Python's lambda functions

Have you ever seen the word lambda used in Python?

sorted_by_values = sorted(items, key=lambda i: i[1])

That's called a "lambda expression" and it defines a "lambda function".

What are lambda expressions and lambda functions? And how are lambda functions different from other functions in Python?

Lambda expressions define functions

A lambda expression looks like this:

>>> square = lambda n: n**2

This square variable points to some object now. What do you think its type might be?

>>> type(square)
<class 'function'>

It's a function!

We could call this function just like any other function in Python:

>>> square(3)
9

A lambda expression is a way of making a function. And a lambda function is the object we get back from a lambda expression:

>>> square
<function <lambda> at 0x7f7f221eaca0>

But wait... don't we already have a syntax for making functions in Python?

We do and you've probably seen it many times before:

def square(n):
    return n**2

We can use this def syntax for defining a new function.

So why do lambda expressions exist?

Lambda expressions can be defined on the same line they're used

Let's say we'd like to sort the items in a dictionary by their values.

Python's built-in sorted function will sort iterable items in ascending order (using the < operator):

>>> fruits = ['lemon', 'apple', 'lime', 'pear', 'watermelon', 'banana']
>>> sorted(fruits)
['apple', 'banana', 'lemon', 'lime', 'pear', 'watermelon']

What if we wanted items by something other than their value? For example, what if we wanted to sort strings by their lengths?

The sorted function accepts an optional key argument for this:

>>> fruits = ['lemon', 'apple', 'lime', 'pear', 'watermelon', 'banana']
>>> sorted(fruits, key=len)
['lime', 'pear', 'lemon', 'apple', 'banana', 'watermelon']

The key argument can be any function (technically any callable will do). The sorted function will call that function on each item in the given iterable, sorting each item by the result of those function calls.

What if you don't already have a function to pass to sorted? For example, let's say we wanted to sort dictionary items by their values:

>>> counts = {"apple": 3, "pear": 2, "banana": 4, "lime": 1}
>>> counts.items()
dict_items([('apple', 3), ('pear', 2), ('banana', 4), ('lime', 1)])

We could define a function that accepts each item and returns the value of that item:

def by_value(item):
    """Return the value from a given (key, value) tuple."""
    key, value = item
    return value

Then we can psas that function to sorted:

>>> sorted(counts.items(), key=by_value)
[('lime', 1), ('pear', 2), ('apple', 3), ('banana', 4)]

Note that we needed to define the function before we called sorted with it:

>>> def by_value(item):
...     """Return the value from a given (key, value) tuple."""
...     key, value = item
...     return value
...
>>> sorted(counts.items(), key=by_value)
[('lime', 1), ('pear', 2), ('apple', 3), ('banana', 4)]

Instead of using def to define our function, we could have used a lambda function:

>>> sorted(counts.items(), key=lambda item: item[1])
[('lime', 1), ('pear', 2), ('apple', 3), ('banana', 4)]

Unlike the def syntax for defining a function, you can pass a lambda function into another function on the same line of code that you define the lambda function. So lambda expressions allows us to write shorter code by defining a function on the same line that we pass that function to sorted.

The limitations of lambda expressions

Lambda functions are anonymous functions, meaning they have no name.

If we ask for help on a lambda function, it'll say its name is <lambda>:

>>> square = lambda n: n**2
>>> help(square)
Help on function <lambda> in module __main__:

<lambda> lambda n

Lambda functions are also limited to a single Python expression. You can't put multiple lines of code within a lambda expression.

Due to their single expression limitation, lambda expressions have no need for return statements (and don't allow them). The result of calling the single expression within a lambda expression will be the return value of the lambda function.

So lambda functions can only execute a single Python expression and their return value will be the result of that single expression.

So how do lambda functions compare to traditional functions, in terms of costs and benefits? Well, it depends on how you categorize differences.

If I had to classify the pros and cons of using lambda over def, I'd say:

• Pro: Can be immediately passed around (no variable needed)
• Pro: Return their sole expression automatically
• Con/Pro: Can only have a single expression within them
• Con: Don't have a name and can't have a docstring
• Con: Have a very different syntax from the usual def syntax

When should you avoid lambda expressions?

PEP 8, the official Python styleguide, says that code like this:

square = lambda n: n**2

Should always be replaced by code like this:

def square(n): return n**2

Or if you prefer, a multi-line function definition with a nice docstring like this:

def square(n):
    """Return the square of the given number."""
    return n**2

I would take this advice a step further by arguing that lambda expressions should usually be avoided. Giving a function a name often makes its use easier to understand and many lambda functions can be replaced by (fairly well-named) functions that already exist. See Overusing lambda expressions in Python for my opinions on avoiding lambda expressions.

Where is lambda often used?

Lambda expressions are often very handy when using functions or classes that involve passing functions around.

For example, you'll often see lambda expressions used with the built-in sorted function:

>>> items = ["2 tomatoes", "1 avocado", "6 bananas", "4 clementines"]
>>> sorted(items, key=lambda item: int(item.split()[0]))
['1 avocado', '2 tomatoes', '4 clementines', '6 bananas']

Lambda expressions are also often used with the map and filter built-in functions:

>>> numbers = [2, 1, 3, 4, 7, 11, 18, 29]
>>> list(filter(lambda n: n % 2 == 1))
[1, 3, 7, 11, 29]
>>> list(map(lambda n: n**2, numbers))
[4, 1, 9, 16, 49, 121, 324, 841]

Though I usually recommend avoiding Python's map and filter functions.

Pass functions around, but use lambda conservatively

Python's functions can be passed around because functions are objects in Python. If you want to define a function on the same line of code that you pass that function off to another function, you can use a lambda expression.

>>> sorted(items, key=lambda item: int(item.split()[0]))
['1 avocado', '2 tomatoes', '4 clementines', '6 bananas']

Though personally, I always like to ask myself whether the code I'm writing might be clearer if I gave my function a name instead.

>>> def by_quantity(item):
...     quantity, name = item.split()
...     return int(quantity)
...
>>> sorted(items, key=by_quantity)
['1 avocado', '2 tomatoes', '4 clementines', '6 bananas']

It's up to you when and where you use lambda expressions, but keep in mind that lambda expressions are just a special syntax for making a function that doesn't have a name. Anything you can do with a lambda function, you can do with a non-lambda function.