exp_i];
  mooseAssert(cast_int<int64_t>(index) < a_coeffs.numel(), "Incorrect LMC coefficient index.");
  auto basis = torch::zeros_like(a_coeffs);
  const auto index_tensor =
      torch::tensor({cast_int<int64_t>(index)},
                    torch::TensorOptions().dtype(torch::kLong).device(a_coeffs.device()));
  basis.index_fill_(0, index_tensor, 1.0);
  grad = torch::outer(basis, a_coeffs) + torch::outer(a_coeffs, basis);
}

Referenced by computedKdhyper().

◆ computeBMatrix()

void LMC::computeBMatrix	(	torch::Tensor &	Bmat,
		const unsigned int	exp_i
	)		const

protected

Computes the covariance matrix for the outputs (using the latent coefficients) We use a $B = \sum_i a_i a_i^T + diag(lambda_i)$ expansion here where $a_i$ and $lambda_i$ are vectors.

Parameters

Bmat	The matrix which should be populated by the covariance values
exp_i	The expansion index in the latent space

Definition at line 160 of file LMC.C.

{
  const auto & a_coeffs = *_a_coeffs[exp_i];
  const auto & lambda_coeffs = *_lambdas[exp_i];
  Bmat = torch::outer(a_coeffs, a_coeffs) + torch::diag(lambda_coeffs);
}

Referenced by computeCovarianceMatrix(), and computedKdhyper().

◆ computeCovarianceMatrix()

void LMC::computeCovarianceMatrix	(	torch::Tensor &	K,
		const torch::Tensor &	x,
		const torch::Tensor &	xp,
		const bool	is_self_covariance
	)		const

overridevirtual

Generates the Covariance Matrix given two sets of points in the parameter space.

Parameters

K	Reference to a matrix which should be populated by the covariance entries
x	Reference to the first set of points
xp	Reference to the second set of points
is_self_covariance	Switch to enable adding the noise variance to the diagonal of the covariance matrix

Implements CovarianceFunctionBase.

Definition at line 63 of file LMC.C.

{
  const auto options = x.options().dtype(at::kDouble);
  // Create temporary vectors for constructing the covariance matrix
  torch::Tensor K_params = torch::zeros({x.sizes()[0], xp.sizes()[0]}, options);
  torch::Tensor B = torch::zeros({_num_outputs, _num_outputs}, options);
  K = torch::zeros({x.sizes()[0] * _num_outputs, xp.sizes()[0] * _num_outputs}, options);
  torch::Tensor K_working;
 
  // For every expansion term we add the contribution to the covariance matrix
  for (const auto exp_i : make_range(_num_expansion_terms))
  {
    _covariance_functions[exp_i]->computeCovarianceMatrix(K_params, x, xp, is_self_covariance);
    computeBMatrix(B, exp_i);
    K_working = torch::kron(B, K_params);
    K += K_working;
  }
}

◆ computedKdhyper()

bool LMC::computedKdhyper	(	torch::Tensor &	dKdhp,
		const torch::Tensor &	x,
		const std::string &	hyper_param_name,
		unsigned int	ind
	)		const

overridevirtual

Redirect dK/dhp for hyperparameter "hp".

Returns false is the parameter has not been found in this covariance object.

Parameters

dKdhp	The matrix which should be populated with the derivatives
x	The input vector for which the derivatives of the covariance matrix is computed
hyper_param_name	The name of the hyperparameter
ind	The index within the hyperparameter. 0 if it is a scalar parameter. If it is a vector parameter, it should be the index within the vector.

Reimplemented from CovarianceFunctionBase.

Definition at line 86 of file LMC.C.

