Validate inputs before training starts

[AtheerAS]

Description of feature

Thanks for the amazing work on MOFA-Flex!
One issue I ran into: after ~1.5h of training, the run failed at the saving step because one of the inputs, save_path, wasn’t valid.
It would be super helpful if all inputs could be validated at the very beginning (e.g., file paths, formats, etc.), so that small mistakes don’t cause an entire training run to be wasted.

MOFA-Flex version: 0.1.0rc2

Thank you!

[ilia-kats]

In particular for save_path, the status can change during the training run (e.g. network shares being mounted/unmounted), so it's impossible to guarantee validity before training. But if you can post the exact error you got, we can try to make it a bit more robust.

[AtheerAS]

I used Path from pathlib in save_path to set the output directory:

output_path = Path("output_path")
save_path= (output_path/ "model.h5mu")

That’s when the error occurred. Unfortunately I was too quick to restart the run and didn’t copy the exact error message.