def aux_gamma_beta(self, i, gamma, j, beta):
# logging.info("(%d,%d): g %.4f; b %.4f", i, j, gamma, beta)
stv_kls = np.zeros(self.n_tests)
sntv_liars_kls = np.zeros(self.n_tests)
sntv_kls = np.zeros(self.n_tests)
for test_nr in range(self.n_tests):
true_preferences = PreferenceCreator(self.n, self.m, self.political_spectrum).create_preferences()
poll_results = get_poll(true_preferences, gamma)
man_ballot = manipulate(true_preferences, np.nonzero(poll_results < self.electoral_threshold)[0], beta)
first_distribution = get_first_choice_dist(true_preferences)
stv_scores, *_ = voting.STV_scores(true_preferences, self.electoral_threshold, percentage=True)
stv_outcome = apportion.largest_remainder(stv_scores, self.seats)
stv_kls[test_nr] = utils.kl_divergence(stv_outcome / self.seats, first_distribution)
# SNTV outcome truthful
sntv_scores, *_ = voting.SNTV_scores(true_preferences, self.electoral_threshold, percentage=True)
sntv_outcome = apportion.largest_remainder(sntv_scores, self.seats)
sntv_kls[test_nr] = utils.kl_divergence(sntv_outcome / self.seats, first_distribution)
del sntv_scores, sntv_outcome
# SNTV outcome liars
sntv_scores, *_ = voting.SNTV_scores(man_ballot, self.electoral_threshold, percentage=True)
sntv_outcome = apportion.largest_remainder(sntv_scores, self.seats)
sntv_liars_kls[test_nr] = utils.kl_divergence(sntv_outcome / self.seats, first_distribution)
logging.info("(%d,%d): g %.4f; b %.4f; kl-sntv_liars: %.4f", i, j, gamma, beta, sntv_liars_kls.mean())
return i, j, (stv_kls.mean(), sntv_kls.mean(), sntv_liars_kls.mean())
def lalala(electoral_threshold, m, n, n_tests, political_spectrum, seats, beta, poll_covid):
stv_propor = []
sntv_propor = []
for test_nr in range(n_tests):
true_preferences = PreferenceCreator(n, m, political_spectrum).create_preferences()
poll_results = get_poll(true_preferences, poll_covid)
man_ballot = manipulate(true_preferences, np.nonzero(poll_results < electoral_threshold)[0], beta)
# STV outcome
# TODO reestablish the natural order
stv_scores, *_ = voting.STV_scores(true_preferences, electoral_threshold, percentage=True)
stv_outcome = apportion.largest_remainder(stv_scores, seats)
# SNTV outcome
sntv_scores, *_ = voting.SNTV_scores(man_ballot, electoral_threshold, percentage=True)
sntv_outcome = apportion.largest_remainder(sntv_scores, seats)
first_distribution = get_first_choice_dist(true_preferences)
# Proportionality
stv_propor.append(utils.kl_divergence(stv_outcome / seats, first_distribution))
sntv_propor.append(utils.kl_divergence(sntv_outcome / seats, first_distribution))
return np.array(stv_propor).mean(), np.array(sntv_propor).mean()
def test_gamma_beta(electoral_threshold, m, n, n_tests, political_spectrum, seats):
steps = 3
gammass = np.geomspace(0.0001, 1, num=steps)
betass = np.linspace(0, 1, num=steps)
results = {'stv': np.zeros((steps, steps)), 'sntv': np.zeros((steps, steps))}
for i, gamma in enumerate(gammass):
for j, beta in enumerate(betass):
logging.info("g %.4f; b %.4f", gamma, beta)
stv_kls = np.zeros(n_tests)
sntv_kls = np.zeros(n_tests)
for test_nr in range(n_tests):
true_preferences = PreferenceCreator(n, m, political_spectrum).create_preferences()
poll_results = get_poll(true_preferences, gamma)
man_ballot = manipulate(true_preferences, np.nonzero(poll_results < electoral_threshold)[0], beta)
first_distribution = get_first_choice_dist(true_preferences)
stv_scores, *_ = voting.STV_scores(true_preferences, electoral_threshold, percentage=True)
stv_outcome = apportion.largest_remainder(stv_scores, seats)
stv_kls[test_nr] = utils.kl_divergence(stv_outcome / seats, first_distribution)
# SNTV outcome
sntv_scores, *_ = voting.SNTV_scores(man_ballot, electoral_threshold, percentage=True)
sntv_outcome = apportion.largest_remainder(sntv_scores, seats)
sntv_kls[test_nr] = utils.kl_divergence(sntv_outcome / seats, first_distribution)
results['stv'][i, j] = stv_kls.mean()
results['sntv'][i, j] = sntv_kls.mean()
return np.meshgrid(gammass, betass), results
def test_alpha(electoral_threshold, m, n, n_tests, political_spectrum, seats, poll_covid):
props = []
# progressbar = tqdm(total=50)
alphass = np.linspace(0.01, 1, num=10)
for i, alpha in enumerate(alphass):
print(i)
# progressbar.update()
astv_propor = []
astv_l_propor = []
for test_nr in range(n_tests):
true_preferences = PreferenceCreator(n, m, political_spectrum).create_preferences()
poll_results = get_poll(true_preferences, poll_covid)
man_ballot = manipulate(true_preferences, np.nonzero(poll_results < electoral_threshold)[0], 1.)
first_distribution = get_first_choice_dist(true_preferences)
# a-STV outcome
astv_scores, *_ = voting.alpha_STV_scores(true_preferences, alpha, electoral_threshold, percentage=True)
astv_outcome = apportion.largest_remainder(astv_scores, seats)
# a-STV liars outcome
astv_l_scores, *_ = voting.alpha_STV_scores(man_ballot, alpha, electoral_threshold, percentage=True)
astv_l_outcome = apportion.largest_remainder(astv_l_scores, seats)
astv_propor.append(utils.kl_divergence(astv_outcome / seats, first_distribution))
astv_l_propor.append(utils.kl_divergence(astv_l_outcome / seats, first_distribution))
a = np.array(astv_propor).mean()
al = np.array(astv_l_propor).mean()
# a = lalala(electoral_threshold, m, n, n_tests, political_spectrum, seats, beta, 0.01)
props.append((a, al))
# progressbar.close()
mean_astv_prop, mean_astv_l_prop = list(zip(*props))
a = np.array(mean_astv_prop)
b = np.array(mean_astv_l_prop)
# print(a, b)
plt.plot(alphass, a, label='a-stv')
plt.plot(alphass, b, label='a-stv-liars')
plt.legend()
plt.show()