Key Functions, lambda, itemgetter and attrgetter

18 min · Intermediate

Real data is rarely a plain list of numbers. Orders are tuples, employees are dictionaries or objects. The key argument tells Python which part of each record to sort by.

A key function decides the order

Key function is a function you pass as key=. Python calls it once per item and sorts by the values it returns, while the items in the result stay unchanged.

Product names sort with capitals first, which looks wrong to a customer. A key that lowercases each name fixes the order.

Python

products = ["keyboard", "Monitor", "cable", "Dock"]print(sorted(products))print(sorted(products, key=str.lower))

Sorting with key=str.lower. Each item gets a key once: keyboard becomes keyboard, Monitor becomes monitor, cable stays cable and Dock becomes dock. The items are then ordered by their keys, giving cable, Dock, keyboard, Monitor, and the original items are returned, not the keys.

Example 1 · Sort by distance from zero

Python

adjustments = [-120, 45, -30, 200]print(sorted(adjustments, key=abs))

Output

[-30, 45, -120, 200]

What this code does

abs is a built-in function, so it can be a key directly.
The values are compared as 30, 45, 120, 200, but the original values are returned.

str.lower has no brackets. You pass the function itself, and Python calls it once for each item.

lambda for a one-off key

lambda is a one-line anonymous function written as lambda arguments: expression. It returns the value of the expression. Use it for short keys that are needed in one place.

Orders are tuples of id and amount. To sort by amount you need the second item. A lambda writes that rule in one line.

Python

orders = [("A-1", 250), ("A-2", 40), ("A-3", 900)]print(sorted(orders, key=lambda order: order[1]))

Example 2 · Sort invoice names by the number in them

Python

files = ["inv-10", "inv-2", "inv-33"]print(sorted(files))print(sorted(files, key=lambda name: int(name.split("-")[1])))

Output

['inv-10', 'inv-2', 'inv-33']
['inv-2', 'inv-10', 'inv-33']

What this code does

Plain text sorting puts inv-10 before inv-2, because the character 1 comes before 2.
The lambda cuts out the number and turns it into an int, so 2 sorts before 10.

lambda order: order[1] is a small function that takes one order and returns its amount.

itemgetter for tuples, lists and dictionaries

operator.itemgetter builds a function that returns one item, or several, from whatever you give it. It works with tuples, lists and dictionaries and replaces lambda x: x[i].

Rows loaded from a file are often dictionaries. itemgetter builds the key function for you and reads clearly.

Python

1from operator import itemgetter2
3rows = [4    {"id": "E1", "dept": "Sales", "pay": 52000},5    {"id": "E2", "dept": "IT", "pay": 61000},6    {"id": "E3", "dept": "Sales", "pay": 47000},7]8print([row["id"] for row in sorted(rows, key=itemgetter("pay"))])

Example 3 · itemgetter with two positions

Python

from operator import itemgetter
students = [("Asha", "B", 88), ("Ravi", "A", 72), ("Meena", "B", 95), ("Diya", "A", 72)]ranked = sorted(students, key=itemgetter(1, 2))print([name for name, *_ in ranked])

Output

['Ravi', 'Diya', 'Asha', 'Meena']

What this code does

itemgetter(1, 2) returns the tuple (class, marks) for each student.
Ravi and Diya tie on both, so stability keeps Ravi first.

itemgetter("pay") does the same as lambda row: row["pay"] and is a little faster.

attrgetter for objects

operator.attrgetter builds a function that reads one or more attributes from an object. It is the object version of itemgetter and replaces lambda e: e.name.

Employees stored as objects have attributes, not keys. attrgetter reads an attribute by name.

Python

1from dataclasses import dataclass2from operator import attrgetter3
4
5@dataclass6class Employee:7    name: str8    dept: str9    pay: int10
11
12staff = [Employee("Asha", "Sales", 52000), Employee("Ravi", "IT", 61000), Employee("Meena", "Sales", 47000)]13print([e.name for e in sorted(staff, key=attrgetter("pay"))])

attrgetter("pay") is the same as lambda e: e.pay. It also accepts several names.

Sorting by several fields

A payroll report lists departments in order and, inside each department, the highest pay first. Return a tuple from the key.

Python

print([e.name for e in sorted(staff, key=lambda e: (e.dept, -e.pay))])

Example 4 · Department ascending, pay descending, by two passes

Python

1from operator import itemgetter2
3staff = [("Asha", "Sales", 52000), ("Ravi", "IT", 61000), ("Meena", "Sales", 47000), ("Kiran", "IT", 58000), ("Diya", "Sales", 52000)]4step1 = sorted(staff, key=itemgetter(2), reverse=True)5step2 = sorted(step1, key=itemgetter(1))6print([name for name, *_ in step2])

Output

['Ravi', 'Kiran', 'Asha', 'Diya', 'Meena']

What this code does

Sort by the less important field first, then by the more important one.
The second sort leaves ties alone, so pay order survives inside each department. Asha and Diya tie and keep arrival order.

Tuples compare the first item, then the second on a tie. The minus sign flips numbers only. It cannot flip text.

Next step

Check your understanding

12 quick questions

1from operator import itemgetter2 3rows = [4 {"id": "E1", "dept": "Sales", "pay": 52000},5 {"id": "E2", "dept": "IT", "pay": 61000},6 {"id": "E3", "dept": "Sales", "pay": 47000},7]8print([row["id"] for row in sorted(rows, key=itemgetter("pay"))])

from operator import itemgetter students = [("Asha", "B", 88), ("Ravi", "A", 72), ("Meena", "B", 95), ("Diya", "A", 72)]ranked = sorted(students, key=itemgetter(1, 2))print([name for name, *_ in ranked])

1from dataclasses import dataclass2from operator import attrgetter3 4 5@dataclass6class Employee:7 name: str8 dept: str9 pay: int10 11 12staff = [Employee("Asha", "Sales", 52000), Employee("Ravi", "IT", 61000), Employee("Meena", "Sales", 47000)]13print([e.name for e in sorted(staff, key=attrgetter("pay"))])

1from operator import itemgetter2 3staff = [("Asha", "Sales", 52000), ("Ravi", "IT", 61000), ("Meena", "Sales", 47000), ("Kiran", "IT", 58000), ("Diya", "Sales", 52000)]4step1 = sorted(staff, key=itemgetter(2), reverse=True)5step2 = sorted(step1, key=itemgetter(1))6print([name for name, *_ in step2])