gensvm_optimize.h File Reference

Header file for gensvm_optimize.c. More...

#include "gensvm_sv.h"
#include "gensvm_simplex.h"
#include "gensvm_update.h"
#include "gensvm_zv.h"

Include dependency graph for gensvm_optimize.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Functions
void	gensvm_optimize (struct GenModel *model, struct GenData *data)
	The main training loop for GenSVM. More...
double	gensvm_get_loss (struct GenModel *model, struct GenData *data, struct GenWork *work)
	Calculate the current value of the loss function. More...
void	gensvm_calculate_errors (struct GenModel *model, struct GenData *data, double *ZV)
	Calculate the scalar errors. More...
void	gensvm_calculate_ZV_dense (struct GenModel *model, struct GenData *data, double *ZV)
	Compute the product Z*V for when Z is a dense matrix. More...
void	gensvm_calculate_ZV_sparse (struct GenModel *model, struct GenData *data, double *ZV)
	Compute the product Z*V for when Z is a sparse matrix. More...
void	gensvm_calculate_huber (struct GenModel *model)
	Calculate the Huber hinge errors. More...
void	gensvm_step_doubling (struct GenModel *model)
	Use step doubling. More...

Detailed Description

Header file for gensvm_optimize.c.

Author

G.J.J. van den Burg

Date

2013-08-01

Contains function declarations for functions used to train a single GenModel.

Copyright

Copyright 2016, G.J.J. van den Burg.

This file is part of GenSVM.

GenSVM is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

GenSVM is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with GenSVM. If not, see http://www.gnu.org/licenses/.

Definition in file gensvm_optimize.h.

Function Documentation

gensvm_calculate_errors()

void gensvm_calculate_errors	(	struct GenModel *	model,
		struct GenData *	data,
		double *	ZV
	)

Calculate the scalar errors.

Calculate the scalar errors q based on the current estimate of V, and store these in Q. It is assumed that the memory for Q has already been allocated. In addition, the matrix ZV is calculated here. It is assigned to a pre-allocated block of memory, which is passed to this function.

Parameters

[in,out]	model	the corresponding GenModel
[in]	data	the corresponding GenData
[in,out]	ZV	a pointer to a memory block for ZV. On exit this block is updated with the new ZV matrix calculated with GenModel::V

Definition at line 277 of file gensvm_optimize.c.

Here is the call graph for this function:

gensvm_calculate_huber()

void gensvm_calculate_huber

(

struct GenModel *

model

)

Calculate the Huber hinge errors.

For each of the scalar errors in Q the Huber hinge errors are calculated. The Huber hinge is here defined as

Parameters

[in,out]

model

the corresponding GenModel

Definition at line 242 of file gensvm_optimize.c.

gensvm_calculate_ZV_dense()

void gensvm_calculate_ZV_dense	(	struct GenModel *	model,
		struct GenData *	data,
		double *	ZV
	)

Compute the product Z*V for when Z is a dense matrix.

This function uses cblas_dgemm() to compute the matrix product between Z and V.

Parameters

[in]	model	a GenModel instance holding the model
[in]	data	a GenData instance with the data
[out]	ZV	a pre-allocated matrix of appropriate dimensions

Definition at line 108 of file gensvm_zv.c.

gensvm_calculate_ZV_sparse()

void gensvm_calculate_ZV_sparse	(	struct GenModel *	model,
		struct GenData *	data,
		double *	ZV
	)

Compute the product Z*V for when Z is a sparse matrix.

This is a simple sparse-dense matrix multiplication, which uses cblas_daxpy() for each nonzero element of Z, to compute Z*V.

Parameters

[in]	model	a GenModel instance holding the model
[in]	data	a GenData instance with the data
[out]	ZV	a pre-allocated matrix of appropriate dimensions

Definition at line 70 of file gensvm_zv.c.

gensvm_get_loss()

double gensvm_get_loss	(	struct GenModel *	model,
		struct GenData *	data,
		struct GenWork *	work
	)

Calculate the current value of the loss function.

The current loss function value is calculated based on the matrix V in the given model. Note that the matrix ZV is passed explicitly to avoid having to reallocate memory at every step.

Parameters

[in]	model	GenModel structure which holds the current estimate V
[in]	data	GenData structure
[in]	work	allocated workspace with the ZV matrix to use

Returns

the current value of the loss function

Definition at line 155 of file gensvm_optimize.c.

Here is the call graph for this function:

gensvm_optimize()

void gensvm_optimize	(	struct GenModel *	model,
		struct GenData *	data
	)

The main training loop for GenSVM.

This function is the main training function. This function handles the optimization of the model with the given model parameters, with the data given. On return the matrix GenModel::V contains the optimal weight matrix.

In this function, step doubling is used in the majorization algorithm after a burn-in of 50 iterations.

Parameters

[in,out]	model	the GenModel to be trained. Contains optimal V on exit.
[in]	data	the GenData to train the model with.

Definition at line 56 of file gensvm_optimize.c.

Here is the call graph for this function:

gensvm_step_doubling()

void gensvm_step_doubling

(

struct GenModel *

model

)

Use step doubling.

Step doubling can be used to speed up the maorization algorithm. Instead of using the value at the minimimum of the majorization function, the value ``opposite'' the majorization point is used. This can essentially cut the number of iterations necessary to reach the minimum in half.

Parameters

[in]

model

GenModel containing the augmented parameters

Definition at line 206 of file gensvm_optimize.c.