Introduction to Python

Author: Andrew Montalenti
Date: 2015-06-01

How this was made

This document was created using Docutils/reStructuredText and S5.

It is the introductory Python course given by to the newest generation of Python hackers.

Simplicity begets elegance.


Exercises for this book are taken from four primary sources. These are:

Read slides on your own

Meta Information

Me: I've been using Python for 10 years. I use Python full-time, and have for the last 3 years. I'm the co-founder/CTO of, a tech startup in the digital media space.

E-mail me:

Follow me on Twitter: amontalenti

Connect on LinkedIn:

Other skills are fair game, too

  • Java
  • Groovy
  • JavaScript
  • SVN / Mercurial
  • Web development
  • Computer Science
  • ...

Slide Zero

Simplicity begets elegance.

>>> nums = [45, 23, 51, 32, 5]
>>> for idx, num in enumerate(nums):
...    print idx, num
0 45
1 23
2 51
3 32
4 5


Beautiful is better than ugly.

Explicit is better than implicit.

Simple is better than complex.

Complex is better than complicated.


Flat is better than nested.

Sparse is better than dense.

Readability counts.

Special cases aren't special enough to break the rules.

Although practicality beats purity.


Errors should never pass silently.

Unless explicitly silenced.

In the face of ambiguity, refuse the temptation to guess.

There should be one-- and preferably only one --obvious way to do it.

Although that way may not be obvious at first unless you're Dutch.


Now is better than never.

Although never is often better than right now.

If the implementation is hard to explain, it's a bad idea.

If the implementation is easy to explain, it may be a good idea.

Namespaces are one honking great idea -- let's do more of those!

The Zen of Python

It's embedded in the heart of the language.

>>> import this
The Zen of Python, by Tim Peters

Beautiful is better than ugly.
Explicit is better than implicit.

Principles behind Python

Time to a solution is determined by:
  • how long writing a program takes (human time)
  • how long it takes that program to run (machine time)
Every language was written with a tradeoff between these:
  • Java
  • C
  • C++
  • Perl

Time spent developing

Watch out for the increasing human cost of a project as time goes on!


Didier's decision

Didier is a cofounder and my favorite Python engineer on the planet.

My choices at the time were Java (the Eclipse-driven language), C (a language of last resort), C++ (a language that comes with reference book), Perl (a Babylonian language), and Matlab (closed source).

At the time, I was done with computer science and I needed a language that I could use for prototyping, like Matlab was, but more general.

Python meets expectations

There are many programming languages that meet the minimum criteria for a strong platform for modern software development. Python, like many of those options, is:

...and then exceeds them

Python is unique in that it also has:

My Job

My job is easy because Python was built as a teaching language.

I learned Python using an online book called Dive Into Python, by Mark Pilgrim.

The concept behind that book is that Python is such a readable language, that we can learn it by reading well-written Python programs.

We'll try to use this approach to good effect.

I have an opinion, and so will you

Python is an opinionated language, and I have an opinion about code.

In this class, you will learn to write opinionated code that has good style.

Our first program

>>> nums = [45, 23, 51, 32, 5]
>>> for idx, num in enumerate(nums):
...    print idx, num
0 45
1 23
2 51
3 32
4 5


>>> nums = [45, 23, 51, 32, 5]

Declares a label called nums, and binds it to a list value.

Could just as easily be written this way, although no Pythonista would ever do this.

>>> nums = list()
>>> nums.append(45); nums.append(23)
>>> nums.append(51); nums.append(32)
>>> nums.append(5)

The verbosity of Java

Contrast to Java:

// Java code
List<Integer> nums = new ArrayList<Integer>();
nums.add(45); nums.add(23);
nums.add(51); nums.add(32);

Or, at best:

List<Integer> nums =
    Arrays.asList(new Integer[] {45, 23, 51, 32, 5});

Concise syntax is Pythonic

In Java, I have even tapped into some obscure features, using Arrays.asList to initialize the ArrayList, leveraging the fact that arrays, but not lists, have a concise initialization syntax.

In Python, lists are declared using [item1, item2, item3, ...], a simple and concise syntax that can be easily read.

This is but one example of hundreds made in the language.

Dissection (con't)

>>> for idx, num in enumerate(nums):

Iterates over each item in the nums list, yielding a tuple for each iteration step that contains the index of the list value and the list value itself.

>> list(enumerate(nums))
[(0, 45), (1, 23), (2, 51), (3, 32), (4, 5)]

Dissection (still con't)

... print idx, num

The for loop has created two new bindings, idx and num, for each iteration step of the loop. These bindings are unpacked from the tuples yielded by the enumerate built-in function.

Dissection (final)

We can break this down by looking at how one step would work.

>>> idx, num = (0, 45)
>>> idx
>>> num
>>> print idx, num
0 45

Welcome to Python!

Two readable lines of code that do a whole lot of work for you. You've already been subtly exposed to some features we'll learn about in the upcoming course:

  • concise list syntax
  • variable bindings and scope
  • tuples
  • iterators (and even generators!)
  • value unpacking
  • print keyword

Development Environment Setup

Behind every good Python programmer is a good development environment.

For this course, I've suggested your company set up the following tools:

Quick note on WingIDE


Python, IPython, BPython, DreamPie

One of Python's best features is the REPL loop, or Read, Evaluate, Print Loop.

Open source developers have created a number of "enhanced" REPLs over the years. The most common of these is IPython, but there is also BPython and DreamPie.

WingIDE has its own REPL that is integrated with the IDE and works in debug mode, as well.

BREAK: Confirm system settings

... cue elevator music ...

The basics

In traditional compiled languages like C/C++, you think about user programs which run directly on the operating system.


Python shell

Python introduces one more layer of abstraction, the Python interpreter, that is like a mini operating system running above the actual operating system.

There is no separate "compilation" step like C/C++/Java.

You simply run the python command, and you can start evaluating code.

>>> print 1 + 2
>>> print 'charles' + 'darwin'

Python runtime

The python command, when run with no arguments, opens the interactive shell.

When run with arguments, it acts as a runtime for code that can be stored in files or provided at the command line.

python -c "print 1 + 2"

EXERCISES: Write our first programs

Label assignment

>>> planet = 'Sedna'
>>> print plant # note the deliberate misspelling
Traceback (most recent call last):
    print plant
    NameError: name 'plant' is not defined

Must assign a value to a label before it can be used. No compile-time check that label has already been assigned, so mis-spelled label name raises a NameError exception. This is a common source of bugs.

Strong typing

Though labels can be re-assigned among types at will, the actual values behind the labels do have types.

And Python does not go out of its way to coerce types that cross "semantic boundaries".

Minimal automatic coercion

>>> string = "two"
>>> number = 3
>>> string, integer = integer, string # swap values
>>> print string + number
Traceback (most recent call last)
    string + number
    TypeError: cannot concatenate 'int' and 'str' objects

This makes Python a dynamic, but strongly typed language. Contrast with Perl, which is both dynamic and weakly typed.


"Wait a second. Are you serious? Whitespace is significant in Python?"

"Where the hell are my curly braces!?"

>>> from __future__ import braces
SyntaxError: not a chance (<ipython console>, line 1)

You heard the interpreter.

Why indentation?

Short answer: because we do it anyway.

Better answer: because some of us don't.

Python Style Guide (PEP 8) recommends 4 spaces, and no tab characters.

Line continuation

One problem with significant indentation is how to break long statements on multiple lines.

Python has two options:

  • The \ character, which indicates that the end of a line has been reached and tells the interpreter to treat the next line as a continuation of the current line; best to avoid this one if possible
  • The (...), {...} and [...] characters, which create an implicit continuation; use this one if possible

Empty statements

Another problem with indentation is that it makes it harder to specify "empty statements". e.g., in JavaScript function() {} represents an "empty function". Python does this with a special keyword called pass, that simply does nothing. An example:

if normal_condition:
elif special_condition:

The middle condition is a "no-op". It's also used for stubs, e.g.:

def placeholder(): pass

Line continuation example

# prefer this:
"{foo} {bar} {baz}".format(
    foo=foo, bar=bar, baz=baz)
# to this:
"{foo} {bar} {baz}"\
    .format(foo=foo, bar=bar, baz=baz)

Documentation built-in

>>> import string
>>> dir(string)
['__builtins__', '__doc__', ..., 'atof', 'atof_error', 'atoi', ...]
>>> print string.count.__doc__
count(s, sub[, start[,end]]) -> int

    Return the number of occurrences of substring sub in string
    s[start:end].  Optional arguments start and end are
    interpreted as in slice notation.

help function

With no arguments, the help function opens an interactive help prompt.

