DACO - Practical Lecture 1 - Introduction to Python for Scientific Computing

NOTE: pdf slides for the first part of this lecture can be found here

This year we have decided to move from Matlab to Python for the practical sessions. Some of you maybe will not have worked with this programming language. This first lecture is intended to guide you through your first steps in this programming language, and make you aware of the (super-rich) Python ecosystem for scientific computing.

I really hope that by the end of this course you will be a Python fan, and consider abandoning Matlab once and forever! This is an overview of what you will be learning today:

  1. Motivation and Goals. What is Python?

  2. Python Installation. Accompanying Tools

    1. Anaconda Python Distribution
    2. Executing Python code
  3. Introduction to the Python Programming Language

    1. Fundamental Python data types
    2. Everything is an object
    3. Python Modules
    4. Python Basic Operators
    5. Sequences and Object Containers
    6. Python typing
    7. Flow Control
    8. Python Functions
    9. Classes and Object-Oriented Programming
  4. Complementary Python Scientific Computing Tools: Numpy

    1. Introduction to NumPy Arrays
    2. Slicing NumPy Arrays
    3. Indexing NumPy Arrays
    4. Other numpy arrays manipulation techniques
  5. Complementary Python Scientific Computing Tools: Matplotlib

    1. Basic Matplotlib usage
    2. Plot Customization
  6. Homework 😱

  7. Sources and References

So… let’s move on.

1.- Motivation and Goals. What is Python?

First thing, Python is free. Second, it is simple. Third, it is increasingly becoming the tool of choice for data science projects. Fourth, it’s multi-platform, it can run in Windows, Linux, Mac, your mobile phone… And last, there is a huge community of contributors to lots of open-source projects that complement it. This manifests in the form of a large ecosystem of scientific computing tools that grow along with the number of users.

However, to add all this to your tool-belt, the first step is to familiarize yourself with the Python language itself. Today we will quickly review the main notions to get started on it.

But first, let us install Python!

2. Python Installation. Accompanying tools

2.1 Anaconda Python Distribution

In this course we will use the Python language programming. However, as mentioned above, one of the main strengths of Python is the large amount of available scientific libraries. This means that, together with the core Python language, we’ll be using several Python packages to perform scientific computations. But instead of installing all these packages manually one at a time, we will be using a Python distribution. A distribution consists of the core Python package and several hundred modules, and available through a single download and installation. In this course, we will use the Anaconda Python distribution. Apart of Python itself and several hundred libraries, Anaconda also includes two very useful development environments: Jupyter NB and Spyder.

Spyder is a Matlab-like IDE (Integrated Development Environment) that we will be using for complex/long programming assignments. On the other hand, Jupyter NB is a functionality that allows you to run sequentially code in your browser. I will be giving you instructions in class on how to install and use both tools.

The Anaconda Python distribution can be found here. Please install Python 3. For windows users, if you are finding troubles, let me know asap.

2.2 Executing Python code

Python has two different modes: interactive and standard. The interactive mode is meant for experimenting your code one line or one expression at a time. The standard mode is useful for running programs from start to finish. You will probably find yourself alternating between both modes.

2.2.1 Python Interactive Mode

For a first example on interactive mode, open a command console, type python, and you will start using Python. Experiment a bit.

A more convenient way of building code interactively is through Jupyter iPython notebooks. This also allows us to mix text, code, and maths, which is really cool. To start an iPython notebook, open a console, type jupyter notebook, and navigate in your browser to http://localhost:8888/. In windows, you may need a log-in password, provided automatically after executing jupyter. Here you can find a series of useful tricks and shortcuts.

2.2.2 Python Standard Mode

In this case, we write a text file with Python instructions, and run it. hello_world.py.

Again, there is a more convenient way for building code in standard mode, which is using an IDE. This programming tool will allow you to more easily debug, inspect variables, and quickly modify your code. They typically also include an embedded interactive system. An example of an IDE is Spyder, already contained in the Anaconda distribution.

