I have a list of candidates which looks like this:

candidates = [
    {
        'name': 'John',
        'rank': 13
    },
    {
        'name': 'Joe',
        'rank': 8
    },
    {
        'name': 'Averell',
        'rank': 5
    },
    {
        'name': 'William',
        'rank': 2
    }
]

What I want to do is semi-randomly pick one of these candidates, based on its squared rank. So that a candidate A having a rank twice as big as B, will have 4 times more chances to be picked than B.

Here is a naive implementation of the idea I had to solve this problem:

def pick_candidate(candidates):
    # initiates a global probability space
    prob_space = []

    # fills the probability space with the indexes
    # of 'candidates', so that prob_space looks like:
    # [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1]
    # if candidates[0] has rank 3 and candidates[1] has rank 2
    for i, c in enumerate(candidates):
        rank = c['rank']
        for j in range(int(rank*rank)):
            prob_space.append(i)

    # picks a random index from the probability space
    picked_prob_space_index = random.randint(0, len(prob_space)-1)

    # retrieves the matching candidate
    picked_candidate_index = prob_space[picked_prob_space_index]

    return candidates[picked_candidate_index]

The questions I'm thinking of, concerning the above code, are:

• Concept: Is the core principle of the algorithm (the idea of building prob_space, etc) overkill and solvable more easily in other ways or with builtins
• Implementation: This core principle put aside, what do you think of the implementation? Would you think of a better, cleaner way to write it?