Reductions from matrix to row or column

Collaboration diagram for Reductions from matrix to row or column:

Enumerations
enum	cuv::reduce_functor {   RF_ADD, RF_MEAN, RF_ADD_SQUARED, RF_MAX,   RF_ARGMAX, RF_ARGMIN, RF_MIN, RF_MULT,   RF_LOGADDEXP, RF_ADDEXP }
	Reduce functor to reduce a matrix to a row or column. More...

Functions
template<class V , class __value_type2 , class M , class L >
void	cuv::reduce_to_col (tensor< V, M > &dst, const tensor< __value_type2, M, L > &src, reduce_functor rf=RF_ADD, const __value_type2 &factNew=1.f, const __value_type2 &factOld=0.f)
	Reduce a matrix to one column using specified reduce functor (or add them up by default)
template<class V , class __value_type2 , class M , class L >
void	cuv::reduce_to_row (tensor< V, M > &dst, const tensor< __value_type2, M, L > &src, reduce_functor rf=RF_ADD, const __value_type2 &factNew=1.f, const __value_type2 &factOld=0.f)
	Reduce a matrix to one row using specified reduce functor (or add them up by default)
template<class V , class M , class L >
tensor< V, M >	cuv::sum (const tensor< V, M, L > &src, const int &axis)
	Convenience function that creates a new vector and performs reduction by summing along given axis.

Detailed Description

Enumeration Type Documentation

enum cuv::reduce_functor

Reduce functor to reduce a matrix to a row or column.

• RF_ADD adds columns/rows
• RF_MEAN adds columns/rows, divides by number of entries
• RF_ADD_SQUARED adds squared entries of columns/rows
• RF_MAX uses maximum in colum (when reducing to row) or row (when reducing to column)
• RF_MIN uses minimum in colum (when reducing to row) or row (when reducing to column)
• RF_MULT multiplies by summing in log-space
• RF_LOGADDEXP calculates in a numerically stable way
• RF_ADDEXP calculates in a numerically stable way

Definition at line 185 of file matrix_ops.hpp.

Function Documentation

template<class V , class __value_type2 , class M , class L >

void cuv::reduce_to_col	(	tensor< V, M > &	dst,
		const tensor< __value_type2, M, L > &	src,
		reduce_functor	rf = RF_ADD,
		const __value_type2 &	factNew = 1.f,
		const __value_type2 &	factOld = 0.f
	)

Reduce a matrix to one column using specified reduce functor (or add them up by default)

Parameters

dst	Destination vector, dst.size = src.h()
src	Source matrix
rf	Reduce functor
factNew	Scalar factor for result of reduce functor
factOld	Scalar factor for former entry of dst Calculates dst= factOld * dst + factNew * rf(src) By default, the reduce functor is RF_ADD so that rf(src) is the sum over all columns of src.

template<class V , class __value_type2 , class M , class L >

void cuv::reduce_to_row	(	tensor< V, M > &	dst,
		const tensor< __value_type2, M, L > &	src,
		reduce_functor	rf = RF_ADD,
		const __value_type2 &	factNew = 1.f,
		const __value_type2 &	factOld = 0.f
	)

Reduce a matrix to one row using specified reduce functor (or add them up by default)

Parameters

dst	Destination vector, dst.size = src.w()
src	Source matrix
rf	Reduce functor
factNew	Scalar factor for result of reduce functor
factOld	Scalar factor for former entry of dst Calculates dst= factOld * dst + factNew * rf(src) By default, the reduce functor is RF_ADD so that rf(src) is the sum over all rows of src.

template<class V , class M , class L >

tensor<V, M> cuv::sum	(	const tensor< V, M, L > &	src,
		const int &	axis
	)

Convenience function that creates a new vector and performs reduction by summing along given axis.

Parameters

src	Source matrix
axis	Along which axis (0 = reduce to row, 1 = reduce to col)

Definition at line 239 of file matrix_ops.hpp.