def learn_voting_rule(self, voting_rule, one_vs_one=None):
headers = []
sample_set = []
labels = []
for i in range(len(self.preference_profiles)):
preference_profile = self.preference_profiles[i]
positional_score_matrix = data_generator.positional_score_matrix(
self.candidates, preference_profile)
weighted_majority_graph = data_generator.weighted_majority_graph(
self.candidates, preference_profile)
histogram = data_generator.histogram(self.candidates,
preference_profile)
x = data_generator.get_feature_vector(positional_score_matrix,
weighted_majority_graph,
histogram,
preference_profile,
self.candidates)
y = voting_rule(self.candidates, preference_profile)
feature_names = data_generator.get_feature_names(
positional_score_matrix, weighted_majority_graph)
if i == 0:
headers = feature_names
sample_set.append(x)
labels.append(y)
print(len(labels))
x_data = learn.load_x(headers, sample_set)
y_data = learn.load_y(self.candidates, labels)
#print(x_data)
#print(y_data)
if (one_vs_one): learn.one_vs_one(x_data, y_data)
else: learn.train(x_data, y_data)
def test_distance_calculation_time(candidates, preference_profiles, n,
distance_type):
histograms = []
for pp in preference_profiles:
hist = data_generator.histogram(candidates, pp)
histograms.append([e[1] for e in hist])
start_time = time.time()
for i in range(n):
h1 = random.choice(histograms)
h2 = random.choice(histograms)
distance_type(h1, h2)
print("--- seconds ---", (time.time() - start_time) / n)
def ball_tree_test(candidates, preference_profiles,
preference_profiles_test, voting_rule, distance_type,
k):
labels = []
test_labels = []
data_histograms = []
test_histograms = []
for pp in preference_profiles:
labels.append(voting_rule(candidates, pp))
hist = data_generator.histogram(candidates, pp)
data_histograms.append([e[1] for e in hist])
for pp in preference_profiles_test:
hist = data_generator.histogram(candidates, pp)
test_histograms.append([e[1] for e in hist])
print(len(data_histograms))
tree = KNN.build_ball_tree(data_histograms, distance_type)
count = 0
n = 1000
for i in range(n):
indexs = KNN.ball_tree_query(tree, test_histograms[i], k)
winners = [labels[index] for index in indexs]
print(winners)
winner = max(set(winners), key=indexs.count)
print(winner)
print("this histogram is:", test_histograms[i])
#print("closest_histogram is:",data_histograms[index])
if winner == voting_rule(candidates, preference_profiles_test[i]):
count += 1
print('correct')
else:
print('wrong')
print(i / n, "%")
print(count / n)
return (count / n)
def get_Xs(candidates, preference_profiles):
Xs = []
for i in range(len(preference_profiles)):
preference_profile = preference_profiles[i]
positional_score_matrix = data_generator.positional_score_matrix(
candidates, preference_profile)
weighted_majority_graph = data_generator.weighted_majority_graph(
candidates, preference_profile)
histogram = data_generator.histogram(candidates,
preference_profile)
X = data_generator.get_feature_vector(positional_score_matrix,
weighted_majority_graph,
histogram,
preference_profile,
candidates)
Xs.append(X)
return Xs
def MPSR_tiebreaking(candidates, preference_profile, winners):
max_rankings = []
max = 0
histogram = data_generator.histogram(candidates, preference_profile)
#print(histogram)
for e in histogram:
if e[1] > max:
max_rankings = []
max_rankings.append(e[0])
max = e[1]
elif e[1] == max:
max_rankings.append(e[0])
#print(max_rankings)
if len(max_rankings) == 1:
max_ranking = max_rankings[0]
for i in range(len(max_ranking)):
if max_ranking[i] in winners:
return max_ranking[i]
else:
return None
def get_histograms(candidates, preference_profiles):
histograms = []
for pp in preference_profiles:
hist = data_generator.histogram(candidates, pp)
histograms.append([e[1] for e in hist])
return histograms