For the contents of this lecture, starting from Section 3, a notebook version of this lecture is available in a static view here. Please click the download button in the right top corner, download and open it!

3. Introduction to the Python Programming Language

3.1 - Fundamental Python data types

Python contains several typical built-in data types as part of the core language. The same as in e.g Matlab (and different from e.g. C) you do not need to explicitly declare the type of a variable. Python determines data type of variables by how they are used.

For instance, to create an integer (int) variable you simply type:

# An integer variable a
a = 5
print(type(a))

Other basic/common python types are for instance float, string, or boolean, exemplified below:

# A float variable f
f = 5.0
# A boolean
b = True
# A string
c = 'bom dia'

3.2 - Everything is an object

It is important that you start thinking of the above examples a,f,b,c as objects. Each object in Python has three characteristics: object type, object value, and object identity. Object type tells Python what kind of an object it’s dealing with. A type could be a number, or a string, or a list, or something else. Object value is the data value is contained by the object. This could be a specific number, for example. Finally, you can think of object identity as an identity number for the object. Each distinct object in the computer’s memory will have its own identity number.

Most Python objects have either data or functions or both associated with them. These are known as attributes. The name of the attribute follows the name of the object, and they are separated by a dot in between them. The two types of attributes are called either data attributes or methods. A data attribute is a value that is attached to a specific object. In contrast, a method is a function that is attached to an object. And typically a method performs some function or some operation on that object. Object type always determines the kind of operations that it supports. In other words, depending on the type of the object, different methods may be available to you as a programmer. Finally, an instance is one occurrence of an object. For example, you could have two strings. They may have different values stored in them, but they nevertheless support the same set of methods.

3.3 - Python Modules

Python contains builtin functions, such as print that can be used by all Python programs. However, while the Python library consists of all core elements, such as data types and built-in functions, the bulk of the Python library consists of modules. In order for you to be able to make use of modules in your own code, you first need to import those modules using the import statement.

Example of modules, attributes, methods, in numpy.

The numpy example above made available some specific data types and methods for us. This is an example of a Python module. In general, Python modules are libraries of code that you import and use through import statements. Let’s go through a simple example. Import the math module. This module gives you access to the pi constant. Print its value. This module also gives you access to several mathematical operations. Find out (print) the value of the sine of pi.

If you only need, e.g., the value of pi from the entire math module, you can import it selectively:

from math import pi
print(pi)

3.4 - Python Basic Operators:

Operators are symbols that allow you to use logic and arithmetic in your computations. Python has several operators, the most prominent ones being arithmetic, comparison, and logical.

3.4.1 - Arithmetic operators:

They will take two variables and perform simple mathematical operations on them. They are addition +, subtraction -, multiplication *, division /, modulus %, floor division //, and exponentiation **

print(3+2, 3-2, 3*2, 3/2, 3%2, 3//2, 3**2)

3.4.2 - Comparison operators:

They observe two variables and return a boolean value. They are the usual greater than (>,>=), equal (=), different (!=), and lower than (<,<=) mathematical operators.

print(3>2, 3<2, 3==2, 3!=2, 3>=2, 3<=2)

3.4.3 - Logical operators:

These operators will interpret their input as boolean values, and return a boolean value depending on the truth value of both inputs. They are and, or, not.

print(True and True, False or False, not True)

3.4.5 - Other operators

There are other operators in Python, such as identity (is, is not) or membership (in, not in). I am 100% sure you will be able to guess what is their effect on variables/containers!

Note this subtle difference between == and is:

== tests whether objects have the same value, whereas is tests whether objects have the same identity. Try it with a=[1,2] and b=[1,2].

3.5 - Sequences and Object Containers

3.5.1 - Python Sequences

A sequence is an ordered collection of objects. In Python, you can find three types of sequences: Lists, Tuples, and Range Objects.

