implementing broadcasting and np.mean on non-contiguous arrays in NumPy 1.25
I'm performance testing and I'm having trouble computing the mean of a non-contiguous NumPy array due to some unexpected broadcasting behavior. I created a 2D array from a slice of another array, and when I try to compute the mean across an axis, the result seems off. Here's a minimal example: ```python import numpy as np # Creating a 2D array original_array = np.arange(12).reshape(3, 4) # Creating a non-contiguous view by slicing sliced_array = original_array[:, 1:3] # Attempting to compute the mean across axis 0 mean_result = np.mean(sliced_array, axis=0) print(mean_result) # Expected behavior? ``` The output I get is `array([4., 5.])`, which seems correct. But when I try to use the result to perform calculations on the original array, I'm getting an unexpected result: ```python resulting_array = original_array + mean_result ``` I expected the operation to work seamlessly, but it raises an behavior: ``` ValueError: operands could not be broadcast together with shapes (3,4) (2,) ``` It appears that the mean result is not matching the shape of the original array for the addition operation. I've checked the shapes using `sliced_array.shape` which returns `(3, 2)` and `mean_result.shape` which returns `(2,)`, but I assumed that NumPy would handle the broadcasting automatically. What am I doing wrong here? Is there a way to reshape the mean result to make the addition compatible, or is this a limitation of how non-contiguous arrays interact with broadcasting? Any insights would be appreciated! For context: I'm using Python on Windows. I'd really appreciate any guidance on this. What would be the recommended way to handle this? What am I doing wrong?