Should `fetch_fsaverage()` return a `Path`?

[hoechenberger]

Currently, mne.datasets.fetch_fsaverage() returns a string, while all datasets.*.data_path() functions return a Path. I find this inconsistent, esp. in tutorials, where we'd want to use both, fetch_fsaverage() and data_path(), within a few lines.

Thoughts?

[cbrnr]

I'm always for improving consistency, so if that's what other functions are doing I'm 👍. In addition, it might be worth looking into the mne.datasets package, because to me it seems to be a bit confusing since there are many functions with possibly overlapping functionality and/or naming:

• mne.datasets.fetch_dataset(): Is this supposed to be the wrapper for downloading each and every dataset?
• If that's the case, why is there a separate mne.datasets.fetch_fsaverage()?
• What about the various mne.datasets.XXX.data_path() functions? They also fetch datasets if they are not available, right?
• Then there is mne.datasets.eegbci.load_data() and mne.datasets.limo.load_data() – why not .data_path()?

Maybe there are more inconsistencies or things needing clarification, but that's all I could find for now. Thoughts?

[hoechenberger]

• Then there is mne.datasets.eegbci.load_data() and mne.datasets.limo.load_data() – why not .data_path()?

I think these quite literally load some data, but I'm not exactly sure

[cbrnr]

I think these quite literally load some data, but I'm not exactly sure

Yes, but it just returns the path like all the other .data_path() methods. From the docs: "Get paths to local copies of EEGBCI dataset files. This will fetch data for the EEGBCI dataset, which is also available at PhysioNet."

[drammock]

Some datasets have the option of only downloading certain participants or runs. IIRC those are the ones with "load_data" funcs. Anything where we always download the entire dataset (e.g. because it's stored as a monolithic tar file) gets a "data_path" instead

[hoechenberger]

i'd suggest that for now, let's limit the scope of this issue to fetch_fsaverage() please.

[larsoner]

Argh must have missed it in #10348. +1 for changing, but for backward compat we should probably do what we did there which was make use of specialized Path subclasses to allow people to keep doing + for one release:

https://github.com/mne-tools/mne-python/pull/10348/files#diff-a176c176c64bba7aedc67f81f2ab701041ac293e7c1f6305f8a49e536cf97f1dR529

@hoechenberger are you up for doing this? Would be good to get it into 1.8

[hoechenberger]

@larsoner I don't think I'll have time to tackle this within the next 2 weeks, sorry!

[larsoner]

Okay I'm on it then