ValueError: Raw data must be data for a single arm for non batched trials. for the below code which is part of the tutorial. for v0.2.6 about ax

[mpolson64]

@leeauk21 While I can't go back and try and reproduce this error, I believe it may have be related to this issue from June: #1664 . The TL;DR is that Typeguard, the library we were using at the time for runtime type checking, was not working properly and causing strange errors deep in our stack -- this was quite a tricky issue to investigate and we ended up needing to remove Typeguard completely.

If you are unable to upgrade to the latest version of Ax the best advice I can give you is to grep for TRIAL_RAW_DATA_FORMAT_ERROR_MESSAGE (the string of the value error you're seeing) and remove places where it's raised or where Typeguard is used. If I recall correctly, the primary place will be in ax/core/trial.py where we call a method named checked_cast_complex that calls Typeguard under the hood. Removing that checked_cast_complex and just passing in the value directly should solve your issue.

I'm sorry I can't offer more help, but I am sympathetic to your inability to upgrade to to legacy code so I hope this helps. Best of luck!

[mpolson64]

Hi @leeauk21, unfortunately I'm unable to reproduce this error using the latest release of Ax, v0.3.4 . Please update to this version and the above snippet should work.

If you're still encountering issues after updating don't hesitate to reopen the task and we will take a look.

[leeauk21]

hey Im stuck in a large codebase so im still stuck in python 3.7. Would it be remotely possible for you or others to try the above code with python 3.7 and ax v0.2.6. Thanks!

[leeauk21]

Thanks a lot @mpolson64 !!!