Sorting Arrays

"Show me Virat's innings from lowest to highest." "List every score of the season, best first." Sorting is one of the first things an analyst does with data, and NumPy sorts arrays in fast compiled code — far faster than sorting a Python list for large data.

Definition

Sorting rearranges the elements of an array into ascending order (smallest first). np.sort(a) returns a sorted copy and leaves a unchanged; a.sort() sorts the array in place (changes a itself) and returns None. For multi-dimensional arrays, both sort independently along one axis (the last axis by default).

• Ascending only — NumPy has no reverse=True option. Descending order is done by reversing the result ([::-1]) or sorting the negative values.
• Copy vs in place — the same difference as Python's sorted(lst) vs lst.sort().
• Along one axis — each row (or each column) is sorted on its own; rows are not kept together. To reorder whole rows by one column, use argsort (next topic).

Syntax

np.sort(a, axis=-1, kind=None, order=None, *, stable=None)   # returns a new array
a.sort(axis=-1, kind=None, order=None, *, stable=None)       # in place, returns None

Example — np.sort vs ndarray.sort

import numpy as np

runs = np.array([72, 45, 88, 31, 64, 0, 103])
s = np.sort(runs)
print(s)
print(runs)

r = runs.copy()
out = r.sort()
print(r)
print(out)

[  0  31  45  64  72  88 103]
[ 72  45  88  31  64   0 103]
[  0  31  45  64  72  88 103]
None

How it works: np.sort gave a new sorted array and runs kept its original match order. r.sort() rearranged r itself and returned None. Use np.sort when the original order still matters (here it does — the order is the match sequence!).

Example — descending order

print(np.sort(runs)[::-1])
print(-np.sort(-runs))

[103  88  72  64  45  31   0]
[103  88  72  64  45  31   0]

How it works: [::-1] reverses the ascending result (a cheap view). -np.sort(-runs) sorts the negatives ascending, then flips the sign back — handy for numbers, but it does not work for strings.

Example — sorting a 2D array along an axis

squad = np.array([[72, 45, 88, 31],    # Virat
                  [56,  0, 74, 88],    # Rohit
                  [45, 32,  0, 67]])   # Hardik
print(np.sort(squad))            # axis=-1: each ROW sorted
print(np.sort(squad, axis=0))    # each COLUMN sorted
print(np.sort(squad, axis=None)) # everything, flattened

[[31 45 72 88]
 [ 0 56 74 88]
 [ 0 32 45 67]]
[[45  0  0 31]
 [56 32 74 67]
 [72 45 88 88]]
[ 0  0 31 32 45 45 56 67 72 74 88 88]

How it works:

axis=1 / -1  ->  sort along each row   (each player's scores, low to high)
axis=0       ->  sort down each column (each match's scores, low to high)
axis=None    ->  flatten, then sort    (every score of the season)

Notice that with axis=0 the rows are mixed up: the first row [45 0 0 31] is not any real player — it is the lowest score of each match. Sorting along an axis sorts every row/column independently.

For each player's scores best first, sort the rows then reverse the columns:

print(np.sort(squad, axis=1)[:, ::-1])

[[88 72 45 31]
 [88 74 56  0]
 [67 45 32  0]]

Sorting strings and NaN

names = np.array(["Rohit", "virat", "Bumrah", "Dhoni", "Gill"])
print(np.sort(names))
print(np.sort(np.array([3.5, np.nan, 1.0, 2.0])))

['Bumrah' 'Dhoni' 'Gill' 'Rohit' 'virat']
[1.  2.  3.5 nan]

Strings sort alphabetically by character code, so uppercase comes before lowercase ('virat' lands last). NaN values are always placed at the end.

Python list vs NumPy

Common mistakes

• ❌ sorted_runs = runs.sort() → sorted_runs is None. ✅ sorted_runs = np.sort(runs).
• ❌ np.sort(runs, reverse=True) → TypeError. ✅ np.sort(runs)[::-1].
• ❌ Sorting a 2D table with axis=0 and expecting rows to stay together. ✅ Each column is sorted separately; use argsort to reorder whole rows.
• ❌ arr.sort() on the only copy of match-ordered data. ✅ Use np.sort (or .copy() first) so the original order survives.

Interview tip

Q: What is the difference between np.sort(a) and a.sort()? A: np.sort(a) returns a new sorted copy and leaves a unchanged. a.sort() sorts a in place, uses no extra array, and returns None. np.sort also accepts axis=None to flatten first.