如何生成随机数，每个随机数至少与所有其他元素相差x？

Question

I know this goes against the definition of random numbers, but still I require this for my project. For instance, I want to generate an array with 5 random elements in range(0, 200) .

Now, I want each of the elements to have a difference of at least 15 between them. So the random array should look something like this:

[15, 45, 99, 132, 199]

I can generate random numbers using numpy:

np.random.uniform(low=0, high=200, size=5)

However, I am not able to keep a consistent difference of at least 15.

Answer 1

It would be nice if the question showed more effort towards solving the problem (ie from the Stack Overflow Tour : "Don't ask about... Questions you haven't tried to find an answer for (show your work!)"), but sometimes a question triggers an itch you just have to scratch...

Here's one way you could do it, written as the function random_spaced :

import numpy as np


def random_spaced(low, high, delta, n, size=None):
    """
    Choose n random values between low and high, with minimum spacing delta.

    If size is None, one sample is returned.
    Set size=m (an integer) to return m samples.

    The values in each sample returned by random_spaced are in increasing
    order.
    """
    empty_space = high - low - (n-1)*delta
    if empty_space < 0:
        raise ValueError("not possible")

    if size is None:
        u = np.random.rand(n)
    else:
        u = np.random.rand(size, n)
    x = empty_space * np.sort(u, axis=-1)
    return low + x + delta * np.arange(n)

For example,

In [27]: random_spaced(0, 200, 15, 5)
Out[27]: array([ 30.3524969 ,  97.4773284 , 140.38221631, 161.9276264 , 189.3404236 ])

In [28]: random_spaced(0, 200, 15, 5)
Out[28]: array([ 81.01616136, 103.11710522, 118.98018499, 141.68196775, 169.02965952])

The size argument lets you generate more than one sample at a time:

In [29]: random_spaced(0, 200, 15, 5, size=3)
Out[29]: 
array([[ 52.62401348,  80.04494534,  96.21983265, 138.68552066, 178.14784825],
       [  7.57714106,  33.05818556,  62.59831316,  81.86507168, 180.30946733],
       [ 24.16367913,  40.37480075,  86.71321297, 148.24263974, 195.89405713]])

This code generates a histogram for each component using 100000 samples, and plots the corresponding theoretical marginal PDFs of each component:

import matplotlib.pyplot as plt
from scipy.stats import beta

low = 0
high = 200
delta = 15 
n = 5
s = random_spaced(low, high, delta, n, size=100000)

for k in range(s.shape[1]):
    plt.hist(s[:, k], bins=100, density=True, alpha=0.25)
plt.title("Normalized marginal histograms and marginal PDFs")
plt.grid(alpha=0.2)

# Plot the PDFs of the marginal distributions of each component.
# These are beta distributions.
for k in range(n):
    left = low + k*delta
    right = high - (n - k - 1)*delta
    xx = np.linspace(left, right, 400)
    yy = beta.pdf(xx, k + 1, n - k, loc=left, scale=right - left)
    plt.plot(xx, yy, 'k--', linewidth=1, alpha=0.25)
    if n > 1:
        # Mark the mode with a dot.
        mode0 = k/(n-1)
        mode = (right-left)*mode0 + left
        plt.plot(mode, beta.pdf(mode, k + 1, n - k, loc=left, scale=right - left),
                 'k.', alpha=0.25)

plt.show()

Here's the plot that it generates:

As can be seen in the plot, the marginal distributions are beta distributions . The modes of the marginal distributions correspond to the positions of n evenly spaced points on the interval [low, high] .

By fiddling with how u is generated in random_spaced , distributions with different marginals can be generated (an old version of this answer had an example), but the distribution that random_spaced currently generates seems to be a natural choice. As mentioned above, the modes of the marginals occur in "meaningful" positions. Moreover, in the trivial case where n is 1, the distribution simplifies to the uniform distribution on [ low , high ].

Answer 2

How about trial-and-error? eg throw some random numbers, sort, compute differences... and if too small repeat?

import random as r

def spreadRandom(theRange, howMany, minSpacing):
    while True:
        candidate = sorted([r.randint(*theRange) for _ in range(howMany)])
        minDiff = min([ candidate[i+1]-candidate[i] for i, _ in enumerate(candidate[:-1])])
        if minDiff >= minSpacing:
            return candidate

spreadRandom([0,200], 5, 15)

You're not guaranteed to ever get an answer, but you're not biasing your numbers at all as you might be by enforcing ranges based on neighboring numbers.

Answer 3

Try shuffling the numbers 0-200:

import random
numbers = list(range(200))
random.shuffle(numbers)
distant_numbers = [numbers[0]]
for number in numbers:
    if any(abs(number - x) < 15 for x in distant_numbers):
        continue
    distant_numbers.append(number)
    if len(distant_numbers) >= 5: break

Edit:

Here's a solution that uses z3 for massive overkill:

def spaced_randoms(n, d, R, first=None):
    solver = z3.SolverFor("QF_FD")
    numbers = [z3.Int("x{}".format(x)) for x in range(n)]
    for number in numbers:
        solver.add(number >= 0)
        solver.add(number <= R)
    for ii in range(n):
        for jj in range(ii+1,n):
            solver.add(z3.Or(numbers[ii] - numbers[jj] > d, numbers[ii] - numbers[jj] < -d))
    if first is not None:
        solver.add(numbers[0] == first)
    result = solver.check()
    if str(result) != "sat":
        raise Exception("Unsatisfiable")
    model = solver.model()
    return [model.get_interp(number) for number in numbers]

Call it like this for a random result:

import random
spaced_randoms(n, d, R, random.randint(0,R))

Answer 4

I think this code might help for your specific needs:

import random
import numpy as np
five_list = np.asarray([])
end = False
number = random.randint(0,200)
five_list = np.append(five_list,number)
while True:
    new_number = random.randint(0,200)
    if all(np.absolute(np.subtract(five_list, new_number)) >= 15):
        five_list = np.append(five_list,new_number)
    if np.size(five_list) == 5:
        break
print(np.sort(five_list)) 

Answer 5

Try it with "brute force":

l= [ i for i in range(201) ]
rslt= []
for i in range(5): 
    n=random.choice(l) 
    rslt.append(n) 
    l=[ k for k in l if abs(k-n)>=15 ]
    #if not l:
    #   break

Or smartly:

sgmnts= [(0,200)]
diff= 15
rslt= []

for i in range(5):

    start,stop= sgmnts.pop( random.choice(range(len(sgmnts))) )
    n= random.choice(range(start,stop+1))
    rslt.append(n)
    if n-diff > start:
        sgmnts.append( (start,n-diff) )
    if n+diff < stop:
        sgmnts.append( (n+diff,stop) )
    if not sgmnts:
        break

"sgmnts" stores the suitable ranges. We select a range randomly, too, by index.

Answer 6

This will generate 5 random values between 200 with a step of 5

import random

array = []

randomRange = 200
arrayRange = 5
valueStep = 15

for loop in range(arrayRange):
    randomMaxValue = randomRange - valueStep * (arrayRange - loop) # First loop will the first randomMaxValue be 125 next will be 140, 155, 170, 185, 200
    if not array: # Checks if the array is empty
        array.append(random.randint(0, randomMaxValue)) # Appends a value between 0 and 125 (First will be 125 because 200 - 15 * 5)
    else:
        array.append(random.randint(array[-1] + 15, randomMaxValue)) # Appends the 4 next values

print(array)