Outcome transform in case of data with task indices

[sheikhahnaf]

Hi,
I am trying to use input and outcome transform in multitaskgp having task indices. In my case, I am using botorch.models.transforms.outcome.Standardize. But it requires m arg as dimension of output. But, with task indices, the ouput is flattened and it does not have columns corresponding to tasks. How to get around this? Is there solution already existing for this?

[saitcakmak]

Hi @sheikhahnaf. Outcome transforms do not differentiate between different tasks in MTGPs. If you'd like to use standardize each task independently, you'd have to do this manually. In Ax, we do this using StratifiedStandardizeY

[sheikhahnaf]

thank you so much @saitcakmak