Let us focus on the first two of them. Lists and tuples can be accessed by indexing and can be sliced in several ways:

# A tuple
t = (0,True)
# A list l
l = [0,True]

Note that there is a relevant difference between tuples and lists: tuples are immutable, while lists are not. This means that you won’t be able to modify the content stored at t. We will explain this in a second. Note also that both lists and tuples allow you to mix different data types.

Note: Mutable and immutable objects

The value of some objects can change in the course of program execution. Objects whose value can change are said to be mutable objects, whereas objects whose value is unchangeable after they’ve been created are called immutable.

Continuing with sequences, in order to access the elements that a tuple/list holds, you use square brackets, not parenthesis:

# the first element of the tuple t
t_first = t[0]

Also note that, the same as e.g. C (and different from e.g. Matlab), in Python indexing starts at 0. Be careful with this, because it is a common source of confusion in the beginning.

Sequences can be accessed by indexing, which supports negatives indexes:

t = (0,1,'hola',3,4.0)
l = [0,1,'hola',3,4.0]
print(t[0])
print(t[-1]) # note the behavior of negative indexes

Sequences also support a very handy operation called slicing, that enables access to multiple objects at the same time:

print(t[0:2]) # slicing first two elements, third is excluded
print(l[2:5]) # slicing last three elements
print(l[2:]) # empty spot after : means up to length-of-list index
print(l[:2]) # empty spot before : means from first index
print(l[0:5:2]) # every element, but use a step size of 2

Tuples and lists, as every python object, have several methods that you can use with them. However, since lists are mutable, they have methods that can modify their content:

a = [4,3,2,1]
a.append(5)
print(a)
a.sort()
print(a)

Note that list methods are in-place methods, they modify the original object that called them and return nothing. For a complete list of these, see here.

There are other generic Python functions that work with sequences and provide useful operations:

a = [4,3,2,1]
b = sorted(a)
n = len(b)
s = sum(a)
print(b, n, s)

3.5.2 - Object Containers: Dictionaries

Sequences are a particularly simple example of Object Containers. More generally, Python offers several more advanced object containers. A really useful one are dictionaries.

Dictionaries are unordered sequences that associate key objects to value objects. This means that a dictionary consists of Key:Value pairs, and the keys must be immutable while the values can be anything. Note that dictionaries themselves are mutable objects.

A dictionary is built with curly braces as follows:

ages = {"Me": 33, "You": 22, "Him":24}

If now I want to know your age, and then increase it, I would type:

print(ages["You"])
ages["You"] += 1
print(ages["You"])

You can add new items to your dictionary as follows:

ages["Her"] = 20
print(ages)

Dictionary objects have also their own methods. For instance, you can use keys to find out what are all the keys in the dictionary, or the values method to retrieve are all of the values in the dictionary:

names = ages.keys()
years = ages.values()

3.6 - Python typing

Objects in Python are composed by their name, their content, and a reference that points the name to the content. When an object is created, Python tells the computer to reserve memory locations to store the corresponding data. Depending on the data type of a variable, the interpreter allocates a certain amount of memory.

As we have seen above, in Python, there is no need of declaring objects nor their type before using them. This is because actually what you are doing is not creating a spot in memory and filling it with an object. Rather, you are creating a pointer (that occupies a first memory spot), and then making that pointer point to an object in a second memory spot. For this reason, you can for instance reassign a variable to a different type of object without errors:

a = [0,1,2]
a = (-1,3)
a = True

Note that when you assign a variable to another, you are just creating a second pointer to that same memory spot.

a = [0,1,2]
b = a
b[2] = 0
print(a)

3.7 Flow Control

Typical flow control structures are implemented as usual in Python.

Be careful with Python code indentation: the space you leave to the left of your piece of code implicitly delimits code blocks.

We will learn the behavior of flow control statements in Python by example, to reinforce the idea of how intuitive Python is.

3.7.1 - if-else statements

Observe the following code:

if 3>2:
    print('success')
elif 3==2:
    print('failure')
else:
    print('I do not know')
print('This will be printed either way')

Do not forget about the semicolon in the end of control statements!

