Why does `np.random.default_rng()` prevent seed change compared to `numpy.random.seed`?

Question

I recently learned about the downsides of using numpy.random.seed(seed). Some internal file can modify the seed.I read an online article that suggested I use np.random.default_rng(). But I am not sure how that works.

I came across this related SO post but it did not help my understanding.

Answer 1

np.random.seed() sets a global seed for NumPy’s legacy random number generator. This global state can be unintentionally modified elsewhere in your code, leading to unpredictable results.

In contrast, np.random.default_rng() creates a new, independent instance of NumPy’s modern Generator (based on the PCG64 algorithm). Each instance maintains its own state and does not affect or rely on any global seed.

Key difference:
You can’t “reseed” the global RNG with default_rng(), it intentionally avoids global state for better reproducibility and isolation.

To get repeatable results, pass a seed when creating the generator:

rng = np.random.default_rng(42)
value = rng.random()

This is the recommended approach in modern NumPy (version 1.17 or later).