But usually, you only need help on a specific function or module. So, then you can call help with an argument.

Example of help

>>> import string
>>> help(string)
Help on module string:

    string - A collection of string operations...





>>> print "Hello World!"
Hello World!
>>> cars = 100
>>> print "There are", cars, "available"
There are 100 cars available
>>> print??
Type:       builtin_function_or_method
    print(value, ..., sep=' ', end='\n', file=sys.stdout)
        Prints the values to a stream, or to sys.stdout by default.


#!/usr/bin/env python
# Written by John Doe, 6/5/2011
if __name__ == "__main__": # only runs when script is executed
    print "Hello, World"

Math and Numbers

>>> x = 3
>>> 1 < x < 3
>>> 2 < x < 5
>>> 3 == x
>>> x + 0.1

Python number literals are either int or float, depending on context. If an expression is a float, IEEE 754 floating point rules apply. Otherwise, normal integer math rules apply.

Float context

>>> type(0.1)
<type 'float'>
>>> type(3)
<type 'int'>
>>> type(3 + 0.1)
<type 'float'>


>>> from decimal import Decimal as D
>>> D("3") + D("0.1")
>>> D("3") + D("0.1") == D("3.1")

When dealing with non-scientific computations, you probably want to use the Decimal class. It is a little awkward to use -- you must specify your numbers as strings. However, it supports all expected operations directly on the Decimal object.

To make it a little easier to use, I imported it using a module alias D.


>>> from fractions import Fraction as F
>>> F(1, 3) + F(1, 2)
Fraction(5, 6)


There are two simple ways to do string formatting in Python, and both are very popular.

  • %, defined on str objects.
  • str.format, a more verbose method available on str.
>>> "%s cars crossed the intersection \
    in the last %s hours" % (5, 24)
5 cars crossed the intersection in the last 24 hours
>>> "{num} cars crossed the intersection \
    in the last {hrs} hours".format(num=5, hrs=24)
5 cars crossed the intersection in the last 24 hours


There are many programmers for whom strings play a much more vital role than number types.

I am one of those programmers.

My applications deal with the web and with large-scale text processing.

A core understanding of strings and a language with capabilities to manipulate them is critical to get work done in this environment.

The str literal

In Python, strings can be enclosed with single, double, or triple quotes.

>>> 'single'
>>> "double"
>>> """triple"""
>>> """though single and double are mostly interchangeable,
triple quotes have a special property: they ignore line
breaks. Thus, they can be used as a kind of 'heredoc'."""
"though single and double are mostly interchangeable,\n triple ..."

Special str modes

They may also be preceded with a special character prefix, indicating a special string mode. Currently, only two are supported: raw strings using prefix r, and unicode strings using prefix u.

>>> print('C:\new\node\nell.exe')
>>> print(r'C:\new\node\nell.exe')
>>> print(u"\u2192")

str methods

Strings support a healthy number of methods, including:

  • common transformations like lower and upper
  • conveniences like strip and startswith
  • utilities like find, replace, split, and join
>>> ". ".join("PYTHON IS GREAT".lower().split()) + "."
python. is. great.

str operators

Strings also support some great operators:

  • + for concatenation
  • [idx] for slicing
  • * for repeating
  • in for substring matching
>>> p = "PYTHON"
>>> g = ("GREAT " * 3).strip()
>>> "GREAT" in g
>>> p + (" %s " % "IS") + g[0:15] + "..."

Method chaining

When methods are chained, the intermediary results of computation are simply discarded.


Built-in namespace

Python automatically imports a slew of functions, types, and symbols, which are known collectively as built-ins.

These provide the "basic language constructs" before you start referencing the modules of the standard library.

>>> sorted(vars(__builtins__).keys())[-5:]
['tuple', 'type', 'unichr', 'unicode', 'vars', 'xrange', 'zip']
>>> sorted is __builtins__.sorted

Built-in functions

Built-in functions are covered in this document:

Some early ones to look at are:

All objects are first-class

>>> line = "GOOG,100,490.10"
>>> field_types = [str, int, float]
>>> raw_fields = line.split(",")
>>> fields = [ty(val) for ty, val in zip(field_types, raw_fields)]
>>> fields
['GOOG', 100, 490.1000000000002]

The fact that everything in Python is first-class is often not fully appreciated by new programmers.

How would you have written this in Java/C#?

EXERCISES: Strings, Printing and Input

True, False, and None

Among the imported symbols from __builtins__ are True, False, and None.

These core symbols used in boolean logic, and are typically utilized with keywords such as is, not, and, or, etc.

Object equality and identity

>>> x, y = (0, 1)
>>> y == True # aka, "y is truthy"
>>> x == False # aka, "x is falsy"
>>> y is True # aka, "y is True singleton"
>>> x is False # aka, "x is False singleton"
>>> y is not True
>>> y is not False

Control structures

if login_method == "admin" or user.staff:
    print "allow login for admin"
elif authenticate(username, password):
    print "allow login for user"
    raise AuthenticationError

Truth tables

Logical operations can be visualized with a truth table.


Lack of switch statement

// Java code
switch (file.getType()) {
    case FileTypes.HTML:
        return "HTML Document";
    case FileTypes.DOC:
        return "MS Word";
    case FileTypes.EXCEL:
        return "MS Excel";
        return null;
    // ...

Using an elif chain

ftype = file.type
if ftype == "text/html":
    return "HTML Document"
elif ftype == "application/ms-word":
     return "MS Word"
elif ftype == "application/ms-excel":
    return "MS Excel"
    return None


>>> x = []
>>> x.append(5)
>>> x.extend([6, 7, 8])
>>> x
[5, 6, 7, 8]
>>> x.reverse()
>>> x
[8, 7, 6, 5]

Lists and memory

Lists are reference types, which means they simply contain pointers to objects that exist elsewhere.


List alterations

Lists can be altered (mutated) at will, which does not require recreation of the entire list.


List concatenation

Lists can be concatenated with other lists using extend, which creates a new list but does not require reallocating all the data.


List slicing

Lists have a convenient syntax for accessing single elements or even ranges of elements. This is called slicing and can also be done with the slice builtin function or [i:j:stride] syntax.


List aliasing

Lists can be aliased simply by binding a new label to the list. This sometimes leads to bugs!


List nesting

Lists can also be arbitrarily nested, or contain other types altogether like tuples, sets, or dictionaries.



>>> d = {}
>>> d['a'] = 5
>>> d['b'] = 4
>>> d['c'] = 18
>>> d
{'a': 5, 'c': 18, 'b': 4}
>>> d['a']
>>> e = dict(a=5, b=4, c=18)
>>> e
{'a': 5, 'c': 18, 'b': 4}

Dictionary visually

Dictionaries contain a "magically stored" set of items, which are key to value mappings.


Switch statement to dictionary

long = file.type
long2short = {
    "text/html": "HTML Document"
    "application/ms-word": "MS Word"
    "application/ms-excel": "MS Excel"
return long2short.get(long, None)


>>> ppl = ["Andrew", "Joe", "Bob", "Joe", "Bob", "Andrew"]
>>> ppl = set(ppl)
>>> print ppl
set(["Andrew", "Joe", "Bob"])
>>> trainers = set(["Andrew"])
>>> print ppl - trainers
set(["Joe", "Bob"])

Sets visually

Similarly to dictionaries, sets "magically" store their values. Since sets can only store unique (hashable) values, it will remove duplicates from other collections like lists. Sets also support fast set operations like union, intersection, addition and subtraction.


Using in with dicts and sets (1)


for key in d:
    print key


for key in d.keys():
    print key

Use in where possible (2)

For consistency, use key in dict, not dict.has_key():

# do this:
if key in d: something with d[key]

# not this:
if d.has_key(key): something with d[key]

Think data, not code


A list is a mutable heterogeneous sequence

A tuple is an immutable heterogeneous sequence

i.e., a list that can't be changed after creation

Why provide a less general type of collection?

