def iterate(self, f, search_area):
x_new = generate_vector_in_sphere(self.x, self.radius)
x_new = bound_vector(x_new, search_area)
f_x_new = f(x_new)
if f_x_new < self.f_x:
self.x = x_new
self.f_x = f_x_new
def _move_stars(self, f, search_area):
black_hole = self.stars[self.black_hole_id]
self.stars = [
s + np.random.rand() * (black_hole - s)
for i, s in enumerate(self.stars) if i != self.black_hole_id
]
self.stars = [bound_vector(s, search_area)
for s in self.stars] + [black_hole]
self._calculate_quality_levels(f)
def iterate(self, f, search_area):
center_of_mass = self._get_center_of_mass()
self.points = [center_of_mass + self._generate_delta(x) for x in self.points]
self.points = [bound_vector(x, search_area) for x in self.points]
self.points.append(center_of_mass)
self._calculate_values(f)
self.iteration_number += 1
def iterate(self, f, search_area):
gradient = approx_fprime(self.x, f, epsilon=self.eps)
gradient_length = np.linalg.norm(gradient)
if gradient_length > 0:
gradient /= gradient_length
x_new = self.x - np.random.uniform(0.0, self.radius) * gradient
x_new = bound_vector(x_new, search_area)
f_x_new = f(x_new)
if f_x_new < self.f_x:
self.x = x_new
self.f_x = f_x_new
def test_bound_vector_1():
v = np.array([1, 2, 3])
bounds = np.array([[-5, 5] * len(v)])
npt.assert_equal(v, bound_vector(v, bounds))
def test_bound_vector_N(_):
search_area = np.array([[-10, 10], [-10, 10], [-10, 10]])
bounds = np.array([[-5, 5], [-5, 5], [-5, 5]])
vector = generate_vector_in_area(search_area)
vector = bound_vector(vector, bounds)
assert ((bounds[:, 0] <= vector) & (vector <= bounds[:, 1])).all()
def test_bound_vector_3():
v = np.array([5, -5, 0])
bounds = np.array([[-2, 2] * len(v)])
npt.assert_equal(np.array([2, -2, 0]), bound_vector(v, bounds))
def test_bound_vector_2():
v = np.array([1, 2, 3])
bounds = np.array([[-2, 2] * len(v)])
npt.assert_equal(np.array([1, 2, 2]), bound_vector(v, bounds))