Python numpy tutorial

1 Python

As a high-level dynamically typed multiparadigm programming language, python is attracts more and more attensions. Because of its strict indentation requirement, python code has some beauty in it.

Using Emacs and IPython, I finished the tutorial written by Justin Johnson . The following sections share the same architecture as in the tutorial. However, for the sake of learning, I input all the code and add my understanding. Because I learn Matlab beforehand, some differencec between Matlab and Python are mentioned. During the learning, I use python 3.6 instead of python 3.5 in the tutorial.

1.1 Basic data types

Like most languages, Python has a number of basic types including integers, floats, booleans, and strings.

1.1.1 numbers

Numbers include integer and float types and work as in other language.

x = 3
print(type(x)) # Prints "<class 'int'>"
print(x)       # Prints "3"
print(x+1)     # Addition; prints "4"
print(x-1)     # Subtraction; print "2"
print(x*2)     # Multiplication; prints "6"
print(x**2)    # Exponentiation; print "9"
x += 1
print(x)       # Print "4"
x *= 2
print(x)       # Print "8"
y = 2.5
print(type(y)) # Print "<class 'float'>"
print(y, y+1, y*2, y**2) # Prints "2.5 3.5 5.0 6.25"

Python has += , *= , -= , /= operators. However a++ and a-- are not supported in Python.

1.1.2 Booleans

Different from Matlab, Python use and , or to implement Booleans logic. In Matlab, which was implemented using && and || .

t = True
f = False
print(type(t)) # Prints "<class 'bool'>"
print(t and f) # logical AND; prints "False"
print(t or f)  # logical or; prints "True"
print(not t)   # logical NOT; print "False"
print(t != f)  # logical XOR; prints "True"

In matlab “not equal” was symbolized by ~= instead of != .

1.1.3 Strings

I think Python support more on strings.

hello = 'hello' # String lterals can use single quotes
world = "world" # double quotes; has the same effects as single ones
print(hello)    # Prints hello
print(len(hello)) #String length; prints "5"
hw = hello + ' ' + world # String concatenation
print(hw) # prints "hello world"
hw12 = '%s %s %d' % (hello,world,12)
print(hw12) # Prints "hello world 12"

In Python3.6, there is a handy sting called f-string. The last two lines code above can be written as:

print(f'{hello} {world} 12')

Quite a lot of methods are provided for strings:

s = "hello"
print(s.capitalize())   # Capitalize a string; prints "Hello"
print(s.upper()) # Print "HELLO"
print(s.rjust(7)) # right-justify a string, padding with spaces;
print( # center a string, padding with space;
print(s.replace('l','(ell)')) # replace all instances of one substring
#with another

More methods can be found in the document.

1.2 Containers

There are four built-in container types: lists, dictionaries, sets and tuples.

1.2.1 lists

A list is the Python equivalent of an array, but is resizeable and can contain element of different types.

xs = [3,1,2] # Create a list
print(xs,xs[2]) # prints "[3,1,2] 2"
print(xs[-1]) # print the last element; print "2"
xs[2] = 'foo' # different  types
print(xs) # print "[3,1,'foo']"
print(xs) # print "[3,1,'foo','bar']"
x = xs.pop()
print(x,xs) # prints "bar [3,1,'foo']"

As usual, you can find all the gory details about lists in the document.

If you are impatient to access list elements one at a time, slicing is an excellent choice to access more than one elements in a list.

nums = list(range(5)) # range is a built-in function that create a list of integers
print(nums) # print "[0,1,2,3,4]"
print(nums[2:4]) # get a slice from index 2 to 4 (exclusive);print "[2,3]"
print(nums[2:]) # prints "[2,3,]"
print(nums[:2]) # get a slice from the start to index 2(exclusive). prints "[0,1]"
print(nums[:]) # the whole list
print(nums[:-1]) # slice indices can be negative; prints "[0,1,2,3]"
nums[2:4] = [8,9]
print(nums) # Prints "[0,1,8,9,4]"

You can also loop over the elements of a list like this:

animals = ['cat', 'dog','monkey']
for animal in animals:

