Source code for mapof.elections.cultures.guardians_approx

import itertools as it
import math

import mapof.core.features.mallows as core_mallows
import numpy as np

import mapof.elections.cultures.mallows as mallows
import mapof.elections.cultures.sampling.samplemat as smpl


def distribute_in_matrix(n, m):
    if m == 0:
        return []
    k = n // m
    r = n - k * m
    matrix = []
    for i in range(m):
        row = [k for _ in range(m)]
        for j in range(i, i + r):
            if j >= m:
                j = j - m
            row[j] = row[j] + 1
        matrix.append(row)
    return matrix


def _distribute_in_block_matrix(n, blocks):
    before = 0
    after = sum(blocks)
    matrix = []
    for b in blocks:
        after = after - b
        block = distribute_in_matrix(n, b)
        for row in block:
            matrix.append([0 for _ in range(before)] + row + [0 for _ in range(after)])
        before = before + b
    return (matrix)


def draw_election(matrix):
    return smpl.sample_election_using_permanent(matrix)


def build_perms(matrix):
    if len(matrix[0]) == 0:
        return [[]]
    perms = []
    last_col = []
    for j in range(len(matrix)):
        last_col.append(matrix[j].pop())
    if sum(last_col) > 0:
        for last_cand, elig in enumerate(last_col):
            if elig <= 0:
                continue
            zeroed_vector = matrix[last_cand][:]
            matrix[last_cand][:] = len(matrix[last_cand]) * [0]
            for perm in build_perms(matrix):
                perms.append(perm + [last_cand])
            matrix[last_cand][:] = zeroed_vector[:]
    for j in range(len(matrix)):
        matrix[j].append(last_col[j])
    return perms


# Models
def generate_un_from_list(num_voters=None, num_candidates=None):
    id_perm = list(range(num_candidates))
    m_fac = math.factorial(num_candidates)
    alls = num_voters // m_fac
    rest = num_voters - alls * m_fac
    res = []
    for _ in range(alls):
        res = res + [list(v) for v in it.permutations(id_perm)]
    res = res + [list(v) for v in it.permutations(id_perm)][:rest]
    return res


