I have two numpy a,b the shape of them are (100,2048), and I used sys.getsizeof(a) = 112 and same with array b.
I have question, when I use c = np.concatenate((a,b),axis=0), the shape of c is (200,2048), but the sys.getsizeof(c) = 1638512
Why?
getsizeof has limited value. It can be way off for lists. For arrays it's better, but you have to understand how arrays are stored.
In [447]: import sys
In [448]: a = np.arange(100)
In [449]: sys.getsizeof(a)
Out[449]: 896
But look at the size of a view:
In [450]: b = a.reshape(10,10)
In [451]: sys.getsizeof(b)
Out[451]: 112
This shows the size of the array object, but not the size of the shared databuffer. b doesn't have its own databuffer.
In [453]: a.size
Out[453]: 100
In [454]: b.size
Out[454]: 100
So my guess is that your a and b are views of some other arrays. But the concatenate produces a new array with its own databuffer. It can't be a view of the other two. So its getsizeof reflects that.
In [457]: c = np.concatenate((a,b.ravel()))
In [459]: c.shape
Out[459]: (200,)
In [460]: c.size
Out[460]: 200
In [461]: sys.getsizeof(c)
Out[461]: 1696
The databuffer for a is 100*8 bytes, so the 'overhead' is 96. For c, 200*8, again with a 96 'overhead'.
It do not reproduce your example:
import numpy as np
import sys
a = np.random.rand(100, 2048)
b = np.random.rand(100, 2048)
print(sys.getsizeof(a), sys.getsizeof(b))
# 1638512 1638512
c = np.concatenate((a,b), axis=0)
print(sys.getsizeof(c))
# 3276912 which is about 1638512 + 1638512
getsizeof stackoverflow.com/a/17574104/8069403 , maybe it could help understand what happens