SG++-Doxygen-Documentation
dataMatrixDatabase.cpp File Reference

Functions

int main ()
 This example shows how to initialize a data matrix of offline decompositions (needed for online objects) which enhances the performance since the decomposition usually takes some time. More...
 

Function Documentation

int main ( )

This example shows how to initialize a data matrix of offline decompositions (needed for online objects) which enhances the performance since the decomposition usually takes some time.

A database is always initialized upon a json file which contains paths to different matrix decompositions identified by the configuration of the grid, adaptivity and the density estimation itself.

First the database has to be initialized. This is done by passing the file to the json database file to the constructor of the DBMatDatabase class.

To retrieve an offline decomposition one must define the setup the offline matrix must match. This is done by defining how the sparse grid is structured, how the sparse grid performs refinement and coarsening (adaptivity). Also regularization and density estimation parameters must be defined to identify a decomposition. This is done by initializing respective structures for all of those settings.

First the sparse grid itself will be specified using a structure that inherits from the sgpp::base::GeneralGridConfiguration supertype. Each structure defines an unique type of a sparse grid, such as a regular sparse grid, geometry aware sparse grids.

Next the refinement and coarsening behaviour of the sparse grid is defined. In this example the default values are used.

Also the regularization must be specified, namely the type of the regularization operator and also the regularization strength, lambda. In this case the identity regularization operator is chosen.

Lastly the density estimation it self is configured. Since online / offline learning will be performed the density estimation type will be sgpp::datadriven::DensityEstimationType::Decomposition. Also the method of decomposition has to be specified, in this case Cholesky decomposition (which supports adaptivity of the grid) is chosen.

Before the matrix can be initialized the underlying grid needs to be created

This section shows how to store a decomposition in the database. First however the matrix has to be created and decomposed. This is done using the configuration structures that the database needs to identify a decomposition.

Afte decomposing the data matrix it can be stored in the database. By passing the configuration structures used to create the matrix the database can identify the decomposition. The last parameter of the putDataMatrix method specified whether an entry with the same configuration will be replaced with a new file path. Note that the database only works on file paths, i.e. strings.

Lastly it is shown how to retrieve a file path from the database. To identify a decomposition the entire configuration which would be used to initialize the data matrix must be passed to the database. If a matching entry is found the getDataMatrix method will return the file path associated with this configuration.

References sgpp::datadriven::DBMatOffline::buildMatrix(), sgpp::datadriven::DBMatOfflineFactory::buildOfflineObject(), sgpp::datadriven::Chol, sgpp::base::Grid::createLinearGrid(), sgpp::base::Grid::createModLinearGrid(), sgpp::datadriven::DBMatOffline::decomposeMatrix(), sgpp::datadriven::Decomposition, sgpp::datadriven::DensityEstimationConfiguration::decomposition_, sgpp::base::GeneralGridConfiguration::dim_, sgpp::datadriven::DBMatDatabase::getDataMatrix(), grid(), sgpp::datadriven::Identity, sgpp::datadriven::RegularizationConfiguration::lambda_, sgpp::base::GeneralGridConfiguration::level_, sgpp::base::Linear, sgpp::base::ModLinear, sgpp::datadriven::DBMatDatabase::putDataMatrix(), sgpp::datadriven::DBMatOffline::store(), sgpp::datadriven::RegularizationConfiguration::type_, sgpp::datadriven::DensityEstimationConfiguration::type_, and sgpp::base::GeneralGridConfiguration::type_.