braindecode
braindecode copied to clipboard
Decrease the dependencies on MOABB
Originally we thought of MOABB as a soft dependecy; a nice way to fetch data and start right away. We should probably remove MOABB from our tests and from most of our examples.
if you put it in examples it's a soft dependency. Yes for tests it's maybe an overkill. Having one good dataset like sample or testing (which is a small sample) we use in mne test is to me a good trade off
having moabb in example to demo on real dataset without having to maintain here the fetchers seems a good idea to me
Related, in #196 MOABB as of now requires jumping through a lot of hoops to get working.
@OleAd MOABB now requires fewer hoops to get working than before, see #201.
We still use MOABB in quite some tests (dataset/windower tests) but do we consider this a problem? to me seems ok or at least not high priority to remove, what do you think @gemeinl ...
Hi, MOABB should be easier to integrate now. Let me know how we could help to ease the process. We also have refactor the code to download datasets, it should be more reliable and stable.
We could still see if we could remove moabb dataloading from tests and replace it with fake datasets just to make things more decoupled
Somehow it seems on a fresh install moabb will fail if you do not create ~/mne_data ? this made our CI fail temporarily... is this intended, wasn't it different before? @sylvchev
No, it is a bug! We simplify the storage of datasets, and this error passed through our CI. I'll push a fix on MOABB right now.
Ok, it is corrected : https://github.com/NeuroTechX/moabb/pull/204
great
Hello @gemeinl, @agramfort, @robintibor, @sylvchev,
I understand that this issue has been discussed and debated. Can we close this issue?