Tuple example

>>> primes = (2, 3, 5, 7)
>>> print primes[0], primes[-1]
2 7
>>> empty_tuple = ()
>>> print len(empty_tuple)
>>> one_tuple = (0,)
>>> print len(one_tuple)

Must use (val,) for one-tuples, due to some grammar ambiguity. (...) overloaded for tuples, function invocation syntax, and operator grouping!

Tuples very handy for nested sequences

>>> pairs = ((1, 10), (2, 20), (3, 30), (4, 40))
>>> for low, high in pairs:
...   print low + high

enumerate for indexed iteration

// Java code
List<String> colors = new ArrayList<String>();
colors.add("yellow"); colors.add("magenta"); // ...
for (int i = 0; i < colors.size(); i++) {
    String color = colors.get(i);
    System.out.println(i + " " + color);

Quite a contrast when you compared to this!

# Python code
>>> items = ['yellow', 'magenta', 'lavender']
>>> for i, name in enumerate(colors):
...   print i, name

Sequence protocol

A Python sequence is any type that supports these operations:


Sequence examples

Examples of sequences:

EXERCISES: Lists and Sorting

EXERCISES: Dictionaries and logic


def divide(a, b):
    """Divides operands a and b using integer division.
    Returns a quotient and remainder of division
    operation in a 2-tuple."""
    q = a // b
    r = a - q*b
    return q, r

Using functions

>>> from mymath import divide
>>> help(divide)
Help on function divide in module mymath:

divide(a, b)
    Divides operands a and b using integer division.
    Returns a quotient and remainder of division
    operation in a 2-tuple.

def is like any other statement

Logic flows around the function body, then re-enters it upon invocation.


Calling functions

>>> divide(5, 2)
(2, 1)
>>> quotient, remainder = divide(5, 2)
>>> print "quotient is %s, remainder is %s" \
...     % (quotient, remainder)
quotient is 2, remainder is 1
>>> return_value = divide(5, 2)
>>> print "quotient is %s, remainder is %s" \
...     % return_value
quotient is 2, remainder is 1

Keyword arguments

def connect(host, port=80, scheme="http", timeout=300):
    http = HTTPClient()
        port=port), timeout=timeout)
    return http

Argument types

Type checking anti-pattern

def connect(domain, port=80, scheme="http", timeout=300):
    if not isinstance(domain, basestring):
        raise ValueError("domain must be a string")
    if not isinstance(port, int):
        raise ValueError("port must be an integer")
    if not isinstance(timeout, int):
        raise ValueError("timeout must be an integer")

Why is this bad?

In many languages and communities, the code on the last slide isn't bad.

But, in the Python community, it is.

Arbitrary rigidity

The reason for this is a distaste for arbitrary rigidity:

Take a deep breath

This concept, duck typing, is near and dear to many a Pythonista's heart.

Eager checking anti-pattern

def connect(domain, port=80, scheme="http", timeout=300, client=None):
    if client is not None:
        http = client
        http = HTTPClient()
    # ... setup as before ...
    if not hasattr(http, "connect"):
        raise ValueError("client missing connect attribute")
    if not callable(http.connect):
        raise ValueError("client's connect isn't a method")
    return http

Why is that bad?

Though eager checking is a virtue in many communities (e.g. Java), in Python, this is very much looked down upon.

Python is built on a different principle known affectionately in the community as EAFTP. That's: "Easier to Ask For Forgiveness Than Permission"

Be brave, we're all adults here

In the code on the last slide, it was like we were walking on eggshells with the value passed to us by the user of the function.

The error messages are nice, the code is correct, but the essence of the function is polluted by the excessive checking.

Why not just try to do what your function expects?

Forgiving, but not pedantic, code

def connect(domain, port=80, scheme="http", timeout=300, client=None):
    if client is not None:
        http = client
        http = HTTPClient()
    except (AttributeError, TypeError) as ex:
        log.error("client may not support protocol")
        raise ex

One more deep breath

EAFTP is another concept near-and-dear to a Pythonista's heart.

The general idea is, rely more on exceptions and exception handling, than on validation and checking.

In addition to code cleanliness benefits, this often has performance benefits, too!

Quirks of optional arguments

A subtle, but common, bug

def download(domain, clients=[]):
    if len(clients) == 0:
        # what's wrong here?
        for client in clients:
    except "...":

A subtle, but common, bug

def download(domain, clients=[]):
    if len(clients) == 0:

EXERCISES: Write a functional program to spec

Expect to receive a URL of the format you would find on the web:


Implement a function, url_parse, that splits this string into dictionary with the component parts, including: scheme, port, host, path, fragment (hash), query string. For the above, it would be:

{ "scheme": "http", "host": "",
"path": "/in/andrewmontalenti",
"port": 80, "fragment": None, "query": None }

URL Parser

def url_parse(url):
    assert url is not None
    scheme, rest = url.split(":", 1)
    rest = rest[2:]
    offset = rest.find("/")
    if offset == -1:
        host = rest
        path = None
        host = rest[0:offset]
        path = rest[offset:]
    if ":" in host:
        host, port = host.split(":", 1)
        port = None
    return dict(
        Scheme=scheme, Host=host,
        Port=port, Path=path)

Module self-test

if __name__ == "__main__":
    print "Running test cases... ",
    url = ""
    parts = url_parse(url)
    assert parts["Scheme"] == "http", "scheme must match"
    assert parts["Host"] == "", "host must match"
    assert parts["Port"] is None, "port must match"
    assert parts["Path"] == "/in/andrewmontalenti", "path must match"
    print "OK."

Simple URL formatter

def format_url(d):
    fmt = "{scheme}://{host}:{port}{path}"
    url = fmt.format(**d) #** ignore that operator for now
    if ":80" in url:
        url = url.replace(":80", "")
    return url

Integration test

url = ""
# end-to-end test
assert format_url(url_parse(url)) == url

EXERCISES: Strings and Lists

HOMEWORK: Chapters 1-5

Will give more detail and color on the topics we covered today.

Day 2

Day 1 Recap

Python is a dynamic and opinionated language.

On your first day, you learned the basics:

Day 2 excitement

Day 1 was to learn enough to be dangeorus.

Day 2 is to be dangerous.

Subroutines and repetetive code

# step 1 -- convert strings to numbers
for item in data:
    item[0] = int(item[0])
    item[1] = float(item[1])
# step 2 -- calculate average for column 1
col1_data = []
for item in data:
col1_avg = sum(col1_data) / len(col1_data)
# step 3 -- calculate average for column 2
col2_data = []
for item in data:
col2_avg = sum(col2_data) / len(col2_data)
# step 4 -- output result to file
outfile = open("out.txt", "w")

Rewritten using subroutines

def convert(data):
    for item0, item1 in data:
        item[0] = int(item0)
        item[1] = float(item1)
def calc_avg(data, col_idx):
    col_data = []
    for cols in data:
    col_avg = sum(col_data) / len(col_data)
    return col_avg
def write_file(fname, values):
    outfile = open(fname, "w")
avg0 = calc_avg(data, 0)
avg1 = calc_avg(data, 1)
write_file("out.txt", (avg0, avg1))

Module structure

"""module docstring"""

# imports
# constants
# exception classes
# interface functions
# classes
# internal functions & classes

def main(...):

if __name__ == '__main__':
    status = main()

Module with cmdline parsing (1)

Module docstring.

import sys
import optparse

def process_command_line(argv):
    if argv is None:
        argv = sys.argv[1:]

    # initialize the parser object:
    parser = optparse.OptionParser(
    # ... continued

Module with cmdline parsing (2)

# .. function above ..
        '-h', '--help', action='help',
        help='Show this help message and exit.')

    settings, args = parser.parse_args(argv)

    # check number of arguments, verify values, etc.:
    if args:
        parser.error('program takes no command-line arguments; '
                    '"%s" ignored.' % (args,))

    # further process settings & args if necessary

    return settings, args

Module cmdline parsing (3)

def main(argv=None):
    settings, args = process_command_line(argv)
    return 0

if __name__ == '__main__':
    status = main()

EXERCISES: Your First Python Module

Imagine you have a data format that looks like this:


The first column represents an individual's age and the second column their test score on a standardized test.

