Broken dependency in design space Bayesian optimization

[yuanshen-flexcompute]

Describe the bug
The Bayesian Optimizer MethodBayOpt relies on an external library bayesian-optimization. The external library changed their API in an update that breaks dependency with MethodBayOpt. The most recent version of bayesian-optimization that still works is 1.5.1.

Version history: https://github.com/bayesian-optimization/BayesianOptimization/releases
Update that broke the dependency: bayesian-optimization/BayesianOptimization#447

To Reproduce
Steps to reproduce the behavior:

1. Install the latest version of bayesian-optimization library
2. Open a notebook that uses MethodBayOpt or set up a dummy DesignSpace.
3. Execute the design space using run().

Expected behavior
The design space run() fails with an error message about missing bayes_opt library.

Additional context
The version 1.5.1 of bayesian-optimization has been tested to work successfully without error. Since this is an external library that the user is expected to install by themselves, we can update our documentation to emphasize that they should install the older 1.5.1 version.

[yuanshen-flexcompute]

Per Tom's suggestion: Update the line

from

"Cannot run Bayesian optimization as 'bayes_opt' module not found. Please check installation or run 'pip install bayesian-optimization'."

to

"Cannot run Bayesian optimization as 'bayes_opt' module not found. Please check installation or run 'pip install bayesian-optimization==1.5.1'."

[yaugenst-flex]

Hmm I thought we addressed this in #2281, but it doesn't apply to packaging... I can change the message, but more generally we should probably start bundling packages for packages @daquinteroflex?

[daquinteroflex]

Bundling might be an interesting approach, but also it's a question if people want to share tidy3d dependencies within their own environments and whether we want to support that or have tidy3d work in a kind of self contained way via dependency bundling into executales. I'm also curious how that'd work for distribution in pypi in terms of size. Another aspect is compatibility on the backend with bundling, which could be tricky in the short run but possible in the long run - but we can chat about that via slack.

Maybe for now, we dont have a tidy3d[design] when I've just checked. I guess just change this pyproject.toml requirement line

tidy3d/pyproject.toml

Line 81 in 42afa41

bayesian-optimization = { version = "<2", optional = true }

and go from there. Eventually these "plugin" dependencies groups can become their own pyproject.toml if we decide to split them off at some point I guess too, or even could be a bundled package? Not sure how that'd work?

[yaugenst-flex]

Maybe for now, we dont have a tidy3d[design] when I've just checked. I guess just change this pyproject.toml requirement line

yeah maybe there was a misunderstanding, but this is what I meant