Avoiding np.diag 2.4.0 memory leak bug #30862 in unscented_transform.py

[devvaibhav455]

Hi Roger,

Thank you for creating this wonderful repository and for sharing your Kalman Filter knowledge through the book and tutorials. I actually learned Kalman Filters primarily through your material.

While working on a project, I ran into an issue when using the UKF implementation with NumPy 2.4.0, where np.diag causes a memory leak:
https://github.com/numpy/numpy/issues/30862

The issue originates from this line:

P = np.dot(y.T, np.dot(np.diag(Wc), y))

I replaced it with a broadcasting-based implementation that produces the same result but avoids constructing the diagonal matrix, uses less memory, less computations overall, and faster :

P = y.T @ (y * Wc[:, None])

This avoids the memory leak triggered by np.diag in NumPy 2.4.0.

I understand this is a NumPy-specific bug, but it took me a couple of days to trace and diagnose the issue. I hope this small change helps others avoid running into the same problem.

Thanks again for your excellent contribution to the community.