Releases: JuliaNLSolvers/LineSearches.jl
Bugfix: Missing Compat in REQUIRE
v5.2.1 Update REQUIRE
Ensure finite values with Static
Ensure function value is finite for Static step (#101) We should probably add some sort of debug / verbosity flag here to warn if this happens
Revamp Static
Static no longer takes the alpha
and scaled
arguments, but instead takes the step provided at each call to the line search. The old functionality can still be accessed by combining Static
with InitialStatic
for the initial step guess.
Fix InitialStatic bug
Merge pull request #98 from JuliaNLSolvers/fixinstat Fix InitialStatic bug
Better arbitrary precision, new native complex support, and v0.7-pre compatibility.
Update output of value_gradient! (#96) * Update output of value_gradient! * Fix value_gradient call output in alpha test.
A simplified LineSearches
This major version of LineSearches introduces a slim-lined internal machinery of LineSearches.jl. The actual line searches no longer depend on AbstractObjective
s from NLSolversBase.jl, but only on univariate functions for phi, the derivative of phi and a composite version.
This version is breaking, as it removes some of the Optim specific things such as LineSearchResults from the package. This makes the individual line searchers more stand-alone in nature.
The initial guesses are still tied to the JuliaNLSolvers setup.
Update for AbstractArray input
Allow AbstractArray instead of Array to support more input types.
Use vecnorm
Use vecnorm instead of norm on gradients and decision vectors. Fixes problems with optimization with arbitrary inputs (3-dimensional arrays etc.)
Fix DomainError bug
v3.2.3 Fix DomainError and calculate number of f/g evaluations correctly in …
Fix bug in More-Thuente
Fixes bug in the translation from the original MATLAB/FORTRAN code of More-Thuente.