Full-range random number in Python

Question

I'm generating a series of random floats using this line:

random.random()*(maxval-minval) + minval

I'm using it to add variable noise to a given variable, and the amount of noise added depends on a series of factors. In some cases, the noise should be so high that in practice the original value is lost, and I have a completely random value.

In this context, the code works works with finite values, but if I use "inf" it returns NaN. Is there a workaround to allow a continuos random range that might include the infinity? I don't want to tamper with os.random() as it is machine-specific.

score 5 · Answer 1 · answered May 08 '15 at 07:52

5

If you define a uniform random distribution over an infinite domain, the probability of any value in the domain being chosen is infinitesimal. What you're asking for doesn't make any mathematical sense.

answered May 08 '15 at 07:52

Asad Saeeduddin

46,193
6
90
139

score 2 · Answer 2 · answered May 08 '15 at 08:42

As it was said before, you can't have uniform distribution over the whole real line, but you can use other random distributions which have real line support. Consider Cauchy distribution. It has 'heavy-tails', which simply means that there is a decent probability of getting very big numbers.

score 1 · Accepted Answer · answered May 08 '15 at 08:20

1

After the discussion in comments i suggest the following :

>>> m=sys.maxint
>>> np.random.uniform(-m,m,5)
array([ -5.32362215e+18,  -2.90131323e+18,   5.14492175e+18,
        -5.64238742e+18,  -3.49640768e+18])

As is said the you can get the max integer with sys.maxint then you can use np.random.randint to get a random number between the maxint and -maxint.

answered May 08 '15 at 08:20

Mazdak

105,000
18
159
188

The minimum value in Python 2 is `-sys.maxint - 1`, [source](https://docs.python.org/2/library/stdtypes.html#numeric-types-int-float-long-complex). – Agostino Jun 09 '15 at 02:02

score 1 · Answer 4 · edited May 23 '17 at 10:24

1

As @Asad says, what you are trying is mathematically not quite sound. But what you could do, is the following:

define a very big number (maybe this post helps: What is the range of values a float can have in Python?)
use random.uniform(0, biggestValue) as an approximation for random values according to your needs.

Maybe this is what you are looking for.

edited May 23 '17 at 10:24

Community

1
1

answered May 08 '15 at 08:21

WWhisperer

97
1
7

Full-range random number in Python

4 Answers4

Linked