mir_eval.key.validate(reference_key, estimated_key)¶ when the key or the mode are unknown

[PRamoneda]

Hello,

First of all, thank you very much for the library. It is fantastic!

At mir_eval.key:

When you don't know the key or the mode (in the reference or in the estimated) you should be able to compute the metric.

see the examples :

x minor
C

Well, x minor can be computed but I think that if your algorithm is performing the mode, the score should be more than 0.

Thank you so much!

[PRamoneda]

On the other hand, why do you only allow to evaluate (key, modes) that have major or minor modes? I believe that there are key datasets with much more extended vocabularies.

Thank you so much!!

[craffel]

The key module was created to match the conventions used by MIREX. If there are other benchmarks that use a richer set of annotations, we welcome PRs to implement that kind of evaluation.

[PRamoneda]

Hey @craffel!

As is seen in this Angel Faraldo phd Thesis figure there are some datasets with extended vocabularies. (I think that all the vocabularies can be reduced to diatonic modes).

There are more datasets with diatonic modes annotations. As:
https://zenodo.org/record/1290737#.X5BjPC3FRhE
https://zenodo.org/record/1095691#.X5BjUi3FRhE
https://zenodo.org/record/1101082#.X5BjaC3FRhE