LBFGS++ v0.4.0

Added

• Added functions final_approx_hessian() and final_approx_inverse_hessian() to LBFGSSolver to retrieve the final approximate Hessian information (#42, #43)
• Added CMake scripts to build examples (#38), contributed by @pjknowles

LBFGS++ v0.3.0

Added

• Added functions final_grad() and final_grad_norm() to LBFGSSolver and LBFGSBSolver to retrieve the final gradient information (#12)

Changed

• LBFGS++ now requires C++11
• The line search classes now have a unified API for both LBFGSSolver and LBFGSBSolver
• The Moré-Thuente line search algorithm LineSearchMoreThuente now can also be used in the L-BFGS solver LBFGSSolver
• Improved the numerical stability of LineSearchNocedalWright (#27)
• Removed the unused variable dg_hi in LineSearchNocedalWright (#35)
• Fixed some compiler warnings regarding shadowed variables (#36)

LBFGS++ v0.2.0

Added

• Added a CMake script for installation (#24), contributed by @steinmig

Changed

• The default line search method for LBFGSSolver has been changed from LineSearchBacktracking to LineSearchNocedalWright, per the suggestion of @mpayrits (#25)
• Fixed a few critical issues (#9, #15, #21), with big thanks to @mpayrits (#25)
• Fixed one inconsistency with Moré and Thuente (1994) in the LineSearchMoreThuente line search algorithm, pointed out by @mpayrits (#23)
• The source code is now formatted using Clang-Format

LBFGS++ v0.1.0

Initial Github release.