def test_fast_segmentation(self):
n = 360
k = 3
epsilon = 1
generate_input_file(n)
data = np.genfromtxt("input.csv", delimiter=" ")
p = np.c_[np.mgrid[1:n + 1], data]
D = Coreset.build_coreset(p, k, epsilon)
print len(D)
x = np.empty((0, 4))
for coreset in D:
print "coreset range", coreset.e - coreset.b + 1
pts = utils.pt_on_line(xrange(int(coreset.b), int(coreset.e) + 1), coreset.g)
# TODO: 2nd parameter should be epsilon
w = Coreset.PiecewiseCoreset(len(pts[0]), epsilon)
p_coreset = np.column_stack((pts[0], pts[1], pts[2], w))
p_coreset_filtered = p_coreset[p_coreset[:, 3] > 0]
# print "weighted points", p_coreset_filtered
x = np.append(x, p_coreset_filtered, axis=0)
print "num of weighted points", len(x)
dividers = ksegment.coreset_k_segment_fast_segmentation(x, k)
print "dividers", dividers
print "dividers-cost:", utils.calc_cost_dividers(p, dividers)
utils.visualize_3d(p, dividers)
def test_basic_demo(self):
# dimension = 2
k = 3
epsilon = 0.5
n = 600
generate_input_file(n)
data = np.genfromtxt("input.csv", delimiter=" ")
p = np.c_[np.mgrid[1:n + 1], data]
coreset = Coreset.build_coreset(p, k, epsilon)
dividers = ksegment.coreset_k_segment(coreset, k)
utils.visualize_3d(p, dividers)
def test_fast_segmentation(self):
n = 600
k = 3
epsilon = 10
generate_input_file(n)
data = np.genfromtxt("input.csv", delimiter=" ")
p = np.c_[np.mgrid[1:n + 1], data]
D = Coreset.build_coreset(p, k, epsilon)
dividers = ksegment.coreset_k_segment_fast_segmentation(D, k, epsilon)
print "dividers", dividers
print "dividers-cost:", utils.calc_cost_dividers(p, dividers)
utils.visualize_3d(p, dividers)
def test_coreset_merging(self):
# generate points
n = 120
# dimension = 2
k = 3
epsilon = 0.1
# data = random_data(N, dimension)
# for example1 choose N that divides by 6
data = example1(n)
p = np.c_[np.mgrid[1:n + 1], data]
coreset = Coreset.build_coreset(p, k, epsilon)
coreset_of_coreset = Coreset.build_coreset(coreset, k, epsilon, is_coreset=True)
dividers = ksegment.coreset_k_segment(coreset_of_coreset, k)
utils.visualize_3d(p, dividers)
def plot_3d(size, map_indices, path_indices):
# load network
abstraction_vin = Abstraction_VIN_3D(size)
# load network state
if os.path.isfile('network/%s.pt' % abstraction_vin.name):
abstraction_vin.load_state_dict(
torch.load('network/%s.pt' % abstraction_vin.name))
abstraction_vin.to(abstraction_vin.device)
else:
print("Abstraction_VIN was not trained.")
return
for map_index, path_index in zip(map_indices, path_indices):
map = dataset.grids[map_index] # environment map
optimal_path = dataset.expert_paths[map_index][path_index]
start = optimal_path[0]
goal = optimal_path[-1]
print('Map index: %d' % map_index)
print('Path index: %d' % path_index)
print('Optimal path length: %f' %
get_path_length(optimal_path, dim=3))
# predict path
abstraction_vin_path, abstraction_vin_success = _get_path_3d(
abstraction_vin, dataset, map, map_index, start, goal,
2 * len(optimal_path))
abstraction_vin_path = torch.stack(abstraction_vin_path, dim=0)
if abstraction_vin_success:
print('Abstraction_VIN path length: %f' %
get_path_length(abstraction_vin_path, dim=2))
else:
print('Abstraction_VIN: No path found.')
# plot paths
visualize_3d(map, goal, optimal_path, abstraction_vin_path)
