[Bug]: Missing MultiTask API in the latest version and documentation

[chanansh]

What happened?

Earlier versions supported MultiTask which allows transfer learning . see https://ax.dev/docs/0.5.0/tutorials/multi_task/
Today, the new version 1.1.2 does not show any mentioning of this capability neither in the tutorial or the api documentation.

This feature is important for using past experiences on one system and using this information to optimize a new system which follows similar dynamics.

Please provide a minimal, reproducible example of the unexpected behavior.

N/A

Please paste any relevant traceback/logs produced by the example provided.

N/A

Ax Version

1.1.2

Python Version

3.12

Operating System

macOS N/A

(Optional) Describe any potential fixes you've considered to the issue outlined above.

I would expect the new Client API to support multi-task and or transfer learning.

Pull Request

None

Code of Conduct

• I agree to follow Ax's Code of Conduct

[chanansh]

@amyreese @dreiss @javache @zpao

[Balandat]

@chanansh Ax still does support multi-task modeling / transfer learning under the hood.

@mpolson64, @saitcakmak, @hvarfner can you provide some context on why this isn't documented / what the plan is to change that? And in the meantime point at the currently recommended way to do transfer learning / multi-task optimization?

@chanansh to unblock things in the short term, my guess is that the tutorial from 0.5.0 that is referenced above should still work in principle, but will require some adaptation to API changes made in 1.0.0.