superviseddescent::Regulariser Class Reference

#include <regressors.hpp>

Public Types
enum	RegularisationType { RegularisationType::Manual, RegularisationType::MatrixNorm }

Public Member Functions
	Regulariser (RegularisationType regularisation_type=RegularisationType::Manual, float param=0.0f, bool regularise_last_row=true)
cv::Mat	get_matrix (cv::Mat data, int num_training_elements)

Friends
class	cereal::access

Detailed Description

Regulariser class for the LinearRegressor.

Produces a diagonal matrix with a regularisation value (lambda) on the diagonal that can be added to the data matrix. Lambda can either be fixed or calculated from the data matrix.

Member Enumeration Documentation

enum superviseddescent::Regulariser::RegularisationType

strong

The method to calculate the regularisation factor lambda.

Enumerator
Manual	Use the given param value as lambda.
MatrixNorm	Use \( \text{lambda} = \text{param} * \frac{\|\text{data}\|_2}{\text{numTrainingElements}} \) A suitable default for param suggested by the SDM authors is 0.5.

Constructor & Destructor Documentation

superviseddescent::Regulariser::Regulariser ( RegularisationType  regularisation_type = RegularisationType::Manual, float  param = 0.0f, bool  regularise_last_row = true  )

inline

Create a new regulariser. Created with default parameters, it will not do any regularisation. Regulariser::RegularisationType can be used to specify the choice of the regularisation parameter lambda.

regularise_last_row is useful to specify the regularisation behaviour in the case the last row of the data matrix contains an affine (offset or bias) component. In that case, you might not want to regularise it (or maybe you do).

Parameters

[in]	regularisation_type	Specifies how to calculate lambda.
[in]	param	Lambda, or a factor, depending on regularisationType.
[in]	regularise_last_row	Specifies if the last row should be regularised.

Member Function Documentation

cv::Mat superviseddescent::Regulariser::get_matrix ( cv::Mat  data, int  num_training_elements  )

inline

Calculates a diagonal regularisation matrix, given the RegularisationType and param of the regulariser. It will have the same dimensions as the given data matrix.

Parameters

[in]	data	Data matrix that might be used to calculate an automatic value for lambda.
[in]	num_training_elements	Number of training elements.

Returns

Returns a diagonal regularisation matrix with the same dimensions as the given data matrix.

---

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

• include/superviseddescent/regressors.hpp