You want to write a Python module that will process a full file of that data, but the file will be perhaps tens of thousands of records long.

tabulardata module

def get_col(rows, idx=0, debug=False):
    """Returns a list of values from the column with index idx."""
    col_vals = []
    for row in rows:
        if debug:
            print col_vals
    return col_vals

def avg(nums):
    """Returns a single number value that is the average of all
    values in sequence nums."""
    return sum(nums) / len(nums) script

from tabulardata import get_col, avg
# step 1: read in the data from the file
rows = []
for line in open("in.txt"):
    # step 2: clean the line and convert into columns
    line = line.strip()
    line = line.split(",")
    age = int(line[0])
    score = float(line[1])
    row = (age, score)
# step 3: calculate averages
ages, scores = get_col(rows), get_col(rows, idx=1)
avg_age, avg_score = avg(ages), avg(scores)
# step 4: output average to a file
outfile = open("out.txt", "w")
outline = "%s,%s" % (avg_age, avg_score)


data = []
for line in open('data/commented-data.txt'):
    if line.startswith("#"):
print data

Writing to files

data = [1, 2, 3, 4, 5]
f = open('data/output.txt', 'w')
for item in data:
    # alternative technique
    print >> f, item

String IO

data = [1, 2, 3, 4, 5]
f = cStringIO()
for item in data:

Bufferred reading

reader = open('data/huge.txt')
data =
while data != '':
    data =

This tends not to be necessary; only for larger files.

Lazy programmer reading

reader = open('data/sizable.txt')
lines = reader.readlines()
sorted(int(line) for line in lines)

Explanation of that last line will come soon!

Quick aside on line endings

Unix: newline ('n')

Windows: carriage return + newline ('rn')

Oh dear... Python converts 'rn' to 'n' and back on Windows. To prevent this (e.g., when reading image files) open the file in binary mode:

reader = open('mydata.dat', 'rb')

Modules and scripts

The organization of your code source files can help or hurt you with code re-use.

Most people start their Python programming out by putting everything in a script:

#!/usr/bin/env python
for i in range(0, 10):
   print i**2

This is great for experimenting, but you can't re-use this code at all!

Short note on UNIX

#!/usr/bin/env python

UNIX folk: note the use of #!/usr/bin/env python, which tells UNIX to execute this script using whatever python program is first in your path. This is more portable than putting #!/usr/local/bin/python or #!/usr/bin/python in your code, because not everyone puts python in the same place.)

Scripts are not reusable

#!/usr/bin/env python
def squares(start, stop):
   for i in range(start, stop):
      print i**2
squares(0, 10)

I think that's a bit better for re-use -- you've made squares flexible and re-usable -- but there are two mechanistic problems.

Problems with scripts

First, it's named, which means it can't readily be imported. (Import filenames have to be valid Python names, of course!)

And, second, were it importable, it would execute squares(0, 10) on import - hardly what you want!

Why import mechanism exists

Namespaces are a great idea!


Convert a script to a module

To fix the first, just change the name:

#!/usr/bin/env python
def squares(start, stop):
  for i in range(start, stop):
      print i**2
squares(0, 10)

Good, but now if you do import calc_squares, the squares(0, 10) code will still get run!

There are a couple of ways to deal with this. The first is to look at the module name: if it's calc_squares, then the module is being imported, while if it's __main__, then the module is being run as a script.

Module-and-script pattern

#! /usr/bin/env python
def squares(start, stop):
  for i in range(start, stop):
     print i**2
if __name__ == '__main__':
  squares(0, 10)

Now, if you run directly, it will run squares(0, 10); if you import it, it will simply define the squares function and leave it at that. This is probably the most standard way of doing it.

Module-and-tests pattern

I actually prefer a different technique, because of my fondness for testing. (I also think this technique lends itself to reusability, though.) I would actually write two files.

__all__ = ["squares"]
def squares(start, stop):
  for i in range(start, stop):
     print i**2
def _test_squares():
if __name__ == `__main__`:
    import nose

Script file separate

# calc-squares:
#!/usr/bin/env python
import squares
squares.squares(0, 10)

Why split our files this way?

First, this is eminently reusable code, because is completely separate from the context-specific call.

Second, you can look at the directory listing in an instant and see that is probably a library, while calc-squares must be a script, because the latter cannot be imported.

Why split (con't)

Third, you can add automated tests to (as described below), and run them simply by running python

Fourth, you can add script-specific code such as command-line argument handling to the script, and keep it separate from your data handling and algorithm code.


A Python package is a directory full of Python modules containing a special file,, that tells Python that the directory is a package.

Packages are for collections of library code that are too big to fit into single files, or that have some logical substructure (e.g. a central library along with various utility functions that all interact with the central library).

Sample package

For an example, look at this directory tree:

package/    -- contains functions a(), b()       -- contains function c()
  subdir/ -- contains function d()

From this directory tree, you would be able to access the functions like so:

import package
import package.other
import package.subdir

What is

Note that is just another Python file; there's nothing special about it except for the name, which tells Python that the directory is a package directory. is the only code executed on import, so if you want names and symbols from other modules to be accessible at the package top level, you have to import or create them in

Why use packages?

There are two ways to use packages:

  1. you can treat them as a convenient code organization technique, and make most of the functions or classes available at the top level
  2. you can use them as a library hierarchy. packages

In the first case you would make all of the names above available at the top level:

  from other import c
  from subdir import d

which would let you do this:

import package

Flat is better, right?

That is, the names of the functions would all be immediately available at the top level of the package, but the implementations would be spread out among the different files and directories.

I personally prefer this because I don't have to remember as much ;)

Submodule-heavy packages

The down side is that everything gets imported all at once, which (especially for large bodies of code) may be slow and memory intensive if you only need a few of the functions.

If you wanted to keep the library hierarchy, just leave out the top-level imports. The advantage here is that you only import the names you need.

However, you need to remember more.

Discussion on package trees

Some people are fond of package trees, but I've found that hierarchies of packages more than two deep are annoying to develop on.

You spend a lot of your time browsing around between directories, trying to figure out exactly which function you need to use and what it's named.

Using __all__ for cleanup

You can restrict what objects are exported from a module or package by listing the names in the __all__ variable. So, if you had a module that contained this code:

__all__ = ['fn1']
def fn1(...):
def fn2(...):

then only 'some_mod.fn1()' would be available on import. This is a good way to cut down on "namespace pollution" -- the presence of "private" objects and code in imported modules -- which in turn makes introspection useful.

with statement (1 of 2)

# step 1: open file
reader = open('foo.txt')
# step 2: process data in file
for line in reader:
# step 3: close file

This is a common enough pattern that Python has a way to do it automatically.

with statement (2 of 2)

with open('foo.txt') as reader:
    for line in reader:
# automatically closes file upon exit of block

This is one of Python's nicest protocols, called context manager.

Context manager

A context manager is any object that implements two special methods, __enter__ and __exit__.

This ensures common handling for any procedure that tends to have a setup step and a cleanup step surrounding actual processing. Common examples:

  • Closing files cleanly after processing them.
  • Setting a decimal context / precision for a certain bit of code, and returning it to original state afterwards.
  • Acquiring a lock and releasing it afterwards.

Dunder protocols

The context management protocol is an example of a "Dunder Protocol".

Rather than rely on type heirarchies like other languages, Python has a variety of cross-cutting protocols that are implemented using methods with a double-underscore prefix and suffix. These are sometimes referred to as "dunder" methods.

Dunder examples

Some other dunders:

Format dunders

Dunder for deep magic only

It has become popular in some open source projects to make use of dunders for internal names. Imitation is a sincere form of flattery, but this is very much recommended against.

The Python maintainers chose dunder because they figured it was a safe place to put internal Python protocols that are not likely to ever conflict with user code. You can have your own .len(), ._len() and .__len() methods, but __len__ is reserved as a protocol with special systemwide meaning.

But Dunders are necessary!

Conversely, many programmers shy away from dunders altogether, thinking that Python maintainers did not intend you to make use of them.

Quite to the contrary: many dunders are downright essential for Python programming, e.g. __init__, __repr__ and __hash__.

PEP8 Discussion

You may have noticed that a lot of Python code looks pretty similar.

