def init_arms(self, cluster_positions, cluster_fitnesses):
# Default matrix that translate and shrink search space to [0,1]^D
matrices = [
Matrix(positions, fitnesses, self.min_bounds,
self.max_bounds) for positions, fitnesses in zip(
cluster_positions, cluster_fitnesses)
]
if self.plot > 0:
draw_optimization(function_id - 1,
cluster_positions,
matrices,
fig_name='initial.png',
**self.fig_config)
# Random sample n_samples to represent search space for non-overlapped optimization
trans_samples = np.random.uniform(0,
1,
size=(self.n_samples,
self.dimension))
k = len(cluster_positions)
for i in range(k):
exclude = []
for j in range(k):
if i != j:
samples = matrices[j].inverse_transform(trans_samples)
exclude.extend(samples)
exclude = np.array(exclude)
if self.verbose: print('optimizing matrix for arm %d' % i)
arm = Arm(self.obj,
self.n_points,
cluster_positions[i],
cluster_fitnesses[i],
algo_type=self.algo_type,
exclude=exclude,
min_bounds=self.min_bounds,
max_bounds=self.max_bounds)
self.arms.append(arm)
matrices[i] = self.arms[i].matrix
if self.plot > 0:
opt_points = []
for j in range(k):
if i == j:
opt_points.append(cluster_positions[i])
else:
samples = matrices[j].inverse_transform(trans_samples)
opt_points.append(samples)
draw_optimization(function_id - 1,
opt_points,
matrices,
fig_name='initial_optimize_%d.png' % i,
**self.fig_config)
if self.plot > 0:
opt_points = [arm.get_positions() for arm in self.arms]
opt_matrices = [arm.matrix for arm in self.arms]
draw_optimization(function_id - 1,
opt_points,
opt_matrices,
fig_name='initial_optimized.png',
**self.fig_config)
