src_c/util.c File Reference

Source file for utility functions. More...

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include <float.h>
#include "mmio.h"
#include "def.h"
#include "util.h"
#include "l1_logreg.h"
#include "dmatrix.h"

Defines
#define	BIN_EXP_BEGIN   (-7)
	Shows histogram of coefficients in 10-base logscale.
#define	BIN_EXP_END   (+2)
#define	BIN_NUM_PER_DECADE   (8)
#define	BIN_IDX_NUM   ((BIN_EXP_END-BIN_EXP_BEGIN)*BIN_NUM_PER_DECADE)
#define	BIN_IDX_BEGIN   (0)
#define	BIN_IDX_END   (BIN_IDX_BEGIN+BIN_IDX_NUM-1)
Functions
double	find_lambdamax (const dmatrix *X, const double *b, const int sflag)
	Returns the maximum value of the regularization parameter lambda that gives a non-zero solution.
void	standardize_data (dmatrix *X, const double *b, double **average, double **stddev, double **acol, double **arow)
	Standardizes the data.
int	read_mm_new_matrix (const char *file, dmatrix **out_mat)
	Read a matrix file.
int	read_mm_new_matrix_transpose (const char *file, dmatrix **out_mat)
int	read_mm_new_vector (const char *file, double **out_vec)
	Read a vector file.
int	write_mm_vector (const char *file, const int m, const double *vec, const char *comments, const int type)
int	write_mm_matrix (const char *file, dmatrix *mat, const char *comments, const int type)
int	write_mm_matrix_crd_header (FILE *fp, const int m, const int n, const int nnz, const char *comments)
int	write_mm_matrix_crd_column (FILE *fp, const int len, const int col, const double *vec)
int	write_mm_matrix_crd_column_threshold (FILE *fp, const int len, const int col, const double *vec, const double threshold)
int	get_mm_info (const char *file, int *m, int *n, int *nz)
int	show_histogram (const int n, const double *coeff)
double	userinput_threshold (void)
	Gets a threshold value from standard input.
int	thresholding (const int n, double *coeff, double threshold)
	Thresholds coefficients.
void	buffer_new (buffer_t **buf, int elemsize, int length)
void	buffer_free (buffer_t *buf)
int	buffer_write (buffer_t *buf, void *src, int n)
int	condense_solution (const int len, double *sol, int *idx)
	Condense solution by removing zeros.

---

Detailed Description

Source file for utility functions.

---

Define Documentation

#define BIN_EXP_BEGIN   (-7)

Shows histogram of coefficients in 10-base logscale.

Parameters:

	n	number of coefficients
	coeff	vector of coefficients

---

Function Documentation

int condense_solution	(	const int	len,
		double *	sol,
		int *	idx
	)

Condense solution by removing zeros.

Truncates coefficients if their absolute values are below the threshold.

Parameters:

	len	size of the original solution vector
	sol	solution vector
	idx	array of non-zero index in solution

ex: initial sol: [1 0 0 6 0 0 0 7 1] final sol: [1 6 7 1] idx: [1 4 8 9]

double find_lambdamax	(	const dmatrix *	X,
		const double *	b,
		const int	sflag
	)

Returns the maximum value of the regularization parameter lambda that gives a non-zero solution.

Parameters:

	X	feature matrix
	b	class vector
	sflag	standardization flag If sflag is 0, compute the maximum value of lambda without standardization. If sflag is 1, given matrix is standardized first and then the maximum value of lambda is computed.

Returns:

maximum value of lambda

References dmat_copy(), dmat_diagscale(), dmat_duplicate(), dmat_free(), dmat_norminf(), dmat_yAmpqTx(), dmatrix::m, dmatrix::n, and standardize_data().

int read_mm_new_matrix	(	const char *	file,
		dmatrix **	out_mat
	)

Read a matrix file.

Reads a Matrix Market formatted matrix from a file.

Parameters:

	file	matrix file name
	out_mat	matrix data

Returns:

result

References dmatrix::idx, dmatrix::jdx, dmatrix::m, dmatrix::n, dmatrix::nz, dmatrix::rdx, and dmatrix::val.

Referenced by main().

int read_mm_new_vector	(	const char *	file,
		double **	out_vec
	)

Read a vector file.

Reads a Matrix Market formatted vector from a file. Returns pointer to dense vector.

Parameters:

	file	vector file name
	out_vec	vector data

Returns:

result

Referenced by main().

void standardize_data	(	dmatrix *	X,
		const double *	b,
		double **	average,
		double **	stddev,
		double **	acol,
		double **	arow
	)

Standardizes the data.

Standardizes the feature matrix.

• If b is not given, compute
• If b is given, compute where is column average and is column deviation.

Parameters:

	X	feature matrix
	b	class vector
	average	column average
	stddev	column standard deviation
	acol	column vector used for implicit standardization
	arow	row vector used for implicit standardization

References dmat_colavg(), dmat_colstd(), dmat_diagscale(), dmat_elemdivi(), dmat_vcopy(), dmat_vset(), dmatrix::m, dmatrix::n, dmatrix::nz, and dmatrix::val.

Referenced by find_lambdamax(), and l1_logreg_train().

int thresholding	(	const int	n,
		double *	coeff,
		double	threshold
	)

Thresholds coefficients.

Truncates coefficients if their absolute values are below the threshold.

Parameters:

	n	number of coefficients
	coeff	vector of coefficients
	threshold	threshold value

double userinput_threshold

(

void

)

Gets a threshold value from standard input.

• If input value is positive, return it.
• If input value is negative, assume it as a 10-based log value. For example, -3 is converted to 10^(-3).

Returns:

threshold value