Require datasets to be in mir-datasets db before adding loader

[alastair]

Just thinking out loud, as a way to try and improve the dissemination of mir datasets. There has already been some work in the ismir community to make a public list of datasets: https://github.com/ismir/mir-datasets It would be nice to require as part of the checklist of accepting a new dataset/loader here to require that the dataset is also added to mir-datasets so that it appears on the list at http://www.audiocontentanalysis.org/data-sets/

[rabitt]

I like this idea a lot. We can also create a 'checklist' of each of the datasets we've included and which are not yet included to encourage new contributions :)

[magdalenafuentes]

We'll embed the mir-datasets.yaml file on the mirdata repo, and add a column that indicates if the dataset has a mirdata loader or not. Also as part of the checklist for adding a loader you will now have to indicate the loader availability in this yaml file.

[rabitt]

Adding a to-do list here:

• [x] Audit if our supported datasets are on the mir_datasets list
• [x] add missing datasets
• [x] update contributing docs

[lostanlen]

TinySOL was missing from the list. I requested to add it. https://github.com/ismir/mir-datasets/pull/25

DALI is also an open PR: https://github.com/ismir/mir-datasets/pull/23

[rabitt]

Hmm yeah it looks like the turnaround time for adding to the ismir/mir-datasets repo is a bit long, and it might not makes sense for this to be a strict requirement. Maybe we can ask for admin access to that repo so we can approve/add repos to the list?

[lostanlen]

yes, i would be in favor of that. at the moment, faroit and ejhumphrey are the only contributors, and there have been no commits since November 15th, 2019.

[lostanlen]

DALI and Tinysol will be in mir-datasets once https://github.com/ismir/mir-datasets/pull/27 get merged

[magdalenafuentes]

This didn't go to far at the end. Thoughts on this moving forward? @rabitt @nkundiushuti

[nkundiushuti]

Since the process of creating a loader is slower than adding a dataset in a yaml file, I think we can do a check before merging the PR We should do in addition:

• include this in the contributing
• I can check if the current datasets in mirdata are in the list

[rabitt]

I agree we should do this. One concern for this requirement was that we don't have control over the speed of merging in the ismir/mir-datasets repository, and we didn't want this to block mirdata PRs. Maybe a solution is we could see if the maintainers are willing to let us be admins of ismir/mir-datasets as well?

[magdalenafuentes]

Missing entries now are: Beatport EDM key, cante100, Mridangam Stroke, Saraga. Small inconsistencies: GiantSteps tempo has audio available though in the list it says it doesn't.

[magdalenafuentes]

Maybe we can do this every release instead of every time we add a loader?

[magdalenafuentes]

I'm removing this one from the release and will do right after