NumPy Arrays for Beginners: Create Arrays, dtype and shape

NumPy (Numerical Python) provides the ndarray: a grid of values that all have the same type, stored together in memory. That design lets NumPy run maths on whole arrays at C speed instead of looping in Python. pandas, scikit-learn and almost every scientific Python library are built on it.

Creating an array#

import numpy as np

prices = np.array([120, 250, 80, 310])
print(prices)
print(type(prices))
print(prices.dtype, prices.shape, prices.ndim, prices.size)

[120 250  80 310]
<class 'numpy.ndarray'>
int64 (4,) 1 4

• dtype is the type of every element (here 64-bit integers).
• shape is the size along each dimension; (4,) means one dimension with 4 items.
• ndim is the number of dimensions, and size the total number of elements.

Why not just use a list?#

Multiplying every price by 1.18 with a list needs a loop or comprehension; with an array, you write the maths directly:

import numpy as np

prices_list = [120, 250, 80, 310]
print([p * 1.18 for p in prices_list])

prices = np.array(prices_list)
print(prices * 1.18)

[141.6, 295.0, 94.39999999999999, 365.79999999999995]
[141.6 295.   94.4 365.8]

The array version is shorter and, on large data, many times faster, because the loop happens inside NumPy's compiled code. Note that prices_list * 2 would repeat the list, not double the numbers: lists and arrays behave differently with operators.

Data types#

All elements share one dtype. Mixing ints and floats gives floats; NumPy picks a type that can hold everything:

import numpy as np

print(np.array([1, 2, 3]).dtype)
print(np.array([1, 2.5, 3]).dtype)
print(np.array([True, False]).dtype)
print(np.array(["a", "bc"]).dtype)

ints = np.array([1.9, 2.1, -3.7]).astype(int)
print(ints)

int64
float64
bool
<U2
[ 1  2 -3]

astype(int) converts and truncates toward zero (1.9 becomes 1, −3.7 becomes −3); it does not round. <U2 means Unicode strings of up to 2 characters. For data analysis you mostly see int64, float64, bool and, in pandas, text and dates.

Ready-made arrays#

import numpy as np

print(np.zeros(3))
print(np.ones(4, dtype=int))
print(np.full(3, 7))
print(np.arange(0, 10, 2))       # like range: start, stop (excluded), step
print(np.linspace(0, 1, 5))      # 5 evenly spaced values, both ends included

[0. 0. 0.]
[1 1 1 1]
[7 7 7]
[0 2 4 6 8]
[0.   0.25 0.5  0.75 1.  ]

arange is for integer steps; linspace is for "n points between a and b", which is what you want for plotting smooth curves.

Two-dimensional arrays#

A 2D array is a table of rows and columns:

import numpy as np

scores = np.array([
    [72, 85, 90],
    [64, 78, 88],
    [95, 91, 87],
])
print(scores.shape)        # (rows, columns)
print(scores.ndim)
print(scores[1])           # the second row
print(scores[1, 2])        # row 1, column 2
print(np.eye(3))           # identity matrix

(3, 3)
2
[64 78 88]
88
[[1. 0. 0.]
 [0. 1. 0.]
 [0. 0. 1.]]

Rows must all have the same length. NumPy can have any number of dimensions (images are often 3D: height, width, colour channels), but data analysis mostly uses 1D and 2D.

Common mistakes#

Practice#

import numpy as np

celsius = np.array([21.5, 23.0, 19.8, 25.4, 22.1])
print(celsius.dtype, celsius.shape)
fahrenheit = celsius * 9 / 5 + 32
print(np.round(fahrenheit, 1))

float64 (5,)
[70.7 73.4 67.6 77.7 71.8]

Key takeaway#

Next: Vectorised Operations and Broadcasting.