{
  // Early return in the paramter name is longer than the expected [name] prefix.
  // We prefix the parameter names with the name of the covariance function.
  if (name().length() + 1 > hyper_param_name.length())
    return false;
 
  // Strip the prefix from the given parameter name
  const std::string name_without_prefix = hyper_param_name.substr(name().length() + 1);
 
  // Check if the parameter is tunable
  if (_tunable_hp.find(hyper_param_name) != _tunable_hp.end())
  {
    const std::string acoeff_prefix = "acoeff_";
    const std::string lambda_prefix = "lambda_";
 
    // Allocate storage for the factors of the total gradient matrix
    const auto options = x.options().dtype(at::kDouble);
    torch::Tensor dBdhp = torch::zeros({_num_outputs, _num_outputs}, options);
    torch::Tensor K_params = torch::zeros({x.sizes()[0], x.sizes()[0]}, options);
 
    if (name_without_prefix.find(acoeff_prefix) != std::string::npos)
    {
      // Automatically grab the expansion index
      const int number = std::stoi(name_without_prefix.substr(acoeff_prefix.length()));
      computeAGradient(dBdhp, number, ind);
      _covariance_functions[number]->computeCovarianceMatrix(K_params, x, x, true);
    }
    else if (name_without_prefix.find(lambda_prefix) != std::string::npos)
    {
      // Automatically grab the expansion index
      const int number = std::stoi(name_without_prefix.substr(lambda_prefix.length()));
      computeLambdaGradient(dBdhp, number, ind);
      _covariance_functions[number]->computeCovarianceMatrix(K_params, x, x, true);
    }
    dKdhp = torch::kron(dBdhp, K_params);
    return true;
  }
  else
  {
    // Allocate storage for the matrix factors
    const auto options = x.options().dtype(at::kDouble);
    torch::Tensor B_tmp = torch::zeros({_num_outputs, _num_outputs}, options);
    torch::Tensor B = torch::zeros({_num_outputs, _num_outputs}, options);
    torch::Tensor dKdhp_sub = torch::zeros({x.sizes()[0], x.sizes()[0]}, options);
 
    // First, check the dependent covariances
    bool found = false;
    for (const auto dependent_covar : _covariance_functions)
      if (!found)
        found = dependent_covar->computedKdhyper(dKdhp_sub, x, hyper_param_name, ind);
 
    if (!found)
      mooseError("Hyperparameter ", hyper_param_name, "not found!");
 
    // Then we compute the output covariance
    for (const auto exp_i : make_range(_num_expansion_terms))
    {
      computeBMatrix(B_tmp, exp_i);
      B += B_tmp;
    }
 
    dKdhp = torch::kron(B, dKdhp_sub);
 
    return true;
  }
 
  return false;
}

◆ computeLambdaGradient()

void LMC::computeLambdaGradient	(	torch::Tensor &	grad,
		const unsigned int	exp_i,
		const unsigned int	index
	)		const

protected

Computes the gradient of $B$ with respect to the entries in $lambda_i$ in the following expression: $B = \sum_i a_i a_i^T + diag(lambda_i)$.

Parameters

grad	The gradient matrix that should be populated
exp_i	The index of the expansion of B
index	The index within the vector $lambda_i$

Definition at line 183 of file LMC.C.

{
  mooseAssert(index < _num_outputs, "Incorrect LMC lambda index.");
  auto basis = torch::zeros_like(*_lambdas[exp_i]);
  const auto index_tensor =
      torch::tensor({cast_int<int64_t>(index)},
                    torch::TensorOptions().dtype(torch::kLong).device(_lambdas[exp_i]->device()));
  basis.index_fill_(0, index_tensor, 1.0);
  grad = torch::diag(basis);
}

Referenced by computedKdhyper().

◆ dependentCovarianceNames()

const std::vector< UserObjectName > & CovarianceFunctionBase::dependentCovarianceNames ( ) const

inlineinherited

Get the names of the dependent covariances.

Definition at line 67 of file CovarianceFunctionBase.h.

  {
    return _dependent_covariance_names;
  }

Referenced by StochasticTools::GaussianProcess::linkCovarianceFunction().

◆ dependentCovarianceTypes()

void CovarianceFunctionBase::dependentCovarianceTypes ( std::map< UserObjectName, std::string > & name_type_map ) const

inherited

Populate a map with the names and types of the dependent covariance functions.

