Replacing Nelder-Mead with Hooke-Jeeves in MCCGA

jmejia8 / Metaheuristics.jl

High-performance metaheuristics for optimization coded purely in Julia.

Other

252 stars 27 forks source link

Closed jbytecode closed 1 year ago

jbytecode commented 1 year ago

I am going to send a pull request that replaces Nelder-Mead local search part with Hooke-Jeeves in MCCGA algorithm.

This PR gains these items:

Reduces the dependencies of outer packages
Hooke-Jeeves is not only a local search algorithm but is a direct-search algorithm
Hooke-Jeeves does not use derivatives

jmejia8 commented 1 year ago

What a great contribution! It would be nice to have Hooke-Jeeves (Pattern Search) as another implemented algorithm.

optimize(f, bounds, PatternSearch())

Since the Hooke-Jeeves method and other methods require a starting solution, I opened https://github.com/jmejia8/Metaheuristics.jl/issues/46 to solve those possible scenarios.

jbytecode commented 1 year ago

Yes, we can separate hj() from the MCCGA files and implement it as an independent algorithm, of course in the next step.

jmejia8 commented 1 year ago

I'm fine with that.

jbytecode commented 1 year ago

Since the PR is merged, we can close this issue for now.

When we get ready to implement Hooke & Jeeves as a separate algorithm, we can create a new one if needed.