3.7.2 - for loops

Observe the following code and try to predict its output:

for i in [0,1,2,3]:
    print(i)

3.7.3 - while loops

Observe the following code and try to predict its output:

i=0
while i < 4:
    print(i)
    i = i+1

3.7.4 - Other statements: break and continue

Observe the following two pieces of code and try to predict their output:

for i in [1,2,3,4,5,6,7]:
    if i % 3 == 0:
        continue
    print(i)
for i in [1,2,3,4,5,6,7]:
    if i % 3 == 0:
        break
    print(i)

Can you give a definition of both statements based on these experiments?

3.8 Python Functions

Being only able to ``interactively” play with variables is boring. To build more complex code, we need functions. Functions are tools for grouping statements so that they can be executed more than once in the same program. They are useful maximize code reuse and minimize code redundancy, therefore contributing to avoid errors.

Functions are written using the def statement. You can send objects created inside your function back to where it was called with the return statement.

def compute_sum(a,b):
    c = a+b
    return c

To use this function, we simply call it passing appropriate parameters:

a = 5
b = -2
print(compute_sum(a,b))

Arguments to Python functions are matched by position. Tuples are typically used to return multiple values. Note that functions themselves are objects:

print(type(compute_sum))

In general, variables created or assigned in a function are local of that function and exist only while the function runs.

L = [0,1,2]
def modify(my_list):
    c = 3
    my_list[0] += 20
modify(L)
print(L)
print(c)

It is also possible to specify a default value for some argument:

def compute_sum(a,b=2):
    c = a+b
    return c

Likewise, you can have keyword arguments. A keyword argument is an argument which is supplied to the function by explicitly naming each parameter and specifying its value:

print(1, 2, 3, 4, sep=';')

Keyword arguments must always go behind non-keyword arguments.

3.9 Classes and Object-Oriented Programming

How can you go beyond built-in data types and create new object types, with their associated methods and attributes defined by you?

Python allows you to create new classes, and then define (instantiate) new objects of that class and interact with them. This way, you can group data and functions operating on it in a more abstract way, and then instantiate concrete samples and use them. Classes allow for a simplified modeling of our problems, and enables the creation of cleaner code that will be more easily extended in the future.

When dealing with classes, data is usually called attributes, and functions methods.

3.9.2 - Building a new Class from scratch

Every class needs to have a special method, called constructor, that initializes its attributes.

class UP_student:
    def __init__(self, name, math_skills, coding_skills, hard_working, theory_mark, practical_mark):
        self.name = name
        self.math_skills = math_skills
        self.coding_skills = coding_skills
        self.hard_working = hard_working                
        self.theory_mark = theory_mark
        self.practical_mark = practical_mark  

You will note the presence of the self parameter: this is a special inner reference to the object state. It may take some time to understand the use of self, but do not be afraid, we will see some examples afterwards.

As it stands, an object of the UP_student class has very limited value, as it contains only data (attributes). Let us add some spice by giving our class a function (method):

class Y:
    def __init__(self, v0):
        self.v0 = v0
        self.g = 9.81
    def value(self, t):
        return self.v0 * t - 0.5*self.g*t**2

The utility of self starts to become clear now. At this point, you have created a useful class, and you can instantiate an object of this new type easily:

name = 'adrian_galdran'
adrian_math_skills = 0.9
adrian_coding_skills = 0.8
adrian_hard_working = True
adrian_student = UP_student(name, adrian_math_skills, adrian_coding_skills, adrian_hard_working)
print(type(a_student))

As you can see, we call our class as if it was a normal Python function, and Python automatically invokes the constructor method. __init__ requires several parameters to be specified at instantiation time. If you do not specify them correctly, you will get an error.

Now, attributes and methods are exposed to the user:

print(adrian_student.coding_skills)
print(adrian_student.hard_working)
print('Global Mark: ', adrian_student.compute_global_mark(0.2)) # Let us give more weight to the practical part!

