How to generate a random float in range [-1,1] using numpy.random.rand()?

Question

Numpy's random.rand() function produces random floats between 0 and 1. How do I transform the output of this function to produce random floats between -1 and 1?

Current output:

In[]: numpy.random.rand(3, 2)
Out[]: 
array([[0.13568674, 0.72465483],
       [0.48726461, 0.68867378],
       [0.50463821, 0.28619853]])

Desired (example) output:

In[]: numpy.random.rand(3, 2)
Out[]: 
array([[-0.13568674, 0.72465483],
       [-0.48726461, 0.68867378],
       [0.50463821, -0.28619853]])

I would not like to use random.uniform().

What's the reason not to use np.random.uniform? That's an odd restriction; do you have a good reason for avoiding it, and are there other, similar restrictions? You seem be operating under an unusual set of constraints, and it would be useful to know what those are. — Mark Dickinson
– Mark Dickinson, Commented Dec 18, 2019 at 11:18

norok2 · Accepted Answer · 2019-12-21 08:04:10Z

3

You basically need to scale and shift the (0, 1) range normally generated by np.random.rand() to the (-1, 1) target range:

import numpy as np

N = 10
max_val, min_val = 1, -1
range_size = (max_val - min_val)  # 2
np.random.rand(N) * range_size + min_val

or other equivalent algebra.

Alternatively, you could use np.random.randint() to generate a int range to be divided by (a fraction of) its size.

For your specific range, this would look like:

import numpy as np


N = 10
L = 100
np.random.randint(-L, L, N) / L

This would be (slightly) slower than the np.random.random() approach, but would give you control over the "density" of the result.

(EDITED: explicitly write the relationship between the algebra and the target range)

edited Dec 21, 2019 at 8:04

answered Dec 18, 2019 at 9:57

norok2

27.1k6 gold badges83 silver badges110 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

CypherX Over a year ago

The second option will create a more stratified or binned random sampling, since there are only 200/10 = 20 possible values now. This gives the control over density. However, this may or may not be applicable to the specific use-case. The best thing, perhaps is to apply np.random.rand and then use some form of binning if necessary. But yes, both methods could be applicable ways of sampling in some scenarios. If someone is blindly applying uniform random sampling, I would suggest to use the first method.

CypherX · Accepted Answer · 2019-12-21 08:15:30Z

1

If you want to generate a random float between two values a and b, where a < b, then it is done as follows. This is equivalent to redistributing or re-scaling the random-distribution between [0,1] to the desired interval [a,b] (equivalent to stretching or compressing, depending on whether (b - a) is <, = or > (1 - 0).

import numpy as np
# in your case:
a, b = -1, 1
(b - a)*np.random.rand() + a

edited Dec 21, 2019 at 8:15

answered Dec 21, 2019 at 8:09

CypherX

7,4034 gold badges29 silver badges39 bronze badges

1 Comment

Ícaro Pires Over a year ago

This is correct, are the same instructions provided in the official documentation. Quote: (...). To sample Unif[a, b), b > a multiply the output of random by (b-a) and add a. (...)

Collectives™ on Stack Overflow

How to generate a random float in range [-1,1] using numpy.random.rand()?

2 Answers 2

1 Comment

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related