My thoughts on enhancements to make random support a broader set of iterables.

better_random.py

import random
import itertools

def sample(population, k):
    "How random.sample should really be defined."

    if callable(getattr(population, '__len__', None)) and callable(getattr(population, '__getitem__', None)):
        return random.sample(population, min(k, len(population)))

    p = iter(population)
    sample = list(itertools.islice(p, 0, k))

    for p_index, item in enumerate(p, k):
        r_index = random.randint(0, p_index)
        if r_index < k:
            sample[r_index] = item

    return sample

def choice(seq):
    "How random.choice should really be defined."

    if callable(getattr(seq, '__len__', None)):
        if callable(getattr(seq, '__getitem__', None)):
            return random.choice(seq)
        
        return next(itertools.islice(seq, random.randrange(0, len(seq)), None))
    else:
        return sample(seq, 1)[0]

def subset(population, k=1):
    "Really how sample & choice could & should be one function."
    if (k == 1):
        try:
            return [choice(population)]
        except (IndexError, ValueError):
            return []

    return sample(population, k)