How can you add new methods to your class? For instance, we can add a print_global_mark method that computes and prints the final mark automatically. This method only needs as input parameter theory_weight, and outputs a string:

class UP_student:
    def __init__(self, name, math_skills, coding_skills, hard_working, theory_mark = 5, practical_mark = 4):
        self.name = name
        self.math_skills = math_skills
        self.coding_skills = coding_skills
        self.hard_working = hard_working                
        self.theory_mark = theory_mark
        self.practical_mark = practical_mark 
    def compute_global_mark(self, theory_weight = 0.6):
        return theory_weight*self.theory_mark + (1-theory_weight)*self.practical_mark
    def print_global_mark(self, theory_weight = 0.6):
        global_mark = self.compute_global_mark(theory_weight)
        print('The final mark of ', self.name, ' is ', global_mark)

Note that even if the print_global_mark method only needs the theory_weight argument, we still must add the self argument so that it can call self.global_mark. This is omitted in the method call.

Notice also that inside the class, compute_global_mark is known to the object and needs no self parameter.

name = 'adrian_galdran'
adrian_math_skills = 0.9
adrian_coding_skills = 0.8
adrian_hard_working = True
adrian_student = UP_student(name, adrian_math_skills, adrian_coding_skills, adrian_hard_working)

adrian_student.print_global_mark()

We know an object consists of both internal data and methods that perform operations on the data. At some point you may find that existing object types do not fully suit your needs. Classes are the tool that allows you to create new types of objects.

3.9.2 - Class Inheritance

Sometimes, even if you have the need for a new type, it may happen that this new object type resembles, in some way, an existing one. Classes have the ability to inherit from other classes, and this is a fundamental aspect of OOP.

Let us see an example of how to build a new class, inheriting from the built-in Python list class. We will add more functionality to it.

class MyList(list):
    

this definition ensure that our new class, derived from list, will inherit the attributes of the base class. However, now we can extend, or redefine those attributes!

For instance, we are going to improve the built-in remove methods, implemented by Python for lists in this way:

L = [0,1,2,5,5]
L.remove(5)

We will add new methods to also be able to remove the maximum and minimum element of a list. For this, we complete the definition of our extended class as follows:

class MyList(list):
    def remove_min(self):
        self.remove(min(self))
    def remove_max(self):
        self.remove(max(self))

Now we can make use of our class:

L2 = MyList(L)
dir(L)
dir(L2)
print(L2.remove_min())

4. Complementary Python Scientific Computing Tools: Numpy

As mentioned in the introduction, one of the most important strengths of Python is the large ecosystem of tools available. One of the most important libraries for scientific computing in general (and for this course in particular) is NumPy, which is designed to perform matrix computations. Here you will learn the fundamental concepts related to Numpy.

4.1 - Introduction to NumPy Arrays

Python allows you to create nested lists, that you could use to work with n-dimensional arrays:

zero_matrix = [[0,0,0],[0,0,0]]
print(len(zero_matrix),len(zero_matrix[0]))

However, you can see this is quite inconvenient, and complexity will grow a lot with high dimensions. Also, Python lists are not designed for linear algebra. For instance, + acting on lists means concatenation:

print([1,2]+[0,0])

NumPy arrays are n-dimensional array objects which conform the core component of scientific computing in Python. They are an additional data type provided by NumPy for representing vectors and matrices. Unlike Python lists, elements of NumPy arrays are all of the same data type, and their size is fixed once defined. By default, the elements are floating point numbers.