Parameters

name_type_map Reference to the map which should be populated

Definition at line 204 of file CovarianceFunctionBase.C.

{
  for (const auto dependent_covar : _covariance_functions)
  {
    dependent_covar->dependentCovarianceTypes(name_type_map);
    name_type_map.insert(std::make_pair(dependent_covar->name(), dependent_covar->type()));
  }
}

Referenced by StochasticTools::GaussianProcess::linkCovarianceFunction().

◆ getCovarianceFunctionByName()

CovarianceFunctionBase * CovarianceInterface::getCovarianceFunctionByName ( const UserObjectName & name ) const

protectedinherited

Lookup a CovarianceFunction object by name and return pointer.

Definition at line 26 of file CovarianceInterface.C.

{
  std::vector<CovarianceFunctionBase *> models;
  _covar_feproblem.theWarehouse()
      .query()
      .condition<AttribName>(name)
      .condition<AttribSystem>("CovarianceFunction")
      .queryInto(models);
  if (models.empty())
    mooseError("Unable to find a CovarianceFunction object with the name '" + name + "'");
  return models[0];
}

Referenced by ActiveLearningGaussianProcess::ActiveLearningGaussianProcess(), CovarianceFunctionBase::CovarianceFunctionBase(), GaussianProcessTrainer::GaussianProcessTrainer(), and GaussianProcessSurrogate::setupCovariance().

◆ getTuningData()

bool CovarianceFunctionBase::getTuningData	(	const std::string &	name,
		unsigned int &	size,
		Real &	min,
		Real &	max
	)		const

virtualinherited

Get the default minimum and maximum and size of a hyperparameter.

Returns false is the parameter has not been found in this covariance object.

Parameters

name	The name of the hyperparameter
size	Reference to an unsigned int that will contain the size of the hyperparameter (will be populated with 1 if it is scalar)
min	Reference to a number which will be populated by the maximum allowed value of the hyperparameter
max	Reference to a number which will be populated by the minimum allowed value of the hyperparameter

Definition at line 168 of file CovarianceFunctionBase.C.

{
  // First, check the dependent covariances
  for (const auto dependent_covar : _covariance_functions)
    if (dependent_covar->getTuningData(name, size, min, max))
      return true;
 
  min = 1e-9;
  max = 1e9;
 
  const auto tensor_value = _hyperparameters.find(name);
  if (tensor_value == _hyperparameters.end())
  {
    size = 0;
    return false;
  }
 
  if (isScalarHyperParameter(tensor_value->second))
  {
    size = 1;
    return true;
  }
 
  if (isVectorHyperParameter(tensor_value->second))
  {
    size = tensor_value->second.numel();
    return true;
  }
 
  mooseError("Unsupported hyperparameter rank ", tensor_value->second.dim(), " for ", name, ".");
}

Referenced by StochasticTools::GaussianProcess::generateTuningMap().

◆ isScalarHyperParameter()

bool CovarianceFunctionBase::isScalarHyperParameter ( const torch::Tensor & tensor )

staticinherited

Return true if a hyperparameter tensor stores one scalar value.

Definition at line 44 of file CovarianceFunctionBase.C.

{
  return tensor.dim() == 0;
}

Referenced by CovarianceFunctionBase::buildHyperParamMap(), CovarianceFunctionBase::getTuningData(), GaussianProcessData::initialize(), and CovarianceFunctionBase::loadHyperParamMap().

◆ isTunable()

bool CovarianceFunctionBase::isTunable ( const std::string & name ) const

virtualinherited

Check if a given parameter is tunable.

Parameters

The	name of the hyperparameter

Definition at line 115 of file CovarianceFunctionBase.C.

{
  // First, we check if the dependent covariances have the parameter
  for (const auto dependent_covar : _covariance_functions)
    if (dependent_covar->isTunable(name))
      return true;
 
  if (_tunable_hp.find(name) != _tunable_hp.end())
    return true;
  else if (_hyperparameters.find(name) != _hyperparameters.end())
    mooseError("We found hyperparameter ", name, " but it was not declared tunable!");
 
  return false;
}

Referenced by StochasticTools::GaussianProcess::generateTuningMap().

