Optim.jl
Optim.jl copied to clipboard
What is Optim's mechanism for calculating the gradient?
When using the LBFGS algorithm within Fminbox, what is Optim's mechanism for calculating the gradient if there is no gradient vector for a given objective function? Can you give a detailed description?
It depends on whether you choose an ad algorithm or not. If you do not, it's finite differencing. See some more information here https://julianlsolvers.github.io/Optim.jl/dev/user/gradientsandhessians/ . Does that answer your question?