zhangyiss/liblbfgs
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Release for 1.8.

Browse Source 
git-svn-id: file:///home/svnrepos/software/liblbfgs/trunk@60 ecf4c44f-38d1-4fa4-9757-a0b4dd0349fc
This commit is contained in:
naoaki
2009-07-13 14:27:47 +00:00
parent 896f8cbc7c
commit 80589797a3
4 changed files with 25 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ChangeLog
View File

@@ -1,3 +1,10 @@
2009-07-13 Naoaki Okazaki <okazaki at chokkan org>
* libLBFGS 1.8:
- Accepted the patch submitted by Takashi Imamichi; the backtracking method now has three criteria for choosing the step length.
- Updated the documentation to explain the above three criteria.
2009-02-28 Naoaki Okazaki <okazaki at chokkan org>
* libLBFGS 1.7:

2
configure.in
View File

@@ -18,7 +18,7 @@ AC_CONFIG_SRCDIR([lib/lbfgs.c])
dnl ------------------------------------------------------------------
dnl Initialization for automake
dnl ------------------------------------------------------------------
AM_INIT_AUTOMAKE(liblbfgs, 1.7)
AM_INIT_AUTOMAKE(liblbfgs, 1.8)
AC_CONFIG_HEADERS(config.h)
AM_MAINTAINER_MODE

25
include/lbfgs.h
View File

@@ -155,21 +155,24 @@ enum {
/** MoreThuente method proposd by More and Thuente. */
LBFGS_LINESEARCH_MORETHUENTE = 0,
/**
* Backtracking method with the Armijo condition:
* Backtracking method with the Armijo condition.
* The backtracking method finds the step length such that it satisfies
* the sufficient decrease (Armijo) condition,
* f(x + a * d) <= f(x) + lbfgs_parameter_t::ftol * a * g(x) \cdot d,
* - f(x + a * d) <= f(x) + lbfgs_parameter_t::ftol * a * g(x)^T d,
*
* where x is the current point, d is the current search direction, and
* a is the step length.
*/
LBFGS_LINESEARCH_BACKTRACKING_ARMIJO = 1,
LBFGS_LINESEARCH_BACKTRACKING = 1,
/** The backtracking method with the defualt (regular Wolfe) condition. */
LBFGS_LINESEARCH_BACKTRACKING = 2,
/**
* Backtracking method with regular Wolfe condition.
* The backtracking method finds the step length such that it satisfies
* both the Armijo condition (LBFGS_LINESEARCH_BACKTRACKING_ARMIJO)
* and the curvature condition,
* g(x + a * d) \cdot d >= lbfgs_parameter_t::wolfe * g(x) \cdot d,
* - g(x + a * d)^T d >= lbfgs_parameter_t::wolfe * g(x)^T d,
*
* where x is the current point, d is the current search direction, and
* a is the step length.
*/
@@ -179,7 +182,8 @@ enum {
* The backtracking method finds the step length such that it satisfies
* both the Armijo condition (LBFGS_LINESEARCH_BACKTRACKING_ARMIJO)
* and the following condition,
* |g(x + a * d) \cdot d| <= lbfgs_parameter_t::wolfe * |g(x) \cdot d|,
* - |g(x + a * d)^T d| <= lbfgs_parameter_t::wolfe * |g(x)^T d|,
*
* where x is the current point, d is the current search direction, and
* a is the step length.
*/
@@ -587,21 +591,22 @@ This library is used by:
@section download Download
- <a href="http://www.chokkan.org/software/dist/liblbfgs-1.7.tar.gz">Source code</a>
- <a href="http://www.chokkan.org/software/dist/liblbfgs-1.8.tar.gz">Source code</a>
libLBFGS is distributed under the term of the
<a href="http://opensource.org/licenses/mit-license.php">MIT license</a>.
@section changelog History
- Version 1.8 (2009-07-13):
- The backtracking method now has three criteria for choosing the step
- Accepted the patch submitted by Takashi Imamichi;
the backtracking method now has three criteria for choosing the step
length:
- ::LBFGS_LINESEARCH_BACKTRACKING_ARMIJO: sufficient decrease condition
(Armijo condition)
- ::LBFGS_LINESEARCH_BACKTRACKING_ARMIJO: sufficient decrease (Armijo)
condition only
- ::LBFGS_LINESEARCH_BACKTRACKING_WOLFE: regular Wolfe condition
(sufficient decrease condition + curvature condition)
- ::LBFGS_LINESEARCH_BACKTRACKING_STRONG_WOLFE: strong Wolfe condition
This is based on the patch submitted by Takashi Imamichi.
- Updated the documentation to explain the above three criteria.
- Version 1.7 (2009-02-28):
- Improved OWL-QN routines for stability.
- Removed the support of OWL-QN method in MoreThuente algorithm because