◆ isVectorHyperParameter()

bool CovarianceFunctionBase::isVectorHyperParameter ( const torch::Tensor & tensor )

staticinherited

Return true if a hyperparameter tensor stores a vector of values.

Definition at line 50 of file CovarianceFunctionBase.C.

{
  return tensor.dim() == 1;
}

Referenced by CovarianceFunctionBase::buildHyperParamMap(), CovarianceFunctionBase::getTuningData(), GaussianProcessData::initialize(), and CovarianceFunctionBase::loadHyperParamMap().

◆ loadHyperParamMap()

void CovarianceFunctionBase::loadHyperParamMap ( const HyperParameterMap & map )

inherited

Load some hyperparameters into the local map contained in this object.

Parameters

map	Input map of hyperparameters

Definition at line 131 of file CovarianceFunctionBase.C.

{
  // First, load the hyperparameters of the dependent covariance functions
  for (const auto dependent_covar : _covariance_functions)
    dependent_covar->loadHyperParamMap(map);
 
  // Then we load the hyperparameters of this object
  for (auto & iter : _hyperparameters)
  {
    const auto map_iter = map.find(iter.first);
    if (map_iter == map.end())
      continue;
 
    if (!isScalarHyperParameter(map_iter->second) && !isVectorHyperParameter(map_iter->second))
      mooseError(
          "Unsupported hyperparameter rank ", map_iter->second.dim(), " for ", iter.first, ".");
 
    iter.second = map_iter->second.clone();
  }
}

Referenced by LoadCovarianceDataAction::load(), StochasticTools::GaussianProcess::setupCovarianceMatrix(), and StochasticTools::GaussianProcess::tuneHyperParamsAdam().

◆ numOutputs()

unsigned int CovarianceFunctionBase::numOutputs ( ) const

inlineinherited

Return the number of outputs assumed for this covariance function.

Definition at line 90 of file CovarianceFunctionBase.h.

90{ return _num_outputs; }

Referenced by GaussianProcessSurrogate::evaluate(), GaussianProcessTrainer::GaussianProcessTrainer(), and StochasticTools::GaussianProcess::linkCovarianceFunction().

◆ validParams()

InputParameters LMC::validParams ( )

static

Definition at line 18 of file LMC.C.

{
  InputParameters params = CovarianceFunctionBase::validParams();
  params.addClassDescription("Covariance function for multioutput Gaussian Processes based on the "
                             "Linear Model of Coregionalization (LMC).");
  params.addParam<unsigned int>(
      "num_latent_funcs", 1., "The number of latent functions for the expansion of the outputs.");
  params.makeParamRequired<unsigned int>("num_outputs");
  params.makeParamRequired<std::vector<UserObjectName>>("covariance_functions");
  return params;
}

Member Data Documentation

◆ _a_coeffs

std::vector<const torch::Tensor *> LMC::_a_coeffs

private

The vectors in the $B = \sum_i a_i a_i^T + diag(lambda_i)$ expansion.

Definition at line 72 of file LMC.h.

Referenced by computeAGradient(), computeBMatrix(), and LMC().

◆ _covar_feproblem

FEProblemBase& CovarianceInterface::_covar_feproblem

privateinherited

Reference to FEProblemBase instance.

Definition at line 31 of file CovarianceInterface.h.

Referenced by CovarianceInterface::getCovarianceFunctionByName().

◆ _covariance_functions

std::vector<CovarianceFunctionBase *> CovarianceFunctionBase::_covariance_functions

protectedinherited

Vector of pointers to the dependent covariance functions.

Definition at line 124 of file CovarianceFunctionBase.h.

Referenced by CovarianceFunctionBase::buildHyperParamMap(), computeCovarianceMatrix(), computedKdhyper(), CovarianceFunctionBase::CovarianceFunctionBase(), CovarianceFunctionBase::dependentCovarianceTypes(), CovarianceFunctionBase::getTuningData(), CovarianceFunctionBase::isTunable(), and CovarianceFunctionBase::loadHyperParamMap().

