Budget maintenance strategy that merges two vectors. More...
#include <shark/Algorithms/Trainers/Budgeted/MergeBudgetMaintenanceStrategy.h>
Additional Inherited Members | |
Public Types inherited from shark::AbstractBudgetMaintenanceStrategy< InputType > | |
typedef KernelExpansion< InputType > | ModelType |
typedef LabeledData< InputType, unsigned int > | DataType |
typedef DataType::element_type | ElementType |
Public Member Functions inherited from shark::AbstractBudgetMaintenanceStrategy< InputType > | |
AbstractBudgetMaintenanceStrategy () | |
virtual void | addToModel (ModelType &model, InputType const &alpha, ElementType const &supportVector)=0 |
std::string | name () const |
return the class name More... | |
Static Public Member Functions inherited from shark::AbstractBudgetMaintenanceStrategy< InputType > | |
static void | findSmallestVector (ModelType const &model, size_t &minIndex, double &minAlpha) |
Budget maintenance strategy that merges two vectors.
\[ ||\alpha||_2\]
coefficient– only then a second one is searched for, by inspecting the expected weight degradation after merging. The vector with smallest weight degradation is the vector one should merge with the first one. By this heuristic, the merging strategy has complexity\[ \mathcal{O}(B) \]
. Compared with the projection strategy, merging should be faster, and stil obtains similar accuracy. Unluckily any kind of timing numbers are missing in the reference paper of Wang, Crammer and Vucetic.Definition at line 83 of file MergeBudgetMaintenanceStrategy.h.