This is because there's an "official" style guide for Python

It's called PEP 8.

It's worth a quick skim, and an occasional deeper read for some sections.

PEP8 Rules

Here are a few tips that will make your code look internally consistent, if you don't already have a coding style of your own:

  • use four spaces (NOT a tab) for each indentation level;

  • use lowercase, _-separated names for module and function names, e.g.


PEP8 Rules (con't)

  • use CapsWord style to name classes, e.g. MySpecialClass;
  • use '_'-prefixed names to indicate a "private" variable that should not be used outside this module, , e.g. _some_private_variable;
  • use '__'-prefixed names to indicate truly private variables/methods -- Python's object type has special handling of these, in fact

Regular expressions

A regular expression is a powerful domain-specific language for text pattern matching.

Python, like most other modern languages, provides standard library support for this powerful tool.

Basic re module

>>> import re
>>> RE_EMAIL = r'[\S][email protected][\S]+'
>>> re.match(RE_EMAIL, '[email protected]')
<_sre.SRE_Match object at 0x11acac0>
>>> re.match(RE_EMAIL, '[email protected]')

Dissecting the e-mail RegExp

>>> RE_EMAIL = r'[\S]+@[\S]+'

[\S] means the character class containing any non-whitespace character. Could also be written:

[^ \t\n\r\f\v]

+ symbol means that any char from that class must occur one or more times.

@ character is treated as a literal here.

The match object

>>> match = re.match(RE_EMAIL, '[email protected]')
>>> for method in ("span", "start", "end", "groups"):
...     print method, getattr(m, method)()
('span', (0, 20))
('start', 0)
('end', 20)
('groups', ())

Matching with groups

Regular expressions tend to be used for two purposes: validation and partial matching.

Validation: if Match object is returned, it passes; if None is returned, it fails.

Partial Matching: If Match object is returned, we break the string into groups.

Imagine if we needed to refer to the username/domain separately in an email address.

Rewrite email RegExp with groups

>>> import re
>>> RE_EMAIL = r'(?P<username>[\S]+)@(?P<domain>[\S]+)'
>>> match = re.match(RE_EMAIL, '[email protected]')
>>> match.groups()
('test', '')
>>> match.groupdict()
{'domain': '', 'username': 'test'}

Enhanced email matching

import re
RE_EMAIL = re.compile(r'''
    (?P<username>[\S]+) # match username portion
    @                   # at sign (@) literal
    (?P<domain>.*)      # match domain portion
''', re.VERBOSE)
match = RE_EMAIL.match('[email protected]')
print match.span()
print match.groups()
print match.groupdict()
email = match.groupdict()
print "User entered email domain %s" % email["domain"]
bad = re.match(RE_EMAIL, "[email protected]")
print bad

EXERCISES: Sorting and Data

EXERCISES: Working with File Formats

"Public" and "private" functions

Many other programming languages (C++, Java) devote many of their keywords to access control -- think keywords like private, public, and protected.

It may be a surprise that Python has none of that.

Indeed, most Python programmers take "We're all adults here" to its logical conclusion, and declare that privacy is overrated in programming.

Privacy conventions

__all__ = ["square"]
_CACHE = [_square(x) for x in range(_SIZE_CACHE)]
def square(x):
    if x < 0 or x > _SIZE_CACHE:
        # value not in cache
        return _square(x)
    # value is cached, return directly
    return _CACHE[x]
def _square(x):
    return x**2

Leading underscore on _CACHE, _SIZE_CACHE, and _square make it clear they aren't meant to be used directly.

Modules vs. Singletons

Modules are basically singletons.

Python programmers don't create singletons, they use modules.

Modules can contain functions, classes, values, and, through the package mechanism, other modules.

Good module organization == good code organization in Python.

Importing functions

>>> import pprint
>>> from pprint import pprint as pp

Importing functions from other modules is probably the simplest form of code reuse available in the Python language

The import keyword function has a shorthand for aliasing a symbol with the as keyword

The from...import syntax tends to be preferred to direct imports

Consider the alternative

>>> import pprint
>>> pprint = pprint.pprint

Eek, too many pprints!

First, a module.

Then, a label.

Then, a function within a module of the same name!

Aliasing functions

>>> from pprint import pprint, pformat
>>> pf = pformat
>>> pf is pformat


An integer is 32 bits of data...

...that labels can refer to

A string is a sequence of bytes representing characters...

...that labels can refer to

A function is a sequence of bytes representing instructions...

...and yes, labels can refer to them to

This turns out to be very useful, and very powerful

First-class built-ins

>>> def positive(x): return x >= 0
>>> print filter(positive, [-3, -2, 0, 1, 2])
[0, 1, 2]

>>> def negate(x): return -x
>>> print map(negate, [-3, -2, 0, 1, 2])
[3, 2, 0, -1, -2]

>>> def add(x, y): return x+y
>>> print reduce(add, [-3, -2, 0, 1, 2])

First-class functions

>>> from pprint import pprint
>>> import sys
>>> def render(data, methods=None):
...     if methods is None:
...         methods = (,)
...     for method in methods:
...         method(data)
>>> out_methods = (
...         pprint,
...         lambda data: sys.stdout.write(repr(data))
...     )
>>> render(methods=out_methods)

Dissecting first-class functions

out_methods = (
   lambda data: sys.stdout.write(repr(data))

pprint is just another label, it happens to point at a function.

lambda is a special keyword for creating simple anonymous functions using a concise syntax.

functions don't have to live anywhere in particular; here, we're inserting them into a 2-tuple labeled out_methods.

First-class functions go anywhere

>>> render(methods=out_methods)

Yes, you just passed a 2-tuple of functions to another function as an argument.

So what?

Other languages make this needlessly difficult (think Anonymous Inner Classes in Java, or Runnable interface). In Python, no problem.

Higher order function

So, you can pass functions around to other functions.

Just the same, functions can also return functions.

These are sometimes known as higher-order functions.

A function factory

We declare a function that builds functions.

from re import match
def regex_factory(pattern):
    def matcher(string):
        match = re.match(pattern, string)
        return match.groups()
    return matcher

And now, to use it:

>>> m = regex_factory("([A-Z]+)([0-9]+)")
>>> m("ABC009")
('ABC', '009')

Optimizing the factory

def precompile_factory_enhance(slow_factory):
    """Higher-order function that optimizes a slow
    function factory that accepts regex patterns
    as its only arg and makes it a fast function
    that uses a compiled pattern."""
    def fast_factory(pattern):
        compiled = re.compile(pattern)
        return slow_factory(compiled)
    return fast_factory
# replace our old regex_factory with the enhanced one
regex_factory = precompile_factory_enhance(regex_factory)
>>> m = regex_factory("([A-Z]+)([0-9]+)")
>>> m("ABC009")
('ABC', '009')


def simple_func(arg1):
simple_func = memoize(simple_func)
simple_func = debug_log(simple_func)

Decorators were meant to make the use of higher-order functions easier in Python. Code like the above used to be common, but now, it can be written more simply.

Decorator example

def simple_func(arg1):

Star argument application syntax

Before we talk about decorators, though, we have to understand some more mechanics about functions. The star argument syntax:

def decorator(fn):
    def wrapper_fn(*args, **kwargs):
        # ... do something before ...
        val = fn(*args, **kwargs)
        # ... do something after ...
        return val
    return wrapper_fn

def my_fn(arg1, arg2, default1=None):

EXERCISES: Functional programming

  • Write a debug_log higher-order function that outputs arguments passed to a function using pprint before entering a function and outputs the return value after leaving.
  • Write a memoize higher-order function that takes a function as its argument and caches all calls to the underlying function using a dictionary.
  • Use the two together for optimizing a fibonacci sequence generator function.

Day 3

Day 1 was to learn enough to be dangeorus.

Day 2 you became dangerous.

Day 3, young grasshopper, you become Pythonic.

The iterator

it = iter(s)
while 1:
        item =
    except StopIteration:

The for loop is "iteration sugar"

for item in s:

This is the right way to write it.

Implement an iterator

If a class supports __iter__, it is said to be Iterable.

The object returned by __iter__ needs to support two methods: __iter__ and next. This object is an Iterator. Think of it like a "cursor" on the sequence, or as a "yielder of values" from the sequence.