This is a first example of how to build a vector and a matrix with all elements zero.

import numpy as np

zero_vector = np.zeros(4)
zero_matrix = np.zeros((2,3))

Note that in the matrix case, we need to specify the dimensions through a tuple. In order to build an array of ones, you can use the numpy.ones function, with the same syntax:

ones_matrix = np.ones((3,2))

Finally, you can manually initialize them using np.array and a Python list:

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

numpy supports the usual standard matrix operations, such as matrix transposition;

my_transposed_matrix = my_matrix.transpose()

Arithmetic operations work as expected also:

my_matrix + ones_matrix

Note that unlike MATLAB, * is elementwise multiplication, not matrix multiplication. You need to use the dot function to compute products of vectors, to multiply a vector by a matrix, and to multiply matrices. dot is available both as a function in the numpy module and as an instance method of array objects:

x = np.array([[1,2],[3,4]]) # 2x2 Matrix
y = np.array([[5,6],[7,8]]) # 2x2 Matrix

v = np.array([9,10]) # 1x2 vector
w = np.array([11, 12]) # 1x2 vector

# Matrix / vector product; both produce a 1x2 vector
print(x.dot(v))
print(np.dot(x, v))

# Matrix / matrix product; both produce a 2x2 matrix
# [[19 22]
#  [43 50]]
print(x.dot(y))
print(np.dot(x, y))

Finally, you can find out the dimensions of a given numpy array through its shape data attribute:

print(my_matrix.shape)

4.2 - Slicing NumPy Arrays

The same way you can slice Python lists, you can do the same with numpy arrays. Remember the indexing logic. Start index is included but stop index is not, so Python stops just before it reaches the stop index.

my_first_column = my_matrix[:,0]
my_last_row = my_matrix[-1,:]
print(my_first_column.shape)

4.3 - Indexing NumPy Arrays

NumPy arrays can also be indexed with other arrays or other sequence-like objects like lists. For example:

z1 = np.array([2,4,6,8,10])
z2 = z1+1
indexes_arr = np.array([0,1])
indexes_list = [0, 4]
print(z2, z2[indexes_arr], z2[indexes_list])

Another way of indexing numpy arrays is with logical indices:

indexes_logical = [True, True, True, False, False]
print(z2[indexes_logical])

Note the potential of this operation!

print(z2>5)
print(z2[z2>5])

Important difference between slicing and indexing

When you slice an array with the colon operator, you obtain a view of the object. This means that if you modify that view, you will also modify the original array. In contrast, when you index an array, what you obtain is a (new) object, a copy independent of the original one.

z1 = np.array([2,4,6,8,10])
w_view = z1[0:3] # sliced z1
print(z1)
print(w_view)
w_view[0] = 50
print(z1)

Compare the above code snippet with this one:

z1 = np.array([2,4,6,8,10])
indexes = [0,1,2,3,4]
w_copy = z1[indexes] # indexed z1
print(z1)
print(w_copy)
w_copy[0] = 50
print(z1)

4.4 - Other numpy arrays manipulation techniques

In numpy, if you want to build an array with fixed start and end values, such that the other elements are uniformly spaced between them, you can do the following:

np.linspace(0, 100, 10) # stop point is included

We have already seen an example of the attribute shape of a numpy array. You can also check the total size of it:

my_matrix = np.array([[2,1],[3,2],[5,4]])
print(my_matrix.shape)
print(my_matrix.size)

Notice that neither shape nor size are followed by a parenthesis. This is because they are not method attributes for the numpy array class, but rather data attributes.

There are a couple of handy logical operations that work on top of numpy arrays. For instance, you often will want to check if any/all of the elements in an array verifies a given condition. This can be accomplished with the methods any() and all():

x = np.random.random(10)
print(x)
print(np.any(x>0.5))
print(np.all(x>0.5))

In this case, instead of using the Python random library, we used the numpy.random module.

5. Complementary Python Scientific Computing Tools: Matplotlib

Matplotlib is the standard Python plotting library. Even if matplotlib is a very large library, it contains a module called pyplot. Pyplot is a collection of functionalities that make matplotlib work in a similar way as Matlab. In this course, we will use pyplot for our data visualizations.