If you want access to the index of each element within the body of a loop, use the built-in enumerate function

animals = ['cat','dog','monkey']
for idx,animal in enumerate(animals):
    print('#%d: %s' % (idx + 1,animal))

Next, I want to show a point which does not exists in Matlab: list comprehensions. Consider the following code:

nums = [0,1,2,3,4]
squares = []
for x in nums:
    squares.append(x ** 2)

Using the list comprehensions, we implement the above with extremely short code.

nums = [0,1,2,3,4]
squares = [x ** 2 for x in nums]

You do not have to define square before you use it. list comprehensions can also contain conditions:

nums = [0,1,2,3,4]
even_squares = [x ** 2 for x in nums if x % 2 == 0]

1.2.2 Dictionaries

The values in a dictionary are pairs of (key,value) . You can access the value through the key just like what you do when you look up a dictionary.

Dictionary type use {} as the limiters.

d = {'cat':'cute','dot':'furry'} # Create a new dictionary with some data
print(d['cat']) # Get an entry from a dictionary using the key 'cat'
print('cat' in d) # Prints "True"; check if a dictionary has a given entry
d['fish'] = 'wet' # set a new entry in a dictionary
print(d['fish'])  # print "wet"
print(d.get('monkey','N/A')) # get an element with a default; print "N/A"
print(d.get('fish','N/A')) # Print "wet"
del d['fish'] # remove an element from a dictionary
print(d.get('fish','N/A')) # Print "N/A"

More methods for dictionary can be found in the document.

Like in list, you can also loop in a dictionary. The different is that in a dictionary, you loop using the keys.

d = {'person':2, 'cat':4,'spider':8}
for animal in d:
    legs = d[animal]
    print(f'a {animal} has  {legs} legs')

The output will be:

a person has  2 legs
a cat has  4 legs
a spider has  8 legs

Also, there is dictionary comprehensions, which are similar to list comprehensions, but allow you to easily construct dictionaries.

nums = [0,1,2,3,4]
even_num_to_square = {x: x**2 for x in nums if x % 2 == 0}

Notice that you can even print a dictionary directly using the print function.

1.2.3 Sets

A set is a collection of distinct elements. Notice the world distinct. A set has the same meaning what a mathematician mean. In particular, in a set there are not two identical elements. Set use {} as the limiter the same as dictionary.

animals = {'cat','dog'}
print('cat' in animals) # True
print('fish' in animals) # False
animals.add('fish') # add 'fish' in the set
print('fish' in animals) # True
print(len(animals)) # 3
animals.add('cat') # add 'cat' second time, nothing happen
print(len(animals)) # 3
animals.remove('cat') # remove 'cat' from the set
print(len(animals)) # 2

You can find more about set in the document.

You can also iterate over a set. Since sets are unordered, you cannot make assumptions about the order in which you visit the elements of the set.

animals = {'cat','dog','fish'}
for idx,animal in enumerate(animals):
    print(f'# {idx} : {animal}')

And the output will be:

# 0 : fish
# 1 : cat
# 2 : dog

Like lists and dictionarier, sets can be constructed by set comprehensions.

from math import sqrt
nums = {int(sqrt(x)) for x in range(20)}

The output will be:


1.2.4 Tuples

Tuple is the fourth container in Python. A tuple is an immutable ordered list of values. A tuple is in many ways similiar to a list. One of the most important differences is that tuples can be used as keys in dictionaries and as elements of sets, while lists cannot. Let’s take a look at a trivial example.

A tuple is enclosed in parentheses. It is not possible to assign to the individual items of a tuple. However, it is possible to create tuples which contain mutable objects, such as lists.

d = {(x,x+1):x for x in range(10)} # create a dictionary
t = (5,6) # create a tuple
print(type(t)) # <class 'tuple'>
print(d[t]) # 5
print(d[(1,2)]) # 1

You can find more about tuples in the document.

