Intro to NumPy

By: Carlo Sarli

Posted: February 3, 2021 in (Python) (NumPy)

Tech Used:

Python3.x
NumPy

Reading Time: 3 minutes

What is NumPy

NumPy (Numerical Python) or np is a python library for the creation and manipulation of multidimensional arrays, matrices and the statistical/mathematical operation one might want to perform on them.

It is an essential package for python and data analysis that anybody is bound to encounter at one point or another in their journey.

Why NumPy

Numpy’s core library are written in low level languages such as C/C++/Fortran allowing for faster than regular python operation on arrays and matrices. It also generally makes np’s data structures more memory efficient than their vanilla python counterparts.

It is highly compatible with most other data analysis libraries such as Pandas, as it is a de facto standard library in any project of this kind.

Multidimensional Arrays

The most important object in the np library is probably the ndarry object. This is a multidimensional array object that acts as a fast container for large datasets.

Getting started with NumPy

In this example we install np using the pip installer. This could also be done via Conda and other python package managers so chose whichever suits your dev env best.

# Install via Pip in terminal
pip install numpy

# import the library

import numpy as np

# The library will be accessible as np

Creating arrays

# NdArrays (Multidimensional arrays)
# Random data 

data = np.random.randn(2, 3)    
# creates a 2 dimensional array with 3 random elements per dimensions

# find array dimensions
print (data.shape)
# output (2, 3)

# just number of dimensions
print(data.ndim)
# output: 2

# find type of data 
print(data.dtype)
# output: float64

# make array from list 
list1 = [1, 4, 3, 8]

arr1 = np.array(list1)
print(arr1)
# output:  [1 4 3 8]

# nested lists are converted to multidimensional arrays
# note they need to be same length
list2 = [[1, 4, 3, 8], [48, 9, 39, 3]]
arr2 = np.array(list2)
print(arr2)
# output:  [[ 1  4  3  8][48  9 39  3]] 

print(arr2.ndim)
# output: 2

# create array and fill with 0s
zeros = np.zeros(5)
print (zeros)
# [0. 0. 0. 0. 0.]

# create array and fill with 1s
ones = np.ones(5)
print (ones)
# [1. 1. 1. 1. 1.]

# add dimensions 

zeros2D = np.zeros((5, 2))
print (zeros2D) 
# [[0. 0.]
#  [0. 0.]
#  [0. 0.]
#  [0. 0.]
#  [0. 0.]]

# create array of sequential numbers in range
seq = np.arange(10)
print (seq)
# [0 1 2 3 4 5 6 7 8 9]


# specify data type
arrWType = np.array([1, 2, 3], dtype=np.float64)
print (arrWType.dtype)
# float64

# cast array 
arrWDiffType = arrWType.astype(np.int32)
print (arrWDiffType.dtype)
# int32

NumPy Arithmetic operations

# arr1 [1 4 3 8]
sum = arr1 + arr1 
print (sum)
#[ 2  8  6 16]

times = arr1 * arr1 
print (times)
#[ 1 16  9 64]

minus = arr1 - arr1 
print (minus)
#[0 0 0 0]

div = arr1 / arr1
print (div)
#[1. 1. 1. 1.]

ex = np.exp(arr1)
print (ex)
# [2.71828183e+00 5.45981500e+01 2.00855369e+01 2.98095799e+03]

sqrt = np.sqrt(ex)
print (sqrt)
# [ 1.64872127  7.3890561   4.48168907 54.59815003]

Indexing, Slicing and copying

print (arr1[2])
# 3

print (arr1[2:4])
# [3 8]

# avoid aliasing by using .copy()
sli = arr1[0:2].copy()
print (sli)
# [1 4]

sli[0] = 12
print (sli)
# [12  4]

print (arr1)
# [1 4 3 8]

# no aliasing

Transposing and reshaping

tr = np.arange(10).reshape(2,5)
# create array of 10 el 0 to 9 and reshape into a 2d array of 5 el
print (tr)
# [[0 1 2 3 4]
#  [5 6 7 8 9]]


# invert axis
print (tr.T)
# [[0 5]
#  [1 6]
#  [2 7]
#  [3 8]
#  [4 9]]

NumPy Docs

Related Tutorials