为什么 numpy 需要元组列表而不是列表列表？

Question

I'm creating table with information about images:

directory = 'images'
dtype = [
    ('file', np.str_), 
    ('ext', np.str_), 
    ('size', np.int32), 
    ('sizeKB', np.float), 
]
files = []

for file in os.listdir(directory):
    height = 0
    width = 0 
    filepath = os.path.join(directory, file)

    name, ext = os.path.splitext(filepath)
    name = name[7:]
    size = os.path.getsize(filepath)

    filedesc = (name, ext, size, size/1024)
    files.append(filedesc)

files_np = np.array(files, dtype=dtype)

When instead of list of tuples I create list of lists

filedesc = (name, ext, size, size/1024)

I get an error

ValueError: invalid literal for int() with base 10: 'somefilename'

Why is that?

Answer 1

Here's an example, with several levels of nesting of [] and ().

In [32]: dt = np.dtype([('x',int,3),('y','U10')])                               
In [33]: data = np.zeros((3,2),dt)                                              
In [34]: data                                                                   
Out[34]: 
array([[([0, 0, 0], ''), ([0, 0, 0], '')],
       [([0, 0, 0], ''), ([0, 0, 0], '')],
       [([0, 0, 0], ''), ([0, 0, 0], '')]],
      dtype=[('x', '<i8', (3,)), ('y', '<U10')])

In [36]: data['x']=np.arange(18).reshape(3,2,3)                                 
In [37]: data['y']='foobar'                                                     
In [38]: data                                                                   
Out[38]: 
array([[([ 0,  1,  2], 'foobar'), ([ 3,  4,  5], 'foobar')],
       [([ 6,  7,  8], 'foobar'), ([ 9, 10, 11], 'foobar')],
       [([12, 13, 14], 'foobar'), ([15, 16, 17], 'foobar')]],
      dtype=[('x', '<i8', (3,)), ('y', '<U10')])
In [39]: print(data)                                                            
[[([ 0,  1,  2], 'foobar') ([ 3,  4,  5], 'foobar')]
 [([ 6,  7,  8], 'foobar') ([ 9, 10, 11], 'foobar')]
 [([12, 13, 14], 'foobar') ([15, 16, 17], 'foobar')]]

And another with a compound dtype within a compound dtype:

In [41]: np.zeros(2,dt)                                                         
Out[41]: 
array([([0, 0, 0], (0, 0.)), ([0, 0, 0], (0, 0.))],
      dtype=[('x', '<i8', (3,)), ('y', [('f0', '<i8'), ('f1', '<f8')])])

At each level, the use of () versus [] conveys useful information.