Fabian-Robert Stöter
Fabian-Robert Stöter
install the test evironment `pip install .[tests]` and then run `py.test tests/test_regression.py -vs`
@K3nn3th2 hi and welcome to the messy world of music separation evaluation ;-) You can use museval in exactly the same way as you would use `mir_eval`, in fact if...
@K3nn3th2 can you send me a code snippet how you run the evaluation? (what are the tensor shapes)?
@K3nn3th2 what are the shape of `ref_nps` and `estim_nps`? it looks like you are comparing a single reference source with a single estimate (which is not the purpose of bsseval...
@K3nn3th2 can this be closed?
yes, this absolutely makes sense. Our plan was to add this to mir_eval (https://github.com/craffel/mir_eval/pull/272). But we also considered a separate repository. You have some opinion on this?
we got more feedback for `museval` at this years ICASSP. Basically people addressed the need for a separate bsseval package with fewer requirements and focus on performance. Therefore I would...
> One suggestion is to make any torch/tensorflow dependency optional as those packages can be quite big (won’t fit on an AWS lambda). sure. Lets make separate packages `bsseval` (==numpy),...
> I'm also in favor of including SI-SDR / SI-SNR (same thing, different name). I guess we could also consider including the redefined SI-SIR and SI-SAR, although I'm still not...
done -> [bsseval](https://github.com/sigsep/bsseval) lets continue discussion over there