Closed momchil-flex closed 11 months ago
Forgot to mention this can use a third-party optimizer. This seems pretty well established.
Yeah I can work on it in the next few weeks if Emerson is busy with something else.
A few customers have asked about PSO and pyswarms have been my recommendations for them as well. Should be relatively straightforward to incorporate it with Tidy3D I think. One question is can we simply use public packages like this in our published materials? I guess we have been using things like pandas
and scipy
but in general is it ok to directly use these for commercial purposes?
It's under MIT license which permits commercial use so it should be good to go?
Yes, anything that's not GPL is generally fine.
Nice! I've suggested including these global optimization features and Monte Carlo simulations as new study types in GUI in the future. This notebook will help a lot as a starting point.
On Mon, Aug 28, 2023 at 11:30 PM momchil-flex @.***> wrote:
Yes, anything that's not GPL is generally fine.
— Reply to this email directly, view it on GitHub https://github.com/flexcompute/tidy3d/issues/1107#issuecomment-1696677163, or unsubscribe https://github.com/notifications/unsubscribe-auth/A55TUUSJGCZDQQLDQL25K6LXXVH5VANCNFSM6AAAAAA4CG7RXM . You are receiving this because you were assigned.Message ID: @.***>
Sounds good to me and I am in favor of using external optimization packages (eg pyswarms) whenever possible. Just remember to add it here: https://github.com/flexcompute-readthedocs/tidy3d-docs/blob/develop/docs/requirements.txt
The PBS notebook has been merged and the bullseye extractor is about to be merged too. We can close this issue now.
It would be great to add one or a few examples of particle swarm optimization to our notebooks. There is a lot of interest in that because I think it can be pretty good to find big improvement in designs that rely on a small number of parameters that are still too many to handle in a full parameter scan. I think we can even pick one of our existing PIC examples, ideally a single simulation should be small-ish, and we should optimize 3-5 global parameters so that the total number of simulations that have to be done is manageable for an example.
Any other thoughts @tomflexcompute @tylerflex?