Motivation

I am writing a hyperparameter optimization tool that allows (almost) no-code hyperparameter optimization with ax/botorch on HPC systems. I'd love to use other models than the ones provided by ax/botorch, for example, TPE. But in the end, I'd love to have a way of using external hyperparameter optimizers, without the need to change the code, for generating new points. Currently, it's not easily possible to do this, even with ExternalGenerationNode and this tutorial.

https://ax.dev/docs/tutorials/external_generation_node/

Problem there is that, ax_client.to_json_snapshot() doesn't work, and I cannot find a way of writing an encoder that works, but I require the json snapshot to work.

Describe the solution you'd like to see implemented in Ax.

Two things I'd like:

1. Implement something based on TPE in Ax/Botorch
2. Expand the tutorial so it's more clear how to integrate the Random-Forest-Example. I.e. make the to_json_snapshot work, support MOO, support different type of parameters.

1, I think, is optional, as with 2 I can implement it by myself. But I'd really love to see the tutorial for 2 expanded so that it is more useful for more complex applications.

Describe any alternatives you've considered to the above solution.

No response

Is this related to an existing issue in Ax or another repository? If so please include links to those Issues here.

No response

Code of Conduct

• I agree to follow Ax's Code of Conduct