The iterator expression

items = [process(item) for item in s]

Python also supports a special syntactic construct that is particularly popular among Pythonistas, known as the iterator expression or list comprehension.

It is basically a declarative version of this code:

items = []
for item in s:

List comprehension

items = [process(item)
            for item in s
            if item in process_list]

These can be very powerful. The above processes and filters a list at once.

List comprehension example (1)

>>> [n ** 2 for n in range(10) if n % 2]
[1, 9, 25, 49, 81]

List comprehensions example (2)

>>> words = 'The quick brown fox jumps over the lazy dog'.split()
>>> [(w.upper(), w.lower(), len(w)) for w in words]
[('THE', 'the', 3),
('QUICK', 'quick', 5),
('BROWN', 'brown', 5),
('FOX', 'fox', 3),
('JUMPS', 'jumps', 5),
('OVER', 'over', 4),
('THE', 'the', 3),
('LAZY', 'lazy', 4),
('DOG', 'dog', 3)]

List comprehensions example (3)

>>> import os
>>> from glob import glob
>>> [f for f in glob('*.py*') if os.stat(f).st_size > 6000]

dict and set comprehensions?

>>> fstats = dict([(f, os.stat(f)) for f in glob('*py*')])
>>> owners = set([os.stat(f).st_uid for f in glob('*py*')])
>>> fstats[""]
posix.stat_result(st_mode=33188, st_mtime=1307117107, ...)
>>> 1000 in owners

Double listcomps!

container = [[1, 2, 3], [2, 3, 4]]
items = [item
            for nested in container
            for item in nested]

This may be confusing. What does this do?

dict.items vs. dict.iteritems

When you start coding declaratively and operating on large datasets, you start to worry about memory.

dict.items returns all the items of a dictionary as a brand new list. If your dictionary is big, this can be very wasteful.

dict.iteritems returns an iterator over the items of your dictionary. Combined with list comprehensions, this can be a memory-efficient way to work with your data.

In Python 3, iteritems is gone and items now returns an iterator. If you want a true clone, you can use list(d.items()).

EXERCISES: Iterators and Iterator Expressions

Create a list of 1000 elements:

l = range(1000)

And now, perform the following operations on it, generating new lists with list comprehensions:

  • Return a list with the square of every value in the list
  • Return a list with only those numbers whose string representation is exactly two characters, as strings
  • Return a list half the size with pairwise sums

Reading code

>>> [l[i] + l[i+1] for i in range(0, 1000, 2)][0:5]
[1, 5, 9, 13, 17]

Just because you can doesn't mean you should!

The generator

Generators take iterators one step further by allowing them to be easily written and lazily evaluated.

def infinity():
    i = 0
    while 1:
        i += 1
        yield i

Yes, this yields an infinite sequence of numbers. (Well, plus or minus overflow)

But calling infinity doesn't yield them all at once. It gives you a generator object, which is an iterator that will return the values specified in the yield expression!

Generator under the hood

Auto-generates the __iter__() and next() methods.

Saves application state up to the yield keyword.

Raises StopIteration when the generator terminates.

In combination, these features make it easy to create iterators with no more effort than writing a regular function.

The generator object

>>> inf = infinity()
>>> inf
<generator object infinity at ... >
>>> for i in range(1000):
...     print,
3 4 5 6 7 8 9 ...
>>> for idx, val in enumerate(infinity()):
...     if 1000 > idx > 1005:
...         print idx,
...     if idx == 5000:
...         break
1001 1002 1003 1004

yield keyword

You can think of the yield keyword similarly to the return keyword, but with a twist.

The function containing the yield is replaced with one that simply returns a generator.

The generator only executes your function upon first call of the .next() method. The function runs until it reaches a yield, and when it does, it yields the value and returns it from .next(). But then, execution stops.

Until the next .next().


Coroutines are a more exotic form of generator built by using the yield keyword as an expression rather than as a statement, e.g.

def comment_filter(target):
    while True:
        line = (yield)
        if not line.startswith("#"):

Use a coroutine

So, how do we use one of these things?

def printer_sink():
    while True:
        recv = (yield)
        if recv.startswith("#"):
printer = printer_sink() # initialize
sink = comment_filter(printer())
sink.send("# commented line")
sink.send("regular line") # only this will output

Yes, exotic, but great for setting up efficient data pipelines.

Built-ins: range vs. xrange

Remember range from prior exercises/slides?

Well, it has a little brother. His name is xrange.

xrange yields values in a range, rather than preallocating a big list for it.

Similarly to iteritems/items, xrange is removed in Py3 and range operates this way by default.

Our own range

def range(num_ints):
    vals = []
    i = 0
    while i < num_ints:
        i += 1
    return vals

Our own xrange

def xrange(num_ints):
    i = 0
    while i < num_ints:
        yield i
        i += 1

Generator expressions

Similarly to list comprehensions / iterator expressions, generators have a declarative form.

>>> is_even = lambda x: x % 2 == 0
>>> sum(x for x in xrange(10000) if is_even(x))
>>> (x for x in xrange(10000))
<generator object <genexpr> at ...>

Rewriting listcomps as genexps

>>> sum([x**2 for x in range(1000)])
>>> sum(x**2 for x in xrange(1000))

What's the difference between them?

Our listcomps allocate a new list and populate it with all the values specified in your listcomp expression.

Meanwhile, genexps create a generator object.

If output is only needed as an input for a function expecting a sequence, you should prefer genexps. They will keep memory stable.

If you need to mutate the list (delete or append elements), you need to stick with listcomps.

Most expressions of form fn([listcomp]) and be rewritten fn(genexp) safely.

EXERCISES: Generators and Generator Expressions

  • Start with a list of 100,000 numbers.

  • Set up a pipeline of generators that performs these transformations:
    • square each number
    • truncate to first 4 characters
    • convert to integer

Object-oriented vs. functional programming

Is Python a functional language?


Is Python an object-oriented language?


Is Python schizophrenic?


Python's style flexing

Perhaps programming paradigms aren't paradigms, but just states of mind.

If you are writing a math utility library, functions may be the best organizing principle.

If you are writing a database-oriented business application, perhaps classes and objects are appropriate.

If you are writing a framework that is meant to be both used and extended, maybe some combination of both is in order.

The standard library

The standard library shows the flexibility in action.

urllib is basically a set of utility functions for dealing with URLs.

But it can be extended by looking into urllib.FancyURLopener.

pprint and pformat are handy functions for pretty-printing.

But if you need more, your can extend pprint.PrettyPrinter.


import pdb; pdb.set_trace()

Probably the handiest single line of Python, ever.

Let's take a look at it interactively at the prompt.

Rich debugger

Tools like WingIDE provide a richer debugger.


Call stack

Debuggers allow you to navigate the callstack, which is useful for untangling complex, abstracted code.



>>> from os import path
>>> import os
>>> files = os.listdir(".")
>>> for fname in files:
...     print path.abspath(fname)

No matter how x-platform Python is, it seems we always end up having to deal with the quirks of our OS.


>>> import shutil
>>> shutil.copyfile('data.db', 'archive.db')
>>> shutil.move('build/executables', 'installdir')

EXERCISES: libraries, regex, and system

Class syntax

Just like modules, classes create a new namespace. But they also have special handling for class attributes.

Every class object is a function which, when called, creates an instance of the class. There is no new keyword.

>>> from swfmd import Person
>>> help(Person)
>>> person = Person()
>>> person.<TAB> # (to see methods)

World's simplest class

# class with no attributes
class Person(object): pass
person = Person()
# but attributes can be added
person.first_name = "John Doe"
person.age = 42

This is certainly a simple class, but doesn't do much for us.

Roughly interchangeable with using a dict, except don't need to use string literals to get access to named values.