1.3 Functions

Python function are defined using the def keyword. For example:

def fib(n):
    """Print a fibonacci series up to n"""
    a,b = 0,1
    while a < n:
        print(a,end = ' ')
        a,b = b,a+b

The keyword def introducs a function definition which must be followed by the function name and the parenthesized list of formal parameters. The statements that form the body of the function start at the next line, and must be indented.

Sometimes, we want to define functions that have default arguments, like this:

def hello(name, loud=False):
    if loud:
        print(f'HELLO, {name.upper()}')
        print(f'Hello, {name}')

Notice that loud will be False by default. If you call the function hello with asignning any value to loud, it will be False. Of course, you can call hello using hello('Bob',True) . You can also call hello using hello('Bob',loud=True) .

You can find more about define a function in the document.

2 Numpy

As the core library of scientific computing in Python, Numpy make Python a replacement of Matlab even when there are quite a lot of differences between Numpy and Matlab. Because of open course, Numpy gets quite a lots of features to support scientific computation.

2.1 Arrays

A numpy array is a grid of values. All the values share the same type. A numpy array can be indexed by a tuple of nonnegative integers. The number of dimensions is the rank of the array. The shape of an array is a tuple of integers giving the size of the array along each dimension.

We can create a numpy arrays from nested Python lists, and access elements using square brackets.

import numpy as np
a = np.array([1,2,3])
a[0] = 5

b = np.array([[1,2,3],[4,5,6]])

As in Matlab, numpy also provides many functions to create special arrays:

import numpy as np

a = np.zeros((4,4))

b = np.ones((2,2))

c = np.full((2,2),9)

d = np.eye(2)

e = np.random.random((2,2))

Another two useful methods are arange and linspace

import numpy as np

a = np.arange(10) #exclude 10

b = np.linspace(1,4,6) #include 4

More methods can be found in the document.

2.2 Array indexing

You can access the element of an array one by one or more than one at a time. I like slicing very much. Because it is an effective way to access more than one elements at a time.

2.2.1 slicing indexing

import numpy as np

a = np.array([[1,2,3,4],
b = a[:2,1:3]
c = a[::2,0::2]

The output will be:

[[2 3]
 [6 7]]
[[ 1  3]
 [ 9 11]]

Notice that:

  1. In python, index starts from zero instead of one. (In Matlab, index starts from one)
  2. the slicing has the form start:stop:step , which mean start from the start and stop at stop (exclusive) with step as step

In Python, negative integer can also be used as index which provides flexibility for accessing the elements in an array.

One important thing about slicing is that the origin one and the sliced result are closely related. please take a look at this post.

Let’s follow the above example.

b[0,0] = 99

And the output will be:

[[99  3]
 [ 6  7]]

[[ 1 99  3  4]
 [ 5  6  7  8]
 [ 9 10 11 12]]

Notice that the 99 appears at b[0,0] and a[0,1] . I did not set a[0,1] explicitly, but it was changed.

When you index the numply arrays using slicing, the resulting array view will always be a subarray of the original way. If you change the subarray, the origin array will be changed too. This is pretty weird for engineer coming from Matlab.

2.2.2 integer array indexing

Another way of indexing array is integer array indexing. For example:

import numpy as np
a = np.array([[1,2],


Notice that integer array indexing does not have the effect of “view”. You can change the new array resulting from integer array indexing without worry about changing the origin array.

Another advantage of integer array indexing is that you can modify certain positions of an array at a time. For example:

import numpy as np

a = np.array([[1,2,3],


#Create an array of indices
b = np.array([0,2,0,1])
# Select one element from each row of the array
# using the indices in b


a[np.arange(4),b] += 10


Notice that only some positions of the array a are added 10. The positions are represented by np.arange(4),b . I think this is of high efficiency.

2.2.3 boolean array indexing

Boolean array indexing is another method of indexing an array. In matlab, there is boolean array indexing too. According to my experience, boolean array indexing is used to pick up certain elements of an array satisfying cerntain conditions.

import numpy as np
a = np.array([[1,2],
# bool_idx are positions where
# the elements are greater than 2
bool_idx = (a > 2)



The shape of bool_idx is the same as a . The value of bool_idx is True where a>2 satisfies. The last line tells us that we can finish the finding of positions and picking up the elements in one concise statement.

