DataGeneratorBase¶

class DataGeneratorBase¶

Subclassed by DataGeneratorClassification, DataGeneratorRegression

Public Functions

DataGeneratorBase(Config *settings)¶

Class constructor. Generates training data for neural network approaches using spherical harmonics and an entropy functional and the quadrature specified by the options file.

Parameters

settings: config class with global information

void ComputeTrainingData() = 0¶: computes the training data set. Realizable set is sampled uniformly. Prototype: 1D, u in [0,100]

Public Static Functions

DataGeneratorBase *Create(Config *settings)¶

Create a datagenerator (1D or 3D)

Return

Pointer to the createt basis class

Parameters

settings: Pointer to the config file

Protected Functions

void SampleMultiplierAlpha()¶: Sample Lagrange multipliers alpha.

void ComputeRealizableSolution()¶: make u the realizable moment to alpha, since Newton has roundoff errors.

void PrintTrainingData() = 0¶: : Print computed training data to csv file and screen

void PrintLoadScreen()¶: Print screen IO.

void ComputeMoments() = 0¶: Pre-Compute Moments at all quadrature points.

bool ComputeEVRejection(unsigned idx_set)¶: Evalute rejection criterion based on the smallest Eigenvalue of the Hessian corresponding to alpha[idx_set].

bool ComputeReducedEVRejection(VectorVector &redMomentBasis, Vector &redAlpha)¶: Evalute rejection criterion based on the smallest Eigenvalue of the reduced Hessian corresponding to alpha[idx_set].

Protected Attributes

Config *_settings¶: config class for global information

VectorVector _uSol¶: vector with moments. Size: (setSize,basisSize)

VectorVector _alpha¶: vector with Lagrange multipliers. Size: (setSize,basisSize)

unsigned long _setSize¶: Size of the whole training Set.

unsigned short _maxPolyDegree¶: Max Order of Spherical Harmonics.

unsigned _nTotalEntries¶: Total number of equations in the system.

QuadratureBase *_quadrature¶: quadrature to create members below

unsigned _nq¶: number of quadrature points

VectorVector _quadPoints¶: quadrature points, dim(_quadPoints) = (_nq,spatialDim)

Vector _weights¶: quadrature weights, dim(_weights) = (_nq)

VectorVector _quadPointsSphere¶: (my,phi), dim(_quadPoints) = (_nq,2)

SphericalBase *_basisGenerator¶: Class to compute and store current spherical harmonics basis.

VectorVector _momentBasis¶: Moment Vector pre-computed at each quadrature point: dim= _nq x _nTotalEntries.

NewtonOptimizer *_optimizer¶: Class to solve minimal entropy problem.

EntropyBase *_entropy¶: Class to handle entropy functional evaluations.

bool _reducedSampling¶: Flag to show if the reduced optimizer is used.