Use of numerical gradient about roptim

[ypan1988]

Yes, you can use Nelder-Mead. Yi

…

________________________________ From: ***@***.*** ***@***.***> Sent: 26 June 2022 21:56:35 To: ypan1988/roptim Cc: Subscribed Subject: [ypan1988/roptim] Use of numerical gradient (Issue #5) Hi, I have a custom function for which there is no analytical gradient that I can find. Is it still possible to use this package to do optimization in Rcpp (without defining a gradient function)? Thanks, Teng — Reply to this email directly, view it on GitHub<#5>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AB74TW3W4N5LENPQY6L43UDVRC7YHANCNFSM5Z4QJ6PA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[teng-gao]

Hi Yi,

Thanks for the reply. To be more precise, I'm trying to replicate the behavior of the below code in R, where L-BFGS-B is used, but no gradient function is provided.

optim(
    start, 
    function(x) {f(x)},
    method = 'L-BFGS-B',
    lower = lower,
    upper = upper,
    hessian = TRUE
)

Thanks,
Teng

[ypan1988]

Yes, it is possible and please check the following paper:
https://cran.r-project.org/web/packages/roptim/vignettes/article.pdf

Yi

[teng-gao]

Got it - this is super useful for our work! I'm trying to parallelize many individual optim calls using RcppParallel. Do you think Roptim would be thread-safe enough to do so?

[ypan1988]

Sorry, I am not very sure about this. But my understanding is that if you are calling optim in parallel, that means you are calling the underlying C functions of optim in parallel. Roptim is simply a wrapper of the C functions to make it easier to write code in C++, so maybe create a separate Roptim object for each thread?! Do consult someone with more experience with parallel programming.