numpy Numpy n-dimensional array: the ndarray


Example

The core data structure in numpy is the ndarray (short for n-dimensional array). ndarrays are

  • homogeneous (i.e. they contain items of the same data-type)
  • contain items of fixed sizes (given by a shape, a tuple of n positive integers that specify the sizes of each dimension)

One-dimensional array:

x = np.arange(15)
# array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14])
x.shape
# (15,)

Two-dimensional array:

x = np.asarray([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9], [10, 11, 12, 13, 14]])
x
# array([[ 0, 1, 2, 3, 4],
# [ 5, 6, 7, 8, 9],
# [10, 11, 12, 13, 14]])
x.shape
# (3, 5)

Three-dimensional:

np.arange(12).reshape([2,3,2])

To initialize an array without specifying its contents use:

x = np.empty([2, 2])
# array([[ 0., 0.],
# [ 0., 0.]])

Datatype guessing and automatic casting

The data-type is set to float by default

x = np.empty([2, 2])
# array([[ 0., 0.],
# [ 0., 0.]])

x.dtype
# dtype('float64')

If some data is provided, numpy will guess the data-type:

x = np.asarray([[1, 2], [3, 4]])
x.dtype
# dtype('int32')

Note that when doing assignments numpy will attempt to automatically cast values to suit the ndarray's datatype

x[1, 1] = 1.5 # assign a float value
x[1, 1]
# 1 
# value has been casted to int
x[1, 1] = 'z' # value cannot be casted, resulting in a ValueError

Array broadcasting

See also Broadcasting array operations.

x = np.asarray([[1, 2], [3, 4]])
# array([[1, 2],
         [3, 4]])
y = np.asarray([[5, 6]])
# array([[5, 6]])

In matrix terminology, we would have a 2x2 matrix and a 1x2 row vector. Still we're able to do a sum

# x + y
array([[ 6, 8],
       [ 8, 10]])

This is because the array y is "stretched" to:

array([[5, 6],
       [5, 6]])

to suit the shape of x.

Resources: