def generate_testing_point(single_formula, category, dimension, number, lower_bound, large_bound):
number_of_point = int(math.pow(number, (1.0 / dimension)))
output = []
if number_of_point == 1:
for i in range(number):
point = []
for j in range(dimension):
value = np.random.uniform(lower_bound, large_bound)
point.append(value)
output.append(point)
else:
step = (large_bound - lower_bound) / float(number_of_point)
x_list = []
for i in range(number_of_point):
x_list.append(lower_bound + i * step)
output = list(product(x_list, repeat=dimension))
label_list = []
point_list = []
for i in output:
i = list(i)
formula_tester = lt.FormulaLabelTester(single_formula)
if category == formula.POLYHEDRON:
is_true = formula_tester.polycircle_model(single_formula.get_formula()[0],
single_formula.get_formula()[1], i)
else:
is_true = formula_tester.polynomial_model(single_formula.get_formula()[:-1], i,
single_formula.get_formula()[-1])
if is_true:
label_list.append([1.0])
else:
label_list.append([0.0])
point_list.append(i)
return point_list, label_list
def generate_polyhedron_points(formu, to_be_appended_random_points_number,
lower_bound, upper_bound):
formu = formu.get_formula()
dim = len(formu[0][0])
outputX = []
outputY = []
for i in range(to_be_appended_random_points_number):
generated_point = []
for j in range(dim):
generated_point.append(random.uniform(lower_bound, upper_bound))
# k = random.randint(2, 3)
# if k % 2 == 0:
# center = random.randint(0, len(formu[0]) - 1)
# for i in range(dim):
# generated_point.append(random.uniform(int(formu[0][center][i]) - 10, int(formu[0][center][i]) + 10))
# else:
# for i in range(dim):
# generated_point.append(random.uniform(-10, 10))
formula_tester = lt.FormulaLabelTester()
flag = formula_tester.polycircle_model(formu[0], formu[1],
generated_point)
outputX.append(generated_point)
if (flag):
outputY.append([1])
else:
outputY.append([0])
return outputX, outputY
def generate_random_points_for_polynomial(single_formula, lower_bound, upper_bound, num):
formula_tester = lt.FormulaLabelTester(single_formula)
data_list = []
label_list = []
coefficient_list = single_formula.get_formula()[:-1]
y = single_formula.get_formula()[-1]
while len(data_list) < num:
point = []
variable_num = len(coefficient_list)
for i in range(variable_num):
point.append(random.randint(lower_bound, upper_bound))
label = formula_tester.polynomial_model(coefficient_list, point, y)
data_list.append(point)
label_list.append([label])
return data_list, label_list
def generate_partitioned_random_points_for_sphere(single_formula, lower_bound, upper_bound,
positive_num, negative_num): # [[[12,0],[-12,0]],[4,4]]
formu_list = single_formula.get_formula()
dim = len(formu_list[0][0])
positive_list = []
negative_list = []
while len(positive_list) < positive_num or len(negative_list) < negative_num:
data_point = []
center = random.choice(formu_list[0])
index = formu_list[0].index(center)
radius = formu_list[1][index]
move = math.sqrt(radius ** 2 / dim)
r = random.uniform(0, 1)
point = []
for i in range(dim):
length = random.uniform(0, move)
if r > 0.5:
step = length
value = center[i] + step
else:
value = random.uniform(lower_bound, upper_bound)
point.append(value)
formula_tester = lt.FormulaLabelTester(single_formula)
label = formula_tester.polycircle_model(formu_list[0], formu_list[1], point)
if label:
if len(positive_list) < positive_num:
positive_list.append(point)
else:
if len(negative_list) < negative_num:
negative_list.append(point)
return positive_list, negative_list
def generate_random_points_for_sphere(single_formula, lower_bound, upper_bound,
data_point_number): # [[[12,0],[-12,0]],[4,4]]
formu_list = single_formula.get_formula()
dim = len(formu_list[0][0])
data_list = []
label_list = []
for k in range(data_point_number):
data_point = []
point = []
if k % 3 == 0:
center = random.randint(0, len(formu_list[0]) - 1)
for i in range(dim):
point.append(
random.uniform(int(formu_list[0][center][i]) - 300, int(formu_list[0][center][i]) + 300))
else:
for i in range(dim):
point.append(random.uniform(lower_bound, upper_bound))
data_list.append(point)
formula_tester = lt.FormulaLabelTester(single_formula)
label = formula_tester.polycircle_model(formu_list[0], formu_list[1], point)
