SG++-Doxygen-Documentation
|
The GroupLassoFunction class. More...
#include <GroupLassoFunction.hpp>
Public Member Functions | |
double | eval (sgpp::base::DataVector weights) override |
eval evaluates the regularization function for weights. More... | |
GroupLassoFunction (double lambda, sgpp::base::GridStorage *gridStorage) | |
GroupLassoFunction. More... | |
base::DataVector | prox (const sgpp::base::DataVector &weights, double stepsize) override |
prox evaluates the proximal operator for the function for weights. More... | |
The GroupLassoFunction class.
Corresponds to the regularization function \( \sum_{\mathbf{p} \in \mathcal{P}} \left(\sqrt{\vert \mathbf{p} \vert}\right) \Vert \mathbf{p} \Vert_2 \). In this formula \(\mathcal{P}\) is a partition of the weights. We use groups that correspond to the original features and their interactions, e.g. we have one group of basis functions that are constant for all but the first predictor, one that only models the second predictor, one that models the interaction of both mentioned predictors, and so on.
|
inline |
lambda | controls the regularization strength. |
gridStorage | is the grid storage. |
|
inlineoverridevirtual |
eval evaluates the regularization function for weights.
weights |
Implements sgpp::solver::RegularizationFunction.
References sgpp::base::DataVector::getSize(), and python.statsfileInfo::i.
|
inlineoverridevirtual |
prox evaluates the proximal operator for the function for weights.
weights | |
stepsize | is the stepsize used for the proximal step |
Implements sgpp::solver::RegularizationFunction.
References chess::dim, sgpp::base::HashGridStorage::getDimension(), sgpp::base::HashGridStorage::getPoint(), sgpp::base::HashGridStorage::getSize(), sgpp::base::DataVector::getSize(), python.statsfileInfo::i, python.utils.statsfile2gnuplot::j, and friedman::p.