Python Patterns: Named Tuples

A drawing of a Lycodon Modestus snake from the Iconographic Zoologica collection.

In Python, sequences are a great way to hold a set of ordered data. As long as the data is simple, a list or tuple is perfect because they are included in every install of Python. But data is not always simple; you can put any object you want in a sequence, making it easy to lose track of what is where.

For example, one might create cards in a virtual address book like this:

card = (
  "Alex",
  "Gude",
  "me@alexgude.com",
  None,
  "17 St., Smaller Town, CA",
)

Simple, but a little confusing. What does None signify? Writing code to work with these objects is error prone:

def check_email(card):
  """Check if a card has an email
  address that is valid."""
  email = card[2]  # 2?!
  is_valid = email is not None and '@' in email

  return is_valid

Is 2 the correct index to use? Maybe it was 3? Catching mistakes in the code is tough for anyone reading it.

Alternatives

A dictionary is a natural solution to this problem, because we can use strings as keys, for example card["email"] instead of card[2]. But we might need to maintain compatibility with something that expects a sequence, as was the case when passing artists around in my matplotlib blitting post.

Instead, we could build a class that acts like a list or tuple::

class Card:
  def __init__(self, first_name, last_name, ...):
    self.__internal = [first_name, last_name, ...]
    self.first_name = self.__internal[0]
    self.last_name = self.__internal[1]
    ...  # etc.

  def __len__(self):
    return self.__internal.__len__()

  def __getitem__(self, key):
    return self.__internal.__getitem__(key)

  def __next__(self):
    return self.__internal.__next__()

  # and many other methods

Not difficult to write, but tedious due to all of the boilerplate code. Thankfully, someone has already done so.

Named Tuples

The named tuple functions exactly like a tuple, with one addition: you can access each component of the tuple by name. Our card example would now looks like this:

from collections import namedtuple

Card = namedtuple(
    "Card",
    [
        "first_name",
        "last_name",
        "email",
        "phone",  # Our empty field revealed!
        "address",
    ]

)

alex_card = Card(
    "Alex", "Gude", "me@alexgude.com",
    None, "17 St., Smaller Town, CA",
)

This is much cleaner than our original card tuple. We now know the missing value is the phone number! We can access the values with dot operators as well: card.email. And the named tuple stills works exactly as you would expect for a standard tuple:

# For loops work
for item in alex_card:
    print(item)

# We can access with . or []
alex_card[2] == alex_card.email

# And we can unpack
first, last, email, phone, address = alex_tuple

Code that operates on this named tuple is much easier to read as well, because it does not rely on magic numbers:

def new_check_email(card):
  """Check if a card has an email
  address that is valid."""
  email = card.email
  is_valid = email is not None and '@' in email

  return is_valid

Named tuples are not as well known as dictionaries or classes, but they solve a common problem and make your code more readable!