The datatype of a NumPy array is called ndarray. If one looks into what an ndarray is actually made of, one can see that it consists of the following:
- one dimensional contiguous block of memory: raw data
- indexing scheme: how to locate an element
- data type descriptor: how to interpret an element
There are many possible ways of arranging the elements of a N-dimensional array in a 1-dimensional block (i.e. memory). NumPy uses striding where a N-dimensional index (n[0], n[1], ..., n[-1]) corresponds to the offset from the beginning of a 1-dimensional block.
If n[k] is the index in dimension k for an element and s[k] is the stride in that dimension, then the offset for an element is:
offset = sum(s[k] * n[k] for k in range(N))
When one assigns a variable to a slice
b = a[1:8:2, 3:12:3]the variable b has the same raw data as a, but only different
strides. Thus, changing the contents of b will also change the contents of
a, i.e. b is view to a as discussed earlier.
In addition to slicing NumPy allows indexing also with integer arrays or Boolean masks:
a = np.arange(0.0, 1.0, 0.1)
ind = np.array([1, 1, 0, 4])
b = a[ind] # b = array([0.1, 0.1, 0. , 0.4])
m = a > 0.5
b = a[m] # b = array([0.6, 0.7, 0.8, 0.9])In these cases b cannot be created just by modifying the strides, so b
will hold a copy of the data in a. Now, modifications of b are not
affecting a.
a = np.array(...)
: a.flags
: various information about memory layout
`a.strides`
: bytes to step in each dimension when traversing
`a.itemsize`
: size of one array element in bytes
`a.data`
: Python buffer object pointing to start of arrays data
`a.__array_interface__`
: Python internal interface