Validation functions

[dfulu]

This pull request adds a draft of the model validation procedure. This includes a data loader to load samples of data.

There are some things that still might need to be addressed and/or discussed:

• Is there a better data structure than a yaml to include the test / validation t0 times
• API improvements
• File structure for validation.py - should it be split up
• Speed up the implementation of measuring the structural similarity. Right now it uses the version from scikit-image. This leads to a nested loop which could be improved
• Ideally we remove the dependency on ocf_datapipes
• Linting and style

[dfulu]

This had to be removed because of a warning stemming from ocf_datapipes

[dfulu]

Adding these extra model parameters to wandb hasn't yet been implemented

[phinate]

Thanks for all your work here @dfulu! Some of your tasks definitely deserve their own issues to collect comments.

This is quite a large PR, so it's a bit difficult to parse all the moving parts. How about we split this into separate PRs for (+ relevant tests + examples):

• Dataset + Dataloaders
• validation functionality
• wandb integration

How does that sound? This way, we can make sure each part is properly linted + typechecked as we merge it in. We'll also want to make sure we're doing shape-validations too as we go through. I'm completely open to the nature of the splits you'd prefer semantically!

(also, sorry for the late message. Please do not reply outside of your working hours!)

[dfulu]

@phinate, yeh sure. It is rather big. I'll start working to split it up into more manageable chunks