A simple class for coffee

class CoffeeMaker(object):
    def __init__(self, num_cups, bean="arabica"):
        self.num_cups = num_cups
        self.bean = bean

    def make_coffee(self):
        fmt "made {num_cups} cups of coffee \
            using {bean} beans"
        print fmt.format(num_cups=self.num_cups, bean=self.bean)
>>> c = CoffeeMaker(2)
>>> c.make_coffee()
made 2 cups of coffee using arabica beans

Modeling a practical class

from datetime import datetime

class Employee(object):
    def __init__(self, id_, name, birth_year, role_id=None): = id_ = name
        self.birth_year = birth_year
        self.role_id = role_id
    def get_age(self):
        return - self.birth_year
    id2role = {
        "FTE": "Full-time Employee",
        "PT": "Part-time Employee",
        "INT": "Intern",
        "CNT": "Contractor"
    def get_role(self):
        if self.role_id is None:
            return None
        return self.id2role.get(self.role_id, "UNKNOWN")

Using the Employee class

>>> emp1 = Employee(1, "John Doe", 1975, role_id="FTE")
>>> emp2 = Employee(2, "Jane Doe", 1954, role_id="PT")
>>> emp3 = Employee(3, "Peter Travis", 1982)
>>> emp1.get_age()
>>> emp2.get_role()
"Part-time Employee"
>>> emp3.get_role() is None

Stateful methods

>>> employees = (emp1, emp2, emp3)
>>> age_sum = sum(employee.get_age() for employee in employees)
>>> avg_age = age_sum/len(employees)
>>> print avg_age

More advanced classes

import fetchers
import re
class Client(object):
    # base URL for Guardian open content API
    base_url = ''
    # Map paths (e.g. /content/search) to their corresponding methods:
    path_method_lookup = (
        ('^/search$', 'search'),
        ('^/tags$', 'tags'),
        ('^/item/(\d+)$', 'item'),
    def __init__(self, api_key, fetcher=None):
        self.api_key = api_key
        self.fetcher = fetcher or fetchers.best_fetcher()
    def search(self, query):

A real-world example.


Every function declared inside a class body is known as a method, which is automatically bound to an instance of the class at initialization time.

Inside a bound method, the first argument, typically named self, contains the instance of the object in question.

The pesky self convention

Guido van Rossum, Python's creator, has provided a slew of detail about his choice to require that all methods declared on a class receive a first argument that is the instance of the class.

Many programmers find this annoying, and wish they could work around it.

I'm one of those programmers.

But I'm here to tell you, in practice, you just get used to it and it ain't so bad.

Working around self

This seems clever:

class SuperObj(object):
    def __init__(self, **kwargs):
>>> s = SuperObj(name="Andrew", email="[email protected]")

But it's probably just a crutch.

Not just methods

Any non-function attributes are considered "class attributes", and are not treated in any special way. Note these are shared for all instances of the class, somewhat similarly to static properties in other languages.

Any attributes added to the instance directly using self.attr = val are considered "instance attributes".

Sometimes, you'll hear Pythonistas refer to the class dictionary and the instance dictionary. That's because, under the hood, Python classes are basically fancy dictionaries.

How does attribute lookup work?

Attribute lookup is implemented by a special internal method called __getattribute__, which checks the instance __dict__ first, then the class __dict__, and then walks up the class inheritance chain, until it finds a hit or raises AttributeError.

Magic can also happen due to a dunder method called __getattr__.

And further magic can happen due to something called descriptors.


At its core, a class is simply a class definition and typically, an initialization sequence implemented by __init__.

Instances of that class can act either as data containers or, more typically, bundles of state and behavior.


Whereas __init__ controls object initialization, __new__ runs earlier to control object creation.

It is much rarer to use new. One use-case is to cache class instances (re-use classes if they have been constructed from equal arguments and are stateless).

Another is to implement singletons, though typically this is a bad idea.

I tend to stay away from __new__ most of the time.

OO Design with Classes

Classes can let you design object-oriented interfaces that are intuitive for business people to grasp.


EXERCISES: Classes and introspection

An example interface: tokenizer

A tokenizer is an object that converts text strings into tokens using some sort of tokenization algorithm.

For example, a tokenizer might convert the sentence "The quick brown fox jumps over the lazy dog" into a sequence of tokens like ["the", "quick", "brown", ...]

Let's do it together

Let's do a live coding session where we build our own library of text tokenizers.

We'll then also compare them to tokenizers we can find on the web, and talk about filters and stemmers, as well.

EXERCISES: More Classes and interfaces

Your exercise is to define an interface for Tokenizer, and then implement a couple of tokenizers against that interface, such as a whitespace tokenizer, one that recognizes punctuation, etc.

Then, run each tokenizer on the same input using a snippet like this:

for tokenizer in tokenizers:

A data transformation problem

Here is some CSV data::


Problem specification

Imagine we need to do the following:
  • group by the "feature" column and sum those values, to generate a third column, "sum by feature"
  • for each row, calculate the percentage of the "sum by feature" of a given individual row

Data xform (1)

from operator import itemgetter
from collections import defaultdict
from pprint import pprint

class Entry(object):

with open("areadata.txt") as reader:
    lines = reader.readlines()
    lines = [line.strip().split(",") for line in lines]
    header, data = lines[0], lines[1:]

    def make_entry(row):
        """From a sequence with cols of data, makes an
        object (type Entry) whose attrs are the header
        labels and whose values are the column values."""
        entry = Entry()
        for i, val in enumerate(row):
            setattr(entry, header[i], val)
        return entry

Data xform (2)

# ... still under with keyword
    for line in lines:
        rows = []
        for row in data:
            # convert values in columns
            column_types = (str, float)
                            for i, val in enumerate(row)])
        # sort rows by first column
        rows = sorted(rows, key=itemgetter(0))
        # sums dictionary holds a mapping of feature => list of
        # area sums; if key is not present, creates an empty list
        sums = defaultdict(list)
        # group values with same feature
        for row in rows:
            entry = make_entry(row)

Data xform (3)

# ... still under with
    # ... still under for line in lines
        # now that all gathered together, let's sum them
        for key, val in sums.iteritems():
            sums[key] = sum(val)
        # re-iterate over the original list of rows to generate the
        # new derived columns, sum and percent
        for row in rows:
            fmt = "%.2f"
            entry = make_entry(row)
            sum_ = sums[entry.feature]
            percent = fmt % ((entry.area / sum_) * 100)
            row[1] = fmt % row[1]
            row[2] = fmt % row[2]
    # print the result in a pretty way

Deciding between a function and a class

You might wonder, with all of this flexibility...

How do I decide between a function and a class?

Some functionality is reserved for classes

For everything else, prefer functions

Only take on as much complexity as you actually need!

Exception heirarchy

The Python exception heirarchy is easy to grasp, and a good example of a class-based type heirarchy.


Exception handling in detail

Handling exceptions involves thinking about alternative control flows.


Solvable and Unsolvable Anagrams

>>> anagram(3, solvable=True)
>>> solve("dgo")
["god", "dog"]
>>> anagram(3, solvable=False)
>>> solve("xxx")

EXERCISES: Build the Anagram API

Your program must:
  • Generates solvable and unsolvable anagram puzzles.
  • Exposes this API as both a command-line interface and an interactive interface.
  • Is written as a Python module that can be imported by other code.
  • Is written in object oriented and functional styles so that components can be reused.

A better understanding of Classes

Python's object orientation capabilities are truly top-notch. We'll explore some ways how in this section.

Property getters

>>> class B(object):
...   def _get_special(self):
...       return 5
...   special = property(_get_special)
>>> b = B()
>>> b.special

Property setters

>>> class B(object):
...   def _get_special(self):
...      return 5
...   def _set_special(self, value):
...      print 'ignoring', value
...   special = property(_get_special, _set_special)
>>> b = B()
>>> b.special
>>> b.special = 10
ignoring 10
>>> b.special

Rewriting Employee with a property

from datetime import datetime

class Employee(object):
    def __init__(self, birth_year):
        self.birth_year = birth_year
    def _get_age(self):
        return - self.birth_year
    def _set_age(self, age):
        self.birth_year = - age
    age = property(_get_age, _set_age)

Using the new properties

>>> emp = Employee(1984)
>>> emp.age
>>> emp.age = 28
>>> emp.birth_year

Using the decorator version

from datetime import datetime

class Employee(object):
    def __init__(self, birth_year):
        self.birth_year = birth_year
    def age(self):
        return - self.birth_year
    def age(self, age):
        self.birth_year = - age

Descriptor objects

class Alias(object):
    def __init__(self, attr_name):
        self.attr_name = attr_name
    def __get__(self, obj, type=None):
        return getattr(obj, self.attr_name)
    def __set__(self, obj, value):
        return setattr(obj, self.attr_name, value)

Alias instances are descriptors, objects that can be used as a kind of "attribute guardian".