◆ _dependent_covariance_names

const std::vector<UserObjectName> CovarianceFunctionBase::_dependent_covariance_names

protectedinherited

The names of the dependent covariance functions.

Definition at line 118 of file CovarianceFunctionBase.h.

Referenced by CovarianceFunctionBase::CovarianceFunctionBase(), and CovarianceFunctionBase::dependentCovarianceNames().

◆ _dependent_covariance_types

std::vector<std::string> CovarianceFunctionBase::_dependent_covariance_types

protectedinherited

The types of the dependent covariance functions.

Definition at line 121 of file CovarianceFunctionBase.h.

Referenced by CovarianceFunctionBase::CovarianceFunctionBase().

◆ _hyperparameters

HyperParameterMap CovarianceFunctionBase::_hyperparameters

protectedinherited

Map of hyperparameters stored as rank-0 or rank-1 tensors.

Definition at line 109 of file CovarianceFunctionBase.h.

Referenced by CovarianceFunctionBase::addRealHyperParameter(), CovarianceFunctionBase::addVectorRealHyperParameter(), CovarianceFunctionBase::buildHyperParamMap(), CovarianceFunctionBase::getTuningData(), CovarianceFunctionBase::isTunable(), and CovarianceFunctionBase::loadHyperParamMap().

◆ _lambdas

std::vector<const torch::Tensor *> LMC::_lambdas

private

Definition at line 73 of file LMC.h.

Referenced by computeBMatrix(), computeLambdaGradient(), and LMC().

◆ _num_expansion_terms

const unsigned int LMC::_num_expansion_terms

protected

The number of expansion terms in the output ovariance matrix.

Definition at line 67 of file LMC.h.

Referenced by computeCovarianceMatrix(), computedKdhyper(), and LMC().

◆ _num_outputs

const unsigned int CovarianceFunctionBase::_num_outputs

protectedinherited

The number of outputs this covariance function is used to describe.

Definition at line 115 of file CovarianceFunctionBase.h.

Referenced by computeCovarianceMatrix(), computedKdhyper(), computeLambdaGradient(), LMC(), and CovarianceFunctionBase::numOutputs().

◆ _tunable_hp

std::unordered_set<std::string> CovarianceFunctionBase::_tunable_hp

protectedinherited

list of tunable hyper-parameters

Definition at line 112 of file CovarianceFunctionBase.h.

Referenced by CovarianceFunctionBase::addRealHyperParameter(), CovarianceFunctionBase::addVectorRealHyperParameter(), computedKdhyper(), and CovarianceFunctionBase::isTunable().

The documentation for this class was generated from the following files:

stochastic_tools/include/covariances/LMC.h
stochastic_tools/src/covariances/LMC.C

Public Types

Public Member Functions

Static Public Member Functions

Public Attributes

Static Public Attributes

Protected Member Functions

Protected Attributes

Static Private Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

◆ HyperParameterMap

Constructor & Destructor Documentation

◆ LMC()

Member Function Documentation

◆ addRealHyperParameter()

◆ addVectorRealHyperParameter()

◆ buildHyperParamMap()

◆ computeAGradient()

◆ computeBMatrix()

◆ computeCovarianceMatrix()

◆ computedKdhyper()

◆ computeLambdaGradient()

◆ dependentCovarianceNames()

◆ dependentCovarianceTypes()

◆ getCovarianceFunctionByName()

◆ getTuningData()

◆ isScalarHyperParameter()

◆ isTunable()

◆ isVectorHyperParameter()

◆ loadHyperParamMap()

◆ numOutputs()

◆ validParams()

Member Data Documentation

◆ _a_coeffs

◆ _covar_feproblem

◆ _covariance_functions

◆ _dependent_covariance_names

◆ _dependent_covariance_types

◆ _hyperparameters

◆ _lambdas

◆ _num_expansion_terms

◆ _num_outputs

◆ _tunable_hp