label_list.append([label])
return data_list, label_list
def generate_partitioned_random_points_for_polynomial(single_formula, lower_bound, upper_bound, positive_num,
negative_num):
formula_tester = lt.FormulaLabelTester(single_formula)
positive_list = []
negative_list = []
coefficient_list = single_formula.get_formula()[:-1]
y = single_formula.get_formula()[-1]
while len(positive_list) < positive_num or len(negative_list) < negative_num:
point = []
variable_num = len(coefficient_list)
for i in range(variable_num):
point.append(random.randint(lower_bound, upper_bound))
label = formula_tester.polynomial_model(coefficient_list, point, y)
if label:
if len(positive_list) < positive_num:
positive_list.append(point)
else:
if len(negative_list) < negative_num:
negative_list.append(point)
return positive_list, negative_list
def generate_polynomial_points(single_formula,
to_be_appended_random_points_number,
lower_bound, upper_bound):
single_formula = single_formula.get_formula()
coefficientList = single_formula[:-1]
y = single_formula[-1]
outputX = []
outputY = []
for i in range(to_be_appended_random_points_number):
xList = []
variableNum = len(coefficientList)
for j in range(variableNum):
xList.append(random.randint(lower_bound, upper_bound))
formula_tester = lt.FormulaLabelTester(single_formula)
flag = formula_tester.polynomial_model(coefficientList, xList, y)
outputX.append(xList)
if (flag):
outputY.append([1])
else:
outputY.append([0])
return outputX, outputY
upper_bound = 1000
learning_rate = 0.01
training_epochs = 1000
parts_num = 5
data_point_number = 200
util.reset_random_seed()
train_set_x, train_set_y, test_set_x, test_set_y = formula_data_point_generation.generate_partitioned_data(
parent_formula, category,
lower_bound,
upper_bound,
50, 50)
label_tester = lt.FormulaLabelTester(parent_formula)
point_number_limit = 200
util.reset_random_seed()
model_folder = "models/test-method/test-branch"
model_file = "test-branch"
util.reset_random_seed()
train_acc, test_acc = benchmark.generate_accuracy(train_set_x, train_set_y, test_set_x, test_set_y, learning_rate,
training_epochs, lower_bound, upper_bound, model_folder, model_file)
child_label_tester = lt.FormulaLabelTester(child_formula)
train_set_x1, train_set_y1, _, _ = formula_data_point_generation.generate_partitioned_data(
child_formula, category,
-400,
400,
0, 20)
# else:
# ws.write(i, j, str(col))
# else:
# ws.write(i, j, col)
# print()
index = 0
for f in formula_list:
start_time = time.time()
util.reset_random_seed()
print(f.get_formula())
train_set_x, train_set_y, test_set_x, test_set_y = formula_data_point_generation.generate_partitioned_data(
f, category, lower_bound, upper_bound, 50, 50)
label_tester = lt.FormulaLabelTester(f)
point_number_limit = 100
util.reset_random_seed()
tf.reset_default_graph()
train_acc_list, test_acc_list, data_point_number_list, appended_point_list = mal.generate_accuracy(
train_set_x[0:50], train_set_y[0:50], test_set_x, test_set_y,
learning_rate, training_epochs, lower_bound, upper_bound, False,
label_tester, point_number_limit, model_folder, model_file)
tf.reset_default_graph()
util.reset_random_seed()
index += 1
train_acc, test_acc = benchmark.generate_accuracy(
train_set_x, train_set_y, test_set_x, test_set_y, learning_rate,
training_epochs, lower_bound, upper_bound, model_folder, model_file)
end_time = time.time()
upper_bound = 1000
learning_rate = 0.01
training_epochs = 1000
parts_num = 5
data_point_number = 200
util.reset_random_seed()
train_set_x, train_set_y, test_set_x, test_set_y = formula_data_point_generation.generate_partitioned_data(
parent_formula, category,
lower_bound,
upper_bound,
50, 50)
label_tester = lt.FormulaLabelTester(parent_formula)
point_number_limit = 200
util.reset_random_seed()
model_folder = "models/test-method/test-branch"
model_file = "test-branch"
# util.reset_random_seed()
# train_acc, test_acc = benchmark.train(train_set_x, train_set_y, test_set_x, test_set_y, learning_rate,
# training_epochs, lower_bound, upper_bound, model_folder, model_file)
positive_x = []
for i in range(len(train_set_x)):
point = train_set_x[i]
if train_set_y[i][0] == 1:
positive_x.append(point)
