Add tensor noise amplitude parameters for perturbation methods

[dallasfoster]

Earth2Studio Pull Request

Description

This is a backward compatible PR that adds the ability to pass broadcastable tensors as noise amplitudes in relevant perturbation methods.

Checklist

• [x] I am familiar with the Contributing Guidelines.
• [x] New or existing tests cover these changes.
• [x] The documentation is up to date with these changes.
• [x] The CHANGELOG.md is up to date with these changes.
• [ ] An issue is linked to this pull request.

[NickGeneva]

Currently there's not mechanism to ensure the noise vector is on the same device as the input tensor.

I would maybe suggest converting floats to a tensor in the inits.

Then inside each call function, the amplitude is moved onto the same device as x.device is it does not match at the moment.

Because the workflow determines the device, we shouldn't assume that the user will know what the device is at the time of construction.

[dallasfoster]

Currently there's not mechanism to ensure the noise vector is on the same device as the input tensor.

I would maybe suggest converting floats to a tensor in the inits.

Then inside each call function, the amplitude is moved onto the same device as x.device is it does not match at the moment.

Because the workflow determines the device, we shouldn't assume that the user will know what the device is at the time of construction.

Thanks for the catch, just pushed an update to include these suggestions.

[NickGeneva]

LGTM

[dallasfoster]

/blossom-ci