Brian McFee

Confirmed that the revised linalgwarning catch and fallback fixes the issue in this case, at least on my machine. With any luck, this will resolve #376 and we won't have...

Well, bsseval is now about a million times slower, I'm guessing because it's properly detecting ill-conditioning now where it wasn't before. It's kind of a miracle that it worked previously,...

Adding the :bug: label here because I'm becoming convinced that the present implementation shouldn't be trusted. To summarize where we're at: - `np.linalg.solve` was letting ill-conditioned data through silently instead...

As an aside, I do think we could also significantly accelerate everything here by rewriting it with proper vectorization instead of nested for loops. However, I don't have the fortitude...

Nudging @craffel - any chance we could get the org for this setup this week? We're looking to do a bit of in-house code maintenance over here, and AFAICT this...

> I had some similar issues with fft precisions when torchaudio did changes some months ago Yeah, it shouldn't be surprising. TBH I've never done a full deep dive on...

What if I was to create a new org and all you have to do is transfer the repo? (Assuming you're still okay with the entire idea, that is.)

All the good names were already taken :\ but I just made https://github.com/mir-evaluation and sent you an invite.

Circling back on this one today, and it's still pretty ugly. I've had my head down in the code for a little bit now, and am increasingly convinced that this...

Thanks! I think the only other thing that will need doing from your side before a new release is updating the credentials on pypi to allow the publish action from...