在NumPy 1.14中如何将结构化数组切片转换为常规NumPy数组

Got it, let's fix this issue specifically for NumPy 1.14. The problem with your original code is that arr2[0] gives you a 0-dimensional structured scalar, and in NumPy 1.14, you can't directly use view() on a 0d scalar to reshape it into a 2-element float32 array—hence that ValueError.

Here are two reliable solutions that work perfectly with NumPy 1.14:

Solution 1: Use view() on the entire sliced array (not the single element) #

Instead of accessing the 0th element first, apply view() directly to arr2 (which is a 1-dimensional structured array). This works because the array has an explicit axis that NumPy can map to the new dtype structure:

import numpy as np

arr = np.array([(105.0, 34.0, 145.0, 217.0)], dtype=[('a', 'f4'), ('b', 'f4'), ('c', 'f4'), ('d', 'f4')])
arr2 = arr[['a', 'b']]
out = arr2.view((np.float32, 2))
# Result: array([[105.,  34.]], dtype=float32)

Solution 2: Stack field arrays explicitly (more readable) #

If you prefer a more intuitive approach that avoids relying on memory layout details, you can stack the individual field arrays along the second axis. This method is also fully compatible with NumPy 1.14:

import numpy as np

arr = np.array([(105.0, 34.0, 145.0, 217.0)], dtype=[('a', 'f4'), ('b', 'f4'), ('c', 'f4'), ('d', 'f4')])
arr2 = arr[['a', 'b']]
out = np.stack([arr2['a'], arr2['b']], axis=1)
# Result: array([[105.,  34.]], dtype=float32)

Both approaches will give you the regular 2D float32 array you need without triggering that error. The first method is more efficient (it’s just a memory view with no data copying), while the second is easier to read if you want to avoid low-level dtype mapping logic.

内容的提问来源于stack exchange，提问作者Alex