Support for numpy >= 2.0

[charleslparker]

I've done some casual testing and based on the results I'm assuming support for numpy 2.0 and above isn't present yet. Are there plans to support it in the near future?

[ryan-wolbeck]

@charleslparker yes it will be something we support in the future. If you want to accelerate that ask I'd be happy to review a PR.

[charleslparker]

Thanks, @ryan-wolbeck . I'm a bit swamped right now but if I have a free moment I'll take a look. Thanks for excellent library!

[jerome-f]

+1 on this Score -> grad np.linalg.solve fails because 2.0

Changed in version 2.0: The b array is only treated as a shape (M,) column vector if it is exactly 1-dimensional. In all other instances it is treated as a stack of (M, K) matrices. Previously b would be treated as a stack of (M,) vectors if b.ndim was equal to a.ndim - 1.

[sharedw]

@jerome-f did you ever find a workaround for this?

[jerome-f]

@sharedw not really I am working with binary classification so the Fisher Information matrix is a scalar and calculating the natural gradient reduces to grad/metric. This should hold for other scalar FIM (single parameter distributions) but I have not tested them. Since the api supports multi-parameter distribution this is not a functional fix for the api.