Fn_fqmi

Functions
double	ecocpak::compute_sigma (const vector< ClassData > &classes_vector)
template<typename T1 , typename T2 >
T1::elem_type	ecocpak::fqmi (const Base< typename T1::elem_type, T1 > &X, const Base< typename T2::elem_type, T2 > &Y, const typename T1::elem_type sigma)
double	compute_sigma (const vector< ClassData > &classes_vector)

---

Function Documentation

double ecocpak::compute_sigma

(

const vector< ClassData > &

classes_vector

)

• Function compute_sigma() computes the sigma constant for the fqmi() function.

• Input Arguments:
  • classes_vector : Vector of classes.

• Output Arguments:
  • Void.

• Return Argument:
  • A double which represents the maximum distance between the sample vectors divided by 2.

double compute_sigma

(

const vector< ClassData > &

classes_vector

)

• Function compute_sigma() computes the sigma constant for the fqmi() function.

• Input Arguments:
  • classes_vector : Vector of classes.

• Output Arguments:
  • Void.

• Return Argument:
  • A double which represents the maximum distance between the sample vectors divided by 2.

template<typename T1 , typename T2 >

T1::elem_type ecocpak::fqmi	(	const Base< typename T1::elem_type, T1 > &	X,
		const Base< typename T2::elem_type, T2 > &	Y,
		const typename T1::elem_type	sigma
	)

• Function fqmi() computes the mutual information between the elements of samples (X) and their respective labels (Y). The computation is being done by the fast quadratic mutual information method, which is defined in the paper "Feature Extraction by Non-Parametric Mutual Information Maximization, Kari Torkkola, Journal of Machine Learning Research 2003, vol. 3, 1415-1438".

• As samples we consider the row vectors of input matrix X.

• Input Arguments:
  • X : A 2D matrix where each one of its rows represents a sample.
  • Y : A vector representing X's class labels.
  • sigma : Sigma variable of the radial basis function (influences the width of the kernel).

• Output Arguments:
  • Void.

• Return Argument:
  • The function fqmi() returns a scalar, which represents the value of the mutual information between the samples in X and their respective class labels in Y.