5.1 - Basic Matplotlib usage

Let us import it:

import matplotlib.pyplot as plt

The most basic function inside pyplot is plot. Its simplest use case takes only one argument, specifying the y-axis values that are to be plotted. In this case, each y-axis value is plotted against its corresponding index value on the x-axis:

y = np.random.random(10)
plt.plot(y)

If you use plot outside the iPython shell, the plot is created but not shown. To tell Python to show the plot, you just need to add plt.show() to your code.

When you give to plot two arguments, the first argument specifies the x-coordinates and the second the y-coordinates.

x = np.linspace(0,10,100)
y = np.cos(x)
plt.plot(x,y)

You can also supply a third argument to plot in order to give some cosmetic specifications on your plot, like color, line type or marker. They work with key-word arguments.

x = np.linspace(0,10,20)
y = np.cos(x)
plt.plot(x, y, 'ro-')
plt.show()
plt.plot(x, y, 'gs-', linewidth=5, markersize=15)

Note that in this case plt.show() forces Python to show the first plot, which otherwise would be ommited.

5.2 - Plot Customization

Let us see some more advanced plot customization techniques. To add a legend to an already created (even if still not shown) plot, you can use legend(), which takes a string as an argument:

x = np.linspace(-3,3,20)
y = x**2
plt.plot(x, y, 'ro-')

If you want to add information on which quantities are specified on each axis, you can do it as follows:

x = np.linspace(-3,3,20)
y = x**2
plt.plot(x, y, 'ro-')
plt.xlabel('The x axis')
plt.ylabel('The y axis')

You can also customize what part of your plot you want to display with axis():

x = np.linspace(-3,3,20)
y = x**2
plt.plot(x, y, 'ro-')
plt.axis([-0.5, 2, -2, 4]) #xmin, xmax, ymin, ymax

It is also quite easy to overlay several plots:

x = np.linspace(-3,3,20)
y1 = x**2
y2 = x**3
plt.plot(x, y1, 'ro-')
plt.plot(x,y2, 'b+-')

If you need to add an independent legend to each of these, you need to label each of them separately:

x = np.linspace(-3,3,20)
y1 = x**2
y2 = x**3
plt.plot(x, y1, 'ro-', label = 'square')
plt.plot(x,y2, 'b+-', label = 'cubic')
plt.legend(loc = 'upper left')

Note that legend() can take a keyword argument specifying location.

Finally, to save your figure, you simple use savefig. The extension of the file name you choose will determine the format of the output.

x = np.linspace(-3,3,20)
y1 = x**2
plt.plot(x, y1, 'ro-', label = 'square')
plt.savefig('my_plot.png')

Finally, let us illustrate the use of histogram plotting tools in matplotlib, as well as how to build several subplots in the same plot. First, we create a normally distributed array of numbers around zero using numpy as follows:

x = np.random.normal(size = 1000)

To build a histogram plot in pyplot we type:

plt.hist(x)

Now, if you want to plot the same histogram with two different colors in two different subplots, you can use the plt.subplot() function. This function takes three arguments: the first two specify the number of rows and columns in the subplot, and the third one is the plot number.

plt.subplot(1,2,1)
plt.hist(x, color = 'r');
plt.subplot(1,2,2)
plt.hist(x, color = 'b');

Note the use of a semicolon after each plot execution, in order to avoid printing the value returned by matplotlib.

6. Homework

For now, you can access a notebook with an exercise on Python classes here. I will be adding another problem in the next days.

7. Sources and References

Of course, there are tons of wonderful Python resources in the internet. The main sources I used to build this lecture were:

  1. Jake Van der Plas’ book A Whilrwind Tour of Python, available here in pdf, and here in iPython notebook format.

  2. Harvard’s online course Using Python for Research, hosted at edX

  3. Stanford’s CS231n short Python tutorial here

In general, most of the material and presentation is shamelessly inspired in 2.

Regarding the numpy part, if you are a Matlab user, you could find this resource very useful: - Numpy for Matlab users - link

If you need or want more practice with Python and the tools presented today, I would recommend following the free course at datacamp, and doing all the exercises proposed there. Codeacademy exercises, hosted here, can also be very useful to get more experience.