The np.random.uniform() function draws random numbers from a continuous uniform distribution.

import numpy as np

np.random.uniform()

# Expected result like...
# 0.20156508227392989

Basic usage

By default, the range is [0, 1) and the function returns a scalar. But this can be adjusted with the low, high and size arguments. For example, to generate 3 random numbers between -10 and 10:

np.random.uniform(-10, 10, size=3)

# Expected result like...
# array([ 6.64893079, -9.70234671,  4.01869664])

As a convention, I usually pass in low and high as positional arguments and set size as a keyword argument, but this is just my own style preference.

Multi-dimensional arrays

If you want a random array with shape (2, 2), you can pass a tuple in for size:

np.random.uniform(-10, 10, size=(2, 2))

# Expected result like...
# array([[-4.22106242, -7.85748055],
#        [ 1.32769555, -3.05958937]])

Data types

Although many NumPy functions accept a dtype argument, np.random.uniform() will always return np.float64 values, either as a single scalar or as an np.ndarray. But if you want a different data type, you can use the astype() method on the result:

np.random.uniform(size=2).astype("float32")

# Expected result like...
# array([0.02816657, 0.7264831 ], dtype=float32)

Seeds

If you want to set the seed for the random number generator, you can use np.random.seed():

np.random.seed(10)
np.random.uniform()

# Expected result (every time)
# 0.771320643266746

This is an important strategy for testing non-deterministic code.

As a final note, the official NumPy docs now suggest using a default_rng() random number generator instead of np.random.uniform(). I don’t think this is super common yet, but it is something to watch out for.