Improved parameter sweep support for `TerminalComponentModeler` (TCM) and `ComponentModeler` (CM)

[yuanshen-flexcompute]

Current Issue

Since TerminalComponentModeler (TCM) and ComponentModeler (CM) are batch jobs, there is no simple way to do a parameter sweep with them (cannot use web.Batch). The only way I can think of at the moment is to use a manual for loop to execute each TCM/CM, which does not make good use of our parallel infrastructure.

Why this is necessary

Due to the introduction of WavePort and LumpedPort, the TCM will become the primary method for RF users to execute simulations. They will need an easy way to sweep parameters (e.g. geometry) to see how the S-matrix changes.

Proposed solution

I do not currently have a good proposed solution. I think it would be necessary to discuss whether we want to modify web.Batch to allow inclusion of TCM/CM, or whether to modify TCM/CM to allow for some sort of parameter sweep functionality. I think it would also be good to check in with the RF GUI folks to see how they plan to implement parameter sweeps in the front end.

[tylerflex]

This is something that could be handled once we have
#2011

[daquinteroflex]

@dmarek-flex, re our conversation, this could be an example of where having a fully functional approach between the terminal_component_modeller(*parameters) -> list[Simulations] might be more powerful than having to support each new set of operation classes

[Frank-DingYC]

Additionally, we will encounter the same problem when attempting to perform parameter optimization (Particle Swarm Optimization, Genetic Algorithm, etc.). @tylerflex

[dmarek-flex]

@dmarek-flex, re our conversation, this could be an example of where having a fully functional approach between the terminal_component_modeller(*parameters) -> list[Simulations] might be more powerful than having to support each new set of operation classes

After our discussion I had a simple idea that I think would be a good solution for the modelers in the future. That is to make it just like a Simulation and SimulationData pair, so have a Modeler and ModelerData pair,. The ModelerData would have the original Modeler as a field along with the BatchData and any post processing functionality, which is probably close to what you are suggesting. This approach would be more inline with the other classes in Tidy3D.

This wasn't exactly what we were talking about, but I do think it would help.