[docs] def generate_approx_uniformity_votes(num_voters=None, num_candidates=None): """ Generate real election that have UN positionwise frequency_matrix """ matrix = distribute_in_matrix(num_voters, num_candidates) return draw_election(matrix)
[docs] def generate_idan_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate real election between (ID) and (AN) """ if part_share is None: print("IDAN_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) id_share = num_voters - (part_size // 2) op_share = part_size // 2 votes = [[j for j in range(num_candidates)] for _ in range(id_share)] votes = votes + [[(num_candidates - j - 1) for j in range(num_candidates)] for _ in range(op_share)] return votes
[docs] def generate_idun_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate elections realizing linear combinations of pos-matrices between (ID) and (UN) """ if part_share is None: print("IDUN_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) id_share = num_voters - part_size un_share = part_size votes = [[j for j in range(num_candidates)] for _ in range(id_share)] votes = votes + draw_election(distribute_in_matrix(un_share, num_candidates)) return votes
[docs] def generate_idst_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate elections realizing linear combinations of pos-matrices between (ID) and (ST) """ if part_share is None: print("IDST_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) id_share = num_voters - part_size st_share = part_size topsize = num_candidates // 2 bottomsize = num_candidates - topsize votes_id = [[j for j in range(num_candidates)] for _ in range(id_share)] votes_st = draw_election(_distribute_in_block_matrix(st_share, [topsize, bottomsize])) return votes_id + votes_st
[docs] def generate_anun_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate elections realizing linear combinations of pos-matrices between (AN) and (UN) """ if part_share is None: print("ANUN_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) id_share = (num_voters - part_size) // 2 op_share = num_voters - part_size - id_share un_share = num_voters - id_share - op_share votes = [[j for j in range(num_candidates)] for _ in range(id_share)] votes = votes + [[(num_candidates - j - 1) for j in range(num_candidates)] for _ in range(op_share)] votes = votes + draw_election(distribute_in_matrix(un_share, num_candidates)) return votes
[docs] def generate_anst_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate elections realizing linear combinations of pos-matrices between (AN) and (ST) """ if part_share is None: print("ANST_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) id_share = (num_voters - part_size) // 2 op_share = num_voters - part_size - id_share st_share = num_voters - id_share - op_share topsize = num_candidates // 2 bottomsize = num_candidates - topsize votes = [[j for j in range(num_candidates)] for _ in range(id_share)] votes = votes + [[(num_candidates - j - 1) for j in range(num_candidates)] for _ in range(op_share)] votes = votes + draw_election(_distribute_in_block_matrix(st_share, [topsize, bottomsize])) return votes
[docs] def generate_unst_part_votes(num_voters=None, num_candidates=None, part_share=None, **kwargs): """ Generate elections realizing linear combinations of pos-matrices between (UN) and (ST) """ if part_share is None: print("UNST_part generation : params None : random param generated") part_size = np.random.choice(range(num_voters)) else: part_size = part_share * (num_voters) part_size = int(round(part_size)) un_share = num_voters - part_size st_share = part_size topsize = num_candidates // 2 bottomsize = num_candidates - topsize votes = draw_election(distribute_in_matrix(un_share, num_candidates)) votes = votes + draw_election(_distribute_in_block_matrix(st_share, [topsize, bottomsize])) return votes
def generate_idan_mallows_votes(num_voters=None, num_candidates=None, params=None): if params is None or not ('scaled-phi' in params): print("IDAN_mallows generation : params None : random param generated") is_reversed = np.random.choice([True, False]) phi = core_mallows.phi_from_normphi(num_candidates, normphi=np.random.uniform()) else: if params['scaled-phi'] > 0.5: is_reversed = True phi = core_mallows.phi_from_normphi(num_candidates, (1 - params['scaled-phi']) * 2) else: is_reversed = False phi = core_mallows.phi_from_normphi(num_candidates, params['scaled-phi'] * 2) id_share = num_voters // 2 op_share = num_voters - id_share votes_id = [[j for j in range(num_candidates)] for _ in range(id_share)] votes_op = mallows.generate_mallows_votes(op_share, num_candidates, {'phi': phi}) if is_reversed: for v in votes_op: v.reverse() return votes_id + votes_op def generate_idst_mallows_votes(num_voters=None, num_candidates=None, params=None): if params is None or not ('phi' in params): print("IDST_mallows generation : params None : random param generated") phi = core_mallows.phi_from_normphi(num_candidates, normphi=np.random.uniform()) else: phi = params['phi'] better = num_candidates // 2 worse = num_candidates - better votes_better = mallows.generate_mallows_votes(num_voters, better, {'phi': phi}) votes_worse = mallows.generate_mallows_votes(num_voters, worse, {'phi': phi}) votes = [] for b, w in zip(votes_better, votes_worse): w_ = [c + better for c in w] v = b + w_ votes.append(v) return votes def generate_anun_mallows_votes(num_voters=None, num_candidates=None, params=None): mallows_params = {} if params is None or not ('phi' in params): print("IDST_mallows generation : params None : random param generated") mallows_params['phi'] = core_mallows.phi_from_normphi(num_candidates, normphi=np.random.uniform()) else: mallows_params['phi'] = params['phi'] mallows_params['weight'] = 0.5 return mallows.generate_mallows_votes(num_voters, num_candidates, mallows_params) def generate_unst_mallows_votes(num_voters=None, num_candidates=None, params=None): if params is None or not ('phi' in params): print("IDST_mallows generation : params None : random param generated") phi = core_mallows.phi_from_normphi(num_candidates, normphi=np.random.uniform()) / 2 else: phi = params['phi'] / 2 better = num_candidates // 2 worse = num_candidates - better votes = draw_election(_distribute_in_block_matrix(num_voters, [better, worse])) # The next part works poorly for odd number of candidates (the last one is always from worse part) for v in votes: for i in range(better): if np.random.random() < phi: c = v[i] v[i] = v[i + better] v[i + better] = c return votes
[docs] def generate_unst_topsize_votes(num_voters=None, num_candidates=None, top_share=None, **kwargs): """ Generate kind of real elections between (UN) and (ST) """ if top_share is None: print("UNST_topsize generation : params None : random param generated") top_share = np.random.random() else: top_share = top_share top_size = int(round(top_share * num_candidates)) better = top_size worse = num_candidates - top_size matrix = _distribute_in_block_matrix(num_voters, [better, worse]) return draw_election(matrix)
[docs] def generate_idst_blocks_votes(num_voters=None, num_candidates=None, no_blocks=None, **kwargs): """ Generate kind of real elections between (ID) and (UN) """ if no_blocks is None: print("IDST_blocks generation : params None : random param generated") no_blocks = np.random.choice(range(num_candidates + 1)) else: no_blocks = no_blocks no_blocks = int(round(no_blocks)) k = num_candidates // no_blocks r = num_candidates - k * no_blocks blocks = [k for _ in range(no_blocks)] with_one_more = list(np.random.choice(range(no_blocks), r, replace=False)) for i in with_one_more: blocks[i] = blocks[i] + 1 matrix = _distribute_in_block_matrix(num_voters, blocks) return draw_election(matrix)
[docs] def generate_approx_stratification_votes(num_voters=None, num_candidates=None, weight=0.5): """ Generate real election that approximates stratification (ST) """ first_group_size = int(num_candidates * weight) return [list(np.random.permutation(first_group_size)) + list(np.random.permutation([j for j in range(first_group_size, num_candidates)])) for _ in range(num_voters)]