Extensions not checked or added to saved file names

[colleenjg]

Currently, saving a Data object takes a filepath and doesn't check for or add a .h5 extension.

• torch_brain/data/data.py:L485

Similarly for config saving: does not check for yaml (or yml) extension:

• torch_brain/pipeline/config.py:L30

Whatever file type name you pass, it'll use it (with an extension, with the wrong extension).

I think it could be worth handling the following cases:

• No extension: We could automatically add the correct extension.
• Wrong extension: Raising a warning? (Or an error?)

Accordingly, we could also do something similar when loading Data or configs:

• No extension: Add the extension before checking? (Or do we want to first check whether a file with no extension exists to be loaded?)
• Wrong extension: Raise a warning before trying to open? (Or an error?)

This is not necessarily a major problem, since extensions don't matter within the scope of torch_brain. But, this would increase the likelihood of users generating files with recognizable, correct extensions. And certainly, for Windows users, file extensions do matter more.

[AlexandreAndr]

I think saving a Data is an important case, for the config I think it corresponds to the brainset config (that you can set with cli, e.g. setting the raw/processed dir) and might be less important because not directly exposed to users (we can still check it but can require less thinking)

No extension: We could automatically add the correct extension.

so if someone does

data.save("data")` # save as `data.h5

• Do we want to inform the user with a small message (I would vote for no)?
• Do we want to support not given extension when we read the files?

data = Data.load("data") # would load  `data.h5`

Wrong extension: Raising a warning? (Or an error?)

Maybe if we start handeling the extension we can err when the extension is not correct

[colleenjg]

Ok, after discussion in the PR, we settled on:

Save behavior:

* If user specified filename with some extension, just use that. Don't warn/error-out if it is different from `.h5`.

* Add `.h5` extension if user specified filename without any extension. Additionally, do an INFO log saying that the extension was added.

Load behavior:

* The user has to specify full filename. We will not tack on `.h5` from our side.

Additionally, in the docstrings of both Data.load and Data.save: add the fact that we save/load ".h5" files in the first line (like how it is written in the docstrings of numpy.save and numpy.load).