# Nelder Mead tolerance vastly higher when I restart optimization with updated guess for initial parameters

**URL:** https://discourse.julialang.org/t/nelder-mead-tolerance-vastly-higher-when-i-restart-optimization-with-updated-guess-for-initial-parameters/87939
**Category:** General Usage
**Tags:** statistics, optim, optimization
**Created:** [September 28, 2022, 3:17pm UTC](https://discourse.julialang.org/t/nelder-mead-tolerance-vastly-higher-when-i-restart-optimization-with-updated-guess-for-initial-parameters/87939 "2022-09-28T15:17:59Z")
**Posts on this page:** 1
**Showing post:** 8

<div class="post-metadata">

### Author: ![amrods](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/amrods/32/2543_2.png) [@amrods](https://discourse.julialang.org/u/amrods)
#### Post date: [September 29, 2022, 11:30am UTC](https://discourse.julialang.org/t/nelder-mead-tolerance-vastly-higher-when-i-restart-optimization-with-updated-guess-for-initial-parameters/87939/8 "2022-09-29T11:30:00Z")

</div>

As I understand, in general you cannot parallelize optimization algorithms. You could parallelize the objective function as you have done. In principle, however, you can parallelize Nelder-Mead as you mention, but I don’t think `Optim.jl` supports that. The book I linked above has sample code for Nelder-Mead, which you can modify and parallelize at will. Another suggestion I have is to try to optimize the computation of the objective function and log-likelihood and use speed-up tools like `LoopVectorization.jl`. See here [How to speed up these functions?](https://discourse.julialang.org/t/how-to-speed-up-these-functions/65728) for example.

---

_[View the full topic](https://discourse.julialang.org/t/nelder-mead-tolerance-vastly-higher-when-i-restart-optimization-with-updated-guess-for-initial-parameters/87939)._