Brian McFee

(warning: nitpicking to follow. sorry!) I'm going through the chord metrics implementation, just to see if there are any other likely oddities. Here are some things I've noticed that seem...

I've also noticed this while qualitatively testing chord models, and I agree that it's confusing to listen to. It's been a while since I looked at this code, but I...

> I think it makes sense to interpolate over the minimum of the interval length or some pre-defined short interval. How about interpolating over, say, two cycles at the frequency...

> This seems fine unless there's an interval which is shorter than two cycles of the frequency. Then again if the interval is that short the user should expect it...

IIRC the reason for this was to keep the translation from the matlab implementation as direct as possible. Maybe @dawenl can comment?

From reading the comments, I think this is behaving to spec / consistent with the mirex implementation. I also think it's weird, but 🤷

To follow up on this a bit, yes, it's intended to be asymmetric, and the docstring is unambiguous about the behavior: https://github.com/craffel/mir_eval/blob/576aad4e0b5931e7c697c078a1153c99b885c64f/mir_eval/key.py#L99-L115 I suppose this was originally intended to break...

> So apparently the "official" behavior changed from only counting ascending ones to also counting descending ones in 2017. It would be really useful if both approaches were implemented, maybe...

> we could add an arg that specifies whether to use a LIL matrix or an ndarray? I thought about that. It's doable, but kind of ugly from an API...

Bumping this one, since it seems like everyone disappeared after summer?