Custom sorting

Sort dicts by multiple keys, use key= with tuple tie-breaking, and apply functools.cmp_to_key for a custom comparator.

Two techniques for when simple key=len or key=itemgetter('age') is not enough: tuple keys for multi-field sorting, and cmp_to_key for arbitrary custom comparators.

Multi-key sort and custom comparator

Python — editable, runs in your browser

from functools import cmp_to_key

# ── Multi-key sort ────────────────────────────────────────────────────────────

students = [
  {'name': 'Diana', 'grade': 85},
  {'name': 'Alice', 'grade': 90},
  {'name': 'Bob',   'grade': 85},
  {'name': 'Carol', 'grade': 90},
  {'name': 'Eve',   'grade': 85},
]

# Sort: grade descending, name ascending on ties.
# Negate grade so descending order works with the default ascending sort.
sorted_students = sorted(students, key=lambda s: (-s['grade'], s['name']))

print("=== Students sorted by grade desc, name asc ===")
for s in sorted_students:
  print(f"  grade={s['grade']}  name={s['name']}")

# ── functools.cmp_to_key ─────────────────────────────────────────────────────

def largest_number(nums):
  """Arrange nums so that concatenating them forms the largest possible number."""
  def compare(a, b):
      # Compare 'ab' vs 'ba' as strings.
      # Return negative if a should come BEFORE b (a wins).
      if a + b > b + a:
          return -1
      elif a + b < b + a:
          return 1
      return 0

strs = [str(n) for n in nums]
  strs.sort(key=cmp_to_key(compare))
  result = ''.join(strs)
  # Edge case: all zeros
  return '0' if result[0] == '0' else result

print()
print("=== Largest number from digits ===")
test_cases = [
  [10, 2],
  [3, 30, 34, 5, 9],
  [0, 0],
  [1],
  [824, 938, 1399, 5607, 6973, 5703, 9609, 4398, 8247],
]
for nums in test_cases:
  print(f"  {nums}  =>  {largest_number(nums)}")

Why tuple keys work for multi-field sorts

Python compares tuples element by element: first by the first element, then (only if equal) by the second, and so on. (-grade, name) sorts by grade descending first (negating reverses order for integers) and name ascending second. One pass, one key function, stable result.

This is usually cleaner than two-pass sorting. Use two-pass only when one key involves a comparison that cannot be cleanly expressed as a tuple — for instance, case-insensitive descending, or mixed ascending/descending on a non-numeric field.

When to use cmp_to_key

cmp_to_key is for situations where you have a comparison function — a function cmp(a, b) that returns negative/zero/positive — and need to convert it to a key= function. The comparison function is given two actual elements and decides their relative order directly.

The largest_number problem is the classic example. The natural sort order of integers does not work: 9 should come before 34, not after. The custom rule is: a should come before b if the string ab is lexicographically greater than ba. This rule cannot be expressed as a simple per-element key function because it depends on both elements — hence cmp_to_key.

cmp_to_key is slower than a key= function because it must call the comparator O(n log n) times during the sort. For large lists, try to express the ordering as a simple key first. Use cmp_to_key only when the ordering is inherently pairwise — like the concatenation comparison above.

Where to go next

Next: binary search variations — bisect_left, bisect_right, and the unifying idea of binary search over any monotone function.

Finished reading? Mark it complete to track your progress.

Multi-key sort and custom comparator

Why tuple keys work for multi-field sorts

When to use cmp_to_key

Where to go next

On this page