np.broadcast_to optimization guide as expected with non-contiguous arrays in NumPy 1.24.2
I'm trying to figure out I'm trying to implement I'm working on a project and hit a roadblock. I'm having trouble using `np.broadcast_to` on a non-contiguous array. I have a 2D array that I want to broadcast to a larger shape, but it seems like the behavior is inconsistent when the original array is not contiguous in memory. For instance, I created a 2D array and then sliced it, making it non-contiguous: ```python import numpy as np # Creating a contiguous 2D array arr = np.array([[1, 2, 3], [4, 5, 6]]) # Slicing to create a non-contiguous array non_contiguous_arr = arr[:, 1:3] print(non_contiguous_arr.flags['C_CONTIGUOUS']) # This prints False # Now I want to broadcast this non-contiguous array to a new shape broadcasted_arr = np.broadcast_to(non_contiguous_arr, (3, 2, 2)) print(broadcasted_arr) ``` When I run this code, I get the behavior `ValueError: shape mismatch: objects want to be broadcast to a single shape`. I expected it to handle the broadcasting smoothly as per the documentation. I have also checked the shapes of the arrays involved, and the dimensions seem appropriate for broadcasting. I tried reshaping the non-contiguous array using `np.reshape` before broadcasting it, but that didn't help either: ```python reshaped_arr = non_contiguous_arr.reshape(1, 2) # Attempting to broadcast again broadcasted_arr = np.broadcast_to(reshaped_arr, (3, 2, 2)) ``` This still results in the same behavior. I have confirmed that I’m using NumPy 1.24.2, and I’ve looked through the release notes for any changes regarding `np.broadcast_to`. Am I missing something fundamental about how `np.broadcast_to` handles non-contiguous arrays? Any tips or workarounds would be appreciated! I'm working on a service that needs to handle this. Thanks in advance! I'm working with Python in a Docker container on Windows 11. What would be the recommended way to handle this? My development environment is Debian. What are your experiences with this?