NumPy · Random numbers

The NumPy docs recommend the Generator API: rng = np.random.default_rng(seed). The older np.random.seed() + np.random.rand() style relies on global state and is kept for backward compatibility (NEP 19 explains why).

• The same seed gives the same sequence, which is essential for reproducible ML experiments.
• rng.integers, rng.random, rng.normal(mu, sigma, size), rng.choice(a, size, replace=False), rng.permutation.
• To split a dataset, shuffle the indices and slice them.

import numpy as np

rng = np.random.default_rng(42)
print(rng.integers(1, 7, size=10))             # dice rolls
print(rng.random(3).round(4))

samples = rng.normal(loc=170, scale=8, size=100_000)
print(round(samples.mean(), 2), round(samples.std(), 2))

rng_a = np.random.default_rng(7)
rng_b = np.random.default_rng(7)
print(np.array_equal(rng_a.random(5), rng_b.random(5)))

idx = rng.permutation(10)
train, test = idx[:8], idx[8:]
print("train", np.sort(train), "test", np.sort(test))
print(rng.choice(["python", "c++", "numpy", "pandas"], size=2, replace=False))

Output:

[1 5 4 3 3 6 1 5 2 1]
[0.9756 0.7611 0.7861]
169.97 8.03
True
train [2 3 4 5 6 7 8 9] test [0 1]
['pandas' 'c++']

That finishes NumPy. Pandas is built on top of it, so that’s next.