CODEWITHSUNDEEP
pythonnumpyrandom
VSA
VSA

Reputation: 71

Generating random numbers in numpy with strict lower bounds

So according to numpy's documentation here: https://docs.scipy.org/doc/numpy-1.12.0/reference/generated/numpy.random.uniform.html, they say that the random values generated belong to the set [lower, upper) i.e. all values are >= lower, and less than upper. Is there any way to make this a stricter bound, i.e. all values are > lower? I have a particular case in which I want to ensure that all random values lie between 0 and k, but should not be equal to 0, as that will cause my program to crash. Any workarounds/other libraries which can help me?

Upvotes: 1

Views: 2670

Answers (2)

James
James

Reputation: 36608

The chance of actually getting 0 from a call to np.ranodm.uniform(0,k) is practially zero. If you want to guarantee it, you can set a lower value that is very small.

epsilon = np.finfo(np.float).eps
np.random.uniform(epsilon, k)

Edit:

For my machine, epsilon is 2.2204460492503131e-16. The chance of getting any specific number should be roughly 1 / (1/epsilon), or 1 / 4503599627370496.

As an example, the following code checks for 0.0 occurring in a million length array of np.random.normal(0,1):

counter = 0
stop = False
while not stop:
    x = np.random.normal(0,1, size=1000000)
    stop = any(x==0.0)
    counter += 1
    print('\rIteration: {}   '.format(counter), end='')

I am currently over 75,000 iteration without a zero occurring. This is not a perfect test, clearly, but it does demonstrate the minute chance of actually getting a zero.

Upvotes: 4

MB-F
MB-F

Reputation: 23637

I want to ensure that all random values lie between 0 and k, but should not be equal to 0

If you have numbers in the range low <= u < high you can easily convert them to low < r <= high:

r = high - np.random.uniform(0, high - low)

Of course this is only useful if r is allowed to include high but not low.

Upvotes: 1

Related Questions