You can find more on indexing in the document.

2.3 Datatypes

Every numpy array is a grid of elements of the same type. Numpy provides a large set of numeric datatypes that you can use to construct arrays. Numpy tries to guess a datatype when you create an array, but functions that construct arrays usually also include an optional argument to explicitly specify the datatype. For example:

import numpy as np
x = np.array([1,2])

x = np.array([1.0,2.0])

x = np.array([1,2],dtype=np.int64)

x = np.random.random((1,2))

x = np.array([1+5j,2 + 4j])

The output is :


Notice that for a complex number, numpy use the complex128 to indicate the datetype.

2.4 Array math

It is always more efficient to do algebra in vector mode so that the computation is expected to be finished in a parallel way. Because everything in python is object, the operators can be overloaded.

import numpy as np

x = np.array([[1,2],[3,4]],dtype=np.float64)
y = np.array([[5,6],[7,8]],dtype=np.float64)


One point that needs our attention is the matrix multiplication. There are two types of multiplication between matrices: one is elementwise multiplication and the other one is matrix multiplication. In Matlab, we use * to implement the matrix multiplication and .* to implement the elementwise multiplication. In Python, things are different. * is use to implement elementwise multiplication and dot is used to compute matrix multiplication.

import numpy as np

x = np.array([[1,2],[3,4]])
y = np.array([[5,6],[7,8]])

v = np.array([9,10])
w = np.array([11,12])

# all the following 4 statements produce 219

# matrix multiplication


# AB != BA

Like in Matlab, we can sum up the matrix along one axis.

import numpy as np

x = np.array([ [1,2],[3,4] ])

# sum all the elements of x

# sum along the column
# sum along the row

Changing the shape of an array is also one frequent operation. The most used operation is transposing a matrix.

import numpy as np

x = np.array([[1,2],[3,4]])


If you want to reshape an array, shape is used.

import numpy as np

a = np.arange(12)

b = a.reshape((3,4))

The output will be:

[ 0  1  2  3  4  5  6  7  8  9 10 11]

[[ 0  1  2  3]
[ 4  5  6  7]
[ 8  9 10 11]]

Notice that np.arange(n) will exclude the n. and from shape (12,) to shape (3,4), Python write the (3,4) row first. However, in Matlab, it is column first.

2.5 Broadcasting

In Matlab, if you want to add two matrices, they must have the same dimension, or an error will be throwed out. However, in Python there is a mechanism called “Broadcasting” which allow you to work with arrays with different shapes.

Suppose we want to add a constant vector to each row of a matrix. We would like this:

import numpy as np

x = np.array([[1,2,3],

v = np.array([1,0,1])

#create an empty matrix with the same shape as x
y = np.empty_like(x)

for i in range(4):
    y[i,:] = x[i,:] + v


The above code implement what we want. However, from the eye of Python, it is not concise. You have to iterate over x . Python allows you broadcast the v to each row of x

import numpy as np

x = np.array([[1,2,3],
v = np.array([1,0,1])
y = x + v

It is so easy. It works even x has shape (4,3) and v has shape (3,) due to broadcasting.

Broadcasting two arrays together follows thses rules:

  1. if the arrays do not have the same shape, prepend the shape of the lower array with 1s until both shapes have the same length.
  2. The two arrays are said to be compatible in a dimension if they have the same size in the dimension, of if one of the arrays has size 1 in that dimension.
  3. The arrays can be broadcast together if they are compatible in all dimensions.
  4. After broadcasting, each array behaves as if it had shape equal to the elementwise maximum of shapes of the two input arrays.
  5. In any dimension where one array had size 1 and the other array had size greater than 1, the first array behaves as if it were copied along that dimension.

Please check this document and this document for more explainations. Functions that support broadcasting is called the universal function. You can find all the available universal function in this document.

3 summary

This post summrize some important aspects of Numpy. However, it is far to be like the numpy refernence. You should check the numpy reference frequently when you are not sure what you are using.


comments powered by Disqus