Feature request: Seeking stopval or callback, to give greater control over stopping criteria

[donboyd5]

Hi,

I am calling LsqFit.levenberg_marquardt directly because I have found it to outperform other Julia least-squares optimizers.

My problem is potentially time consuming and I am satisfied with results that stop well before default g_tol and x_tol are reached. I find that I cannot adjust these tolerances in a way that stops at a point I like.

I would love to be able to tell the algorithm to stop when a specific objective function (sum-of-squares) value is reached (similar to stopval in NLopt.jl) or to use a callback function that returns a boolean that stops the algorithm at a point I define, similar to what is allowed in Optim.jl.

As far as I can tell neither is possible in LsqFit.levenberg_marquardt (if I am wrong and someone could tell me how to do this, I would appreciate it).

Anyway, I would love it if you could add either a callback feature or a stopval parameter to LsqFit.levenberg_marquardt.

Many thanks.

[pkofod]

Yes, I agree.