Using a custom learner in FLAML

[hari-ushankar]

Hello folks at FLAML,

A really impressive project and happy to have found this. A general request I had was if FLAML could take some of the models in gpytorch (https://docs.gpytorch.ai/en/v1.13/models.html) to create custom learner and do some automatic model building with various hyperparameters?

Is there a docpage to extend the functionalities of FLAML this way? I can't find anything for this feature at the moment.

[thinkall]

Hi @hari-ushankar, thanks for reaching out. Yes, FLAML can support customized learners. Please check out the doc here.

[hari-ushankar]

Hey Jiang @thinkall , thanks for the reference to the docpage. I should have been more specific about the extending to different learners.

Would it be possible to add Gaussian Process Regression from gpytorch into the FLAML's list of learners? Here is an example to a gpytorch regression example: https://docs.gpytorch.ai/en/v1.13/examples/01_Exact_GPs/Simple_GP_Regression.html

My naive thinking would be that it is possible, but I do like your opinion on this. Any resource such as docpage, blogs or a youtube video would be really helpful. Thank you. 😃

[thinkall]

Hi @hari-ushankar , do you mean search hyperparameters for a gpytorch model? Is this doc the example you're looking for?