LineSearches.jl

LineSearches.jl copied to clipboard

HZ: avoid convergence if bisected

26

`secant2` calls `update`, and `update` might switch to bisection in the U3 step (as named in the initial HZ publication). If we did bisect, the line-search "are we making enough...

timholy

Corrected, improved and cleaned BackTracking

4

There is a small mistake in `BackTracking.jl` when computing `α_tmp` with the cubic interpolation: when the leading coefficient is close to zero, the expression of `α_tmp` misses a negative sign,...

AlexandreMagueresse

Suboptimal choice of alpha in Hager-Zhang

10

I've encountered a weird issue with alpha value selection in Hager-Zhang. Let me illustrate. Here are the values computed during bracketing: ```julia alphas = [0.0, 1.0, 0.5, 0.25, 0.11797881082031754, 0.05898940541015877,...

mateuszbaran

Would an AbstractLineSearchType be a good idea?

2

I am using `LineSearches` in one of my packages – and just to be able to write something generic an abstract linesearch type for all linesearch structs would be great....

kellertuer

I have implemented a line search method based on majorize-minimize (MM) principles for cost functions of the form $f(x) = \sum_j f_j(A_j x)$ where each $f_j$ has a Lipschitz gradient...

JeffFessler

Rationale to dimiss strong wolfe linesearch

1

Hello, Thank you for all your work! I was wondering why in the docstring of the strong wolfe linesearch the authors say the following. ``` This algorithm is mostly of...

vroulet

Hager-Zhang converges incorrectly within Optim.ConjugateGradient due to flatness check

3

I found a case where the ConjugateGradient method of Optim.jl terminates early, at a non-stationary point, without raising an error. After some digging, the problem seems associated with LineSearches. I...

kbarros

Add test cases, disable flatness check in HZ

5

This may cause problems of its own, but for the time being it's better than the status quo. Fixes #173 Closes #174 Fixes #175 I am pretty sure that without...

timholy

Julia Formatter, removed zeroT and replaced it with zero(T), changed x == nothing to isnothing(x) and added check if phi_x is finite in Backtracking

1

I am doing a few readability improvememts of this package and will try to refactor StrongWolfe and work off the Todos that are still in the code. This is the...

theplatters

This is implicit in #81, but just to document it: ```julia julia> using LineSearches, Optim julia> f(x) = sum(abs2, x) f (generic function with 1 method) julia> optimize(f, [0.5], [1.5],...

timholy