NumPy Random Permutation

Reordering Data Randomly

Many tasks need data in a random order: shuffling rows before splitting a dataset into training and test sets, randomizing the order of quiz questions, or dealing a shuffled deck of cards. NumPy provides two functions for this - np.random.shuffle and np.random.permutation - and they are easy to mix up because both scramble a sequence, but they behave very differently.

import numpy as np

np.random.seed(1)
original = np.array([10, 20, 30, 40, 50])

shuffled_copy = np.random.permutation(original)

print("Original:", original)
print("Shuffled copy:", shuffled_copy)
# 'original' is untouched - permutation returned a brand new array

np.random.shuffle: In-Place Reordering

np.random.shuffle rearranges the array it is given directly, in memory, and returns None. That means you never assign its result to a variable - you call it on the array and then read the same array afterward. For a multi-dimensional array, shuffle only reorders along the first axis, so a 2D array has its rows shuffled, not the values inside each row.

import numpy as np

np.random.seed(2)
deck = np.array([1, 2, 3, 4, 5])

result = np.random.shuffle(deck)
print("Return value:", result)   # None
print("deck after shuffle:", deck)  # deck itself has changed

# Shuffling a 2D array reorders whole rows, not values inside a row
grid = np.array([[1, 2], [3, 4], [5, 6]])
np.random.shuffle(grid)
print(grid)

np.random.permutation: Returns a New Array

np.random.permutation leaves its input untouched and instead returns a new, shuffled array, which is safer when you still need the original order elsewhere in your program. It also accepts a single integer n instead of an array - in that case it returns a shuffled arrangement of range(n), which is a common way to generate a random ordering of indices without shuffling any real data yet.

• shuffle changes its argument in place; permutation returns a new array and leaves the input alone.
• shuffle always returns None; permutation always returns the shuffled result.
• shuffle only accepts an existing array or list; permutation also accepts a single integer n.
• For a 2D array, both functions reorder rows (the first axis), never the values within a row.

import numpy as np

np.random.seed(3)
features = np.array([[1, 2], [3, 4], [5, 6], [7, 8]])
labels = np.array([0, 1, 0, 1])

# Shuffle indices, not the arrays themselves, to keep them aligned
order = np.random.permutation(len(features))
print("Shuffled order:", order)

shuffled_features = features[order]
shuffled_labels = labels[order]
print(shuffled_features)
print(shuffled_labels)