Python random unique values from two non overlapping ranges

randlist = [r + (end - start + 1) * (r >= start) for r in
            random.sample(range(total - end + start), 100)]

Example / "proof":

• total=10, start=2, end=5
• There are 7 allowed numbers: 0, 1, 6, 7, 8, 9, 10
• range(total-end+start) = range(7) picks from 7 numbers 0..6 (so far so good)
• Numbers larger than or equal to start=2 are shifted upwards by end-start+1=4
• Resulting numbers are in 0, 1, 6, 7, 8, 9, 10.

Demo:

>>> sorted(randlist2(2000000000, 10000000, 1900000000))
[176827, 3235435, 3278133, 3673989, 5148447, 8314140, 8885997, 1900189345, 1902880599,
...
1997494057, 1997538971, 1997854443, 1997907285]

This works until over 2 billion, easily beating the required upper limit of "the number of wikipedia english wikipedia pages, so whatever many million that is" :-). After that it gets OverflowError: Python int too large to convert to C ssize_t. I see no spike in memory usage of my PC and the result is instant. This is using Python 3, obviously.

Try the following function:

def rand_choice(start, end, amount, istart, iend):
    from random import randint
    all = []
    for i in range(amount):
        randnum = istart
        while randnum in range(istart, iend+1) or randnum in all:
            randnum = randint(start, end)
        all.append(randnum)
    return all

>>> rand_choice(1, 1000, 10, 10, 20)
[30, 798, 427, 229, 943, 386, 749, 925, 520, 877]
>>> rand_choice(1, 1000, 10, 10, 20)
[414, 351, 898, 813, 91, 205, 751, 269, 360, 501]
>>> 

Just a slight variation of the original:

def randlist(total, start, end):
    import random
    randset = set()
    while len(randset) < 100:
        temp = random.randint(0, total)
        start <= temp <= end or randset.add(temp)
    return random.sample(randset, 100)

Another answerer earlier had a pretty cool idea of concatenating ranges together into a single Sequence class. The code had some issues, but I managed to make a version that seems to work with random.sample.

import collections, random

class range_duo(collections.Sequence):
    def __init__(self, r1, r2):
        self.r1 = r1
        self.r2 = r2
        self.l1 = len(r1)
        self.l2 = len(r2)
        self.total_length = len(r1) + len(r2)

    def __len__(self):
        return self.total_length

    def __getitem__(self, key):
        if key < self.l1:
            return self.r1[key]
        else:
            return self.r2[key-self.l1]


# Solving the example in the original question:
rd = range_duo(range(0, 10), range(20, 10000))
sample = random.sample(rd, 100)
print(sample)

Obviously this class isn't perfect, but my only goal was to solve the problem with random.sample using a minimum memory footprint. In Python 2.x, xrange should be used instead of range.