Numpy is a package used for scientific computation. It is fast thanks to its optimized C code core, easy to use and open source.

Getting started with Numpy is easy simply start by importing the package.

import numpy as np

Numpy offers us different data types but the most frequent data type is an ndarray. The n stands for number and d stands for dimensional.

# our first ndarray
one_dimensional = np.array([1,2,3])

three_dimensional = np.array([
    [
        [15.5, 20, 44],
        [34, 23, 78.2],
        [43, 64.3, 76]
    ],
    [
        [24, 24, 24],
        [80, 69, 32],
        [12, 15, 19]
    ]
])

An array can have 1 to nth dimension but usually they are named as followed:

• 1D = Vector
• 2D and above = Matrix

Since ndarrays can have different shapes and dimensions we can easily check what we are looking at.

# returns (3,) where 3 are the number of items [1,2,3]
one_dimensional.shape 

# returns (2, 3, 3) where 2 is the number of first level arrays
# 3 is the number of arrays within each array
# and the last 3 is the number of items in each array
three_dimensional.shape

# number of dimension of each ndarray
one_dimensional.ndim # returns 1
three_dimensional.ndim # returns 3

# number of items in each ndarray
one_dimensional.size # returns 3
three_dimensional.size # returns 18

Creating ndarrays

Thus far we looked at how ndarrays are shaped and how they look. Now we will learn how to make our own ndarrays.

# the simplest one
first = np.array([1,2,3])

# create an ndarray with a specific shape and fill it
# this renders an ndarray with 2 arrays inside and each array has 3 arrays
# and each of those inner arrays have 3 items filled with the number 1
# but displayed like a string thanks to the dtype argument we passed
second = np.ones((2,3,3), dtype=str)

array([[['1', '1', '1'],
        ['1', '1', '1'],
        ['1', '1', '1']],

       [['1', '1', '1'],
        ['1', '1', '1'],
        ['1', '1', '1']]], dtype='<U1')

# use a range to fill an ndarray
# here we start at 0 and go up to 20 in steps of 3
array_from_range = np.arange(0, 20, 3)

array([ 0,  3,  6,  9, 12, 15, 18])

# we can also randomly fill an ndarray with a given shape
# randint takes a minium = 1 and a maximum = 100 as well a shape (5,4)
array_with_random_ints = np.random.randint(1, 100, (5,4))

array([[0.90437567, 0.22028631, 0.72788259],
       [0.7913316 , 0.31966372, 0.58556257],
       [0.22827483, 0.88165549, 0.52483884],
       [0.09496781, 0.76239879, 0.31752114],
       [0.64415386, 0.85145683, 0.17401625],
       [0.82328818, 0.95793898, 0.86793474]])

A quick note on randomness. At the end, we use np.random to randomly fill our ndarray. It may look like true randomness but in fact those numbers come from a seed. They are pseudo random numbers which means that we can specify a seed number and numpy will stick to that seed rather than picking a different seed everytime which will make our randomness render the same numbers always.

np.random.seed(5)

Manipulating ndarray values

We can use mathematical operations on our ndarrays to do some fun things and numpy comes with built in methods that allow us to do some statistical operations easier as well.

one_dimensional = np.array([2,4,6])
placeholder_ones = np.ones(3) # array([1., 1., 1.])


one_dimensional + placeholder_ones
one_dimensional - placeholder_ones
one_dimensional * placeholder_ones
one_dimensional / placeholder_ones

Numpy uses broadcasting to make arithmetic operation fast.

The term broadcasting describes how numpy treats arrays with different shapes during arithmetic operations. Subject to certain constraints, the smaller array is “broadcast” across the larger array so that they have compatible shapes. Broadcasting provides a means of vectorizing array operations so that looping occurs in C instead of Python.

two_dimensional * three_dimensional # this fails

The reason the previous multiplication fails is because the shapes don’t match.

When operating on two arrays, NumPy compares their shapes element-wise. It starts with the trailing dimensions, and works its way forward. Two dimensions are compatible when

1. they are equal, or
2. one of them is 1

If these conditions are not met, a ValueError: operands could not be broadcast together exception is thrown, indicating that the arrays have incompatible shapes.

# numpy comes with math methods as well
np.sum(one_dimensional)
np.mean(one_dimensional)
np.max(one_dimensional)
np.min(one_dimensional)
np.std(one_dimensional) # standard variation
np.var(one_dimensional) # variance
np.sqrt(one_dimensional) # square root

And thats about it. We have covered some of the common operations in numpy. Of course, there is more and you can look at the numpy documentation for more information on how to do more advanced things as well.