python 列表转化成数组_将列表转换为NumPy数组

I have a list that consits of 2000 rows and 88200 columns:

testlist = list(split_audio_to_parts(audio, self.sample_rate, self.audio_index))

debugging output of testlist gives

[array([-0.00683594, -0.00689697, -0.00708008, ..., 0. ,

0. , 0. ]), array([-0.01287842, -0.01269531, -0.01257324, ..., 0. ,

0. , 0. ]), array([0.02288818, 0.01940918, 0.01409912, ..., 0. , 0. ,

0. ]), array([0.00772095, 0.00671387, 0.00695801, ..., 0. , 0. ,

0. ]),

and so on.

split_audio_to_parts is a function:

def split_audio_to_parts(x, sample_rate, audio_index):

for i, row in audio_index.iterrows():

x_part = x[int(row['start_samples']):int(row['end_samples'])]

yield x_part

When I try to convert it to numpy array using samples = np.array(testlist) or samples = np.asarray(testlist), it gives me array of shape (2000,), although debugging shows that testlist consits of 2000 entries with 88200 positions. Why so? I'm using 64bit numpy and 64bit Python 3.6.5.

解决方案

The problem is testlist is a list of different size arrays. For example checkout this code:

>>>import numpy as np

>>>import random

>>>random.seed(3240324324)

>>> y=[np.array(list(range(random.randint(1,3)))) for _ in range(3)]

>>> y

[array([0, 1, 2]), array([0, 1, 2]), array([0])]

>>> np.array(y)

array([array([0, 1, 2]), array([0, 1, 2]), array([0])], dtype=object)

>>> np.array(y).shape

(3,)

and the array would be of object type instead of float. the only way for this to work is having same sized arrays.

If you really need to stuff these rows somehow into an array you can pad with zeros, for example:

>>> size = y[max(enumerate(y),key=lambda k:k[1].shape)[0]].shape[0]

>>> z=[np.append(x,np.zeros(size-x.shape[0])) for x in y]

>>> z

[array([ 0., 1., 2.]), array([0, 1, 2]), array([0, 0, 0])]

>>>np.array(z).shape

(3, 3)

but you would have to decide how you do this padding.