Using descriptors

class Person(object):
    def __init__(self, birth_year):
        self.birth_year = birth_year
        self.nicknames = []
    born_in = Alias("birth_year")
    nicks = Alias("nicknames")

These are used often in frameworks, e.g. Django's CharField and IntField types.

An aliased field without data duplication

>>> person = Person(1984)
>>> person.birth_year
>>> person.born_in
>>> person.born_in = 1983
>>> person.birth_year
>>> person.nicknames.extend(["Andy", "Drew"])
>>> person.nicks
["Andy", "Drew"]
>>> person.nicks is person.nicknames

Customizing __repr__ and __str__

class Person(object):
    def __init__(self, name, gender): = name
        self.gender = gender
    def __repr__(self):
        return "<%s name=%s, gender=%s>" % (
  , self.gender)
    char2gender = dict(M="Male", F="Female")
    def __str__(self):
        name =
        gender = self.char2gender.get(self.gender, None)
        if gender is None: return name
        return "%s (%s)" % name, gender

Using repr and str

>>> person = Person("John Doe", "M")
>>> person
<Person name=John Doe, gender=M>
>>> print person
John Doe (Male)
>>> str(person)
"John Doe (Male)"
>>> repr(person)
'<Person name=John Doe, gender=M>'
>>> person.gender = None
>>> print person
John Doe


Called unconditionally to implement attribute accesses for instances of the class.

If the class also defines __getattr__(), the latter will not be called unless __getattribute__() either calls it explicitly or raises an AttributeError.

This method should return the (computed) attribute value or raise an AttributeError exception.


>>> class A:
...    def __getattr__(self, name):
...        if name == 'special':
...           return 5
...        return self.__dict__[name]
>>> a = A()
>>> a.special

In old-style classes, this was the only way to control attribute access.

Nowadays, it's reserved only for "magic" objects, proxies, and frameworks.

EXERCISES: Build a proxy class

A proxy class is a class that can wrap any other object to add some functionality to it, but will forward all original requests onto that original object. For example:

>>> person = Person()
>>> tracer = Tracer(person)
>>> tracer.birth_year = 1984
[tracer] setting attribute "birth_year" => 1984
>>> tracer.compute_age()
[tracer] calling method "compute_age"

Implement Tracer.

EXERCISES: Build a type heirarchy


Build a set of classes that represents the data in this diagram.

More standard library

Let's take a quick whirlwind tour through some of my favorite stdlib modules.


>>> is_even = lambda x: x % 2 == 0
>>> items = sorted(range(1000), key=is_even)
>>> from itertools import groupby
>>> odd, even = groupby(items, is_even)
>>> odd[1].next()
>>> odd[1].next()


>>> import itertools
>>> list(itertools.product('ABC', '123'))
[('A', '1'), ('A', '2'), ('A', '3'),
('B', '1'), ('B', '2'), ('B', '3'),
('C', '1'), ('C', '2'), ('C', '3')]


>>> list(itertools.combinations('ABC', 2))
[('A', 'B'), ('A', 'C'), ('B', 'C')]


>>> list(itertools.chain(range(0, 3), range(10, 13)))
[0, 1, 2, 10, 11, 12]


>>> import random
>>> random.choice(['apple', 'pear', 'banana'])
>>> random.sample(xrange(100), 10)      # sampling without replacement
[30, 83, 16, 4, 8, 81, 41, 50, 18, 33]
>>> random.random()                     # random float
>>> random.randrange(6)                 # random integer chosen from range(6)


>>> from datetime import date
>>> now =
>>> now, 12, 2)
>>> now.strftime("%m-%d-%y. %d %b %Y is a %A on the %d day of %B.")
'12-02-03. 02 Dec 2003 is a Tuesday on the 02 day of December.'
>>> birthday = date(1964, 7, 31)
>>> age = now - birthday
>>> age.days


>>> import math
>>> math.cos(math.pi / 4.0)
>>> math.log(1024, 2)

pickle dump

import pickle

data1 = {'a': [1, 2.0, 3, 4+6],
        'b': ('string', u'Unicode string'),
        'c': None}
selfref_list = [1, 2, 3]

output = open('data.pkl', 'wb')

# Pickle dictionary using protocol 0.
pickle.dump(data1, output)

# Pickle the list using the highest protocol available.
pickle.dump(selfref_list, output, -1)


pickle load

import pprint, pickle

pkl_file = open('data.pkl', 'rb')

data1 = pickle.load(pkl_file)

data2 = pickle.load(pkl_file)



>>> import json
>>> json.dumps(['foo', {'bar': ('baz', None, 1.0, 2)}])
'["foo", {"bar": ["baz", null, 1.0, 2]}]'
>>> print json.dumps({'4': 5, '6': 7}, sort_keys=True, indent=4)
    "4": 5,
    "6": 7
>>> json.loads('["foo", {"bar": ["baz", null, 1.0, 2]}]')
[u'foo', {u'bar': [u'baz', None, 1.0, 2]}]


>>> from collections import defaultdict
>>> s = 'mississippi'
>>> d = defaultdict(int)
>>> for k in s:
        d[k] += 1
>>> d.items()
[('i', 4), ('p', 2), ('s', 4), ('m', 1)]

urllib / urllib2

import urllib
import urllib2
url = ''
values = {'name' : 'Andrew M',
        'location' : 'Tampa, FL',
        'language' : 'Python' }
data = urllib.urlencode(values)
req = urllib2.Request(url, data)
response = urllib2.urlopen(req)
page =


def average(values):
    """Computes the arithmetic mean of a list of numbers.

    >>> print average([20, 30, 70])
    return sum(values, 0.0) / len(values)

import doctest
doctest.testmod()   # automatically validate the embedded tests

sqlalchemy (1)

from sqlalchemy import *

db = create_engine('sqlite:///tutorial.db')

db.echo = True

metadata = MetaData(db)

places = Table('places', metadata,
    Column('id', Integer, primary_key=True),
    Column('lat', Float),
    Column('lng', Float),
    Column('desc', String(50))

sqlalchemy (2)

insert = places.insert()
insert.execute(desc='Wine Toad', lat=39.545, lng=42.4443)
        {'desc': 'SWFMD', 'lat': 38.555, 'lng': 43.55},
        {'desc': 'Hampton Inn', 'lat': 34.555, 'lng': 42.55})

select =
resultset = select.execute()

sqlalchemy (3)

row = resultset.fetchone()
print 'Id:', row[0]
print 'Desc:', row['desc']
print 'Lat:',
print 'Lng:', row[places.c.lng]

for row in resultset:
    print row.desc, 'is @',, row.lng, 'lat/lng'

sqlalchemy ORM (1)

from sqlalchemy import *
from datetime import datetime
db = create_engine("sqlite:///test.db")
db.echo = True
metadata = MetaData(db)

user_table = Table('acct_users', metadata,
                Column('id', Integer, primary_key=True),
                Column('user_name', Unicode(16),
                        unique=True, nullable=False),
                Column('password', Unicode(40), nullable=False),
                Column('display_name', Unicode(255), default=''),
                Column('created', DateTime,


sqlalchemy ORM (2)

from sqlalchemy.orm import *

class User(object):
    def get_dos_password(self):
    def days_since_created(self):
        return ( - self.created).days
mapper(User, user_table)
Session = sessionmaker()
session = Session()

# empty right now
results = session.query(User)

user = User()
user.user_name = "amontalenti"
user.password = "not a chance"

EXERCISES: Use the standard library

Build a module that makes an HTTP request to Wikipedia for any given article page, e.g. Barack Obama. What it returns back is the number of occurrences of a string in the article page.

Some requirements:

  • Use urllib and urllib2 to actually perform the HTTP request.
  • Use a defaultdict to count the number of occurrences of the search term.
  • Use the re module to search for the string in each line of the response page.
  • Use doctest to make sure everything is working properly.

EXERCISES: Compile this presentation!

  • Full source code for this presentation is available
  • You can compile it and refer back to it at any time

Baby Turtles

Use your powers wisely, and always remember...


Magic Turtles!

It's turtles all the way down!


Other great presentations about Python



Your host:

Andrew Montalenti,