Random Sampling with NumPy's choice Function

Syntax

numpy.random.choice(a, size=None, replace=True, p=None)

Parameter	Description
`a`	Aray-like object or enteger. If an integer, samples from `range(a)`. If array-like, samples from the elements directly.
`size`	Output shape. Integer or tuple of integesr. Returns a single element when `None` (default).
`replace`	Boolean flag. When `True`, each element can be selected multiple times. When `False`, sampling without replacement ensures unique elements.
`p`	Probability distribution array. Must have the same length as `a`, with values summing to 1. Each element's selection probability corresponds to its index.

import numpy as np

colors = ['red', 'green', 'blue']
selected = np.random.choice(colors)
print(selected)

Output:

blue

numbers = [10, 20, 30, 40, 50]
samples = np.random.choice(numbers, size=4)
print(samples)

Output:

[20 30 20 50]

dataset = ['alpha', 'beta', 'gamma', 'delta', 'epsilon']
subset = np.random.choice(dataset, size=3, replace=False)
print(subset)

Output:

['beta' 'epsilon' 'alpha']

outcomes = ['low', 'medium', 'high']
weights = [0.6, 0.3, 0.1]
result = np.random.choice(outcomes, p=weights, size=5)
print(result)

Output:

['low' 'low' 'medium' 'low' 'low']

indices = np.random.choice(10, size=5, replace=False)
print(indices)

Output:

[7 2 9 0 4]

When replace=False, specifying size greater than the input length raises a ValueError.
The probability array p must satisfy two constraints: all values must be non-negative, and the total must equal 1.0.
For reproducible results across runs, initialize the random seed:

np.random.seed(42)
result = np.random.choice(['a', 'b', 'c'], size=2)
print(result)

When p is not provided, the function assumes a uniform distribution where each element has equal selection probability.
The function internally leverages the same PRNG engine as other NumPy random functions, making it suitable for integration with np.random.shuffle and similar operations.

Posted on Sat, 15 Aug 2026 16:25:59 +0000 by Izzy1979