def __init__(self, N=64, seed=None, **kwargs):
self.seed = seed
if N == 64:
data = utils.loadmat('pointclouds/david64')
assert data['N'][0, 0] == N
W = data['W']
coords = data['coords']
elif N == 500:
data = utils.loadmat('pointclouds/david500')
assert data['N'][0, 0] == N
W = data['W']
coords = data['coords']
else:
coords = np.random.RandomState(seed).rand(N, 2)
target_dist_cutoff = -0.125 * N / 436.075 + 0.2183
T = 0.6
s = np.sqrt(-target_dist_cutoff**2/(2*np.log(T)))
d = utils.distanz(coords.T)
W = np.exp(-np.power(d, 2)/(2.*s**2))
W[W < T] = 0
W[np.diag_indices(N)] = 0
plotting = {"limits": [0, 1, 0, 1]}
super(DavidSensorNet, self).__init__(W, coords=coords,
plotting=plotting, **kwargs)
def __init__(self, N=64, seed=None, **kwargs):
self.seed = seed
if N == 64:
data = utils.loadmat('pointclouds/david64')
assert data['N'][0, 0] == N
W = data['W']
coords = data['coords']
elif N == 500:
data = utils.loadmat('pointclouds/david500')
assert data['N'][0, 0] == N
W = data['W']
coords = data['coords']
else:
coords = np.random.RandomState(seed).rand(N, 2)
target_dist_cutoff = -0.125 * N / 436.075 + 0.2183
T = 0.6
s = np.sqrt(-target_dist_cutoff**2 / (2 * np.log(T)))
d = utils.distanz(coords.T)
W = np.exp(-np.power(d, 2) / (2. * s**2))
W[W < T] = 0
W[np.diag_indices(N)] = 0
plotting = {"limits": [0, 1, 0, 1]}
super(DavidSensorNet, self).__init__(W,
coords=coords,
plotting=plotting,
**kwargs)
def __init__(self, N=64):
if N == 64:
david64 = PointsCloud("david64")
W = david64.W
coords = david64.coords
elif N == 500:
david500 = PointsCloud("david500")
W = david500.W
coords = david500.coords
else:
coords = np.random.rand(N, 2)
target_dist_cutoff = -0.125 * N / 436.075 + 0.2183
T = 0.6
s = np.sqrt(-target_dist_cutoff**2/(2*np.log(T)))
d = distanz(coords.T)
W = np.exp(-np.power(d, 2)/(2.*s**2))
W[W < T] = 0
W[np.diag_indices(N)] = 0
plotting = {"limits": [0, 1, 0, 1]}
super(DavidSensorNet, self).__init__(W=W, gtype='davidsensornet',
coords=coords, plotting=plotting)
def __init__(self, N=400, a=1, b=4, dim=3, thresh=1e-6, s=None,
noise=False, srtype='uniform', seed=None, **kwargs):
if s is None:
s = np.sqrt(2. / N)
self.a = a
self.b = b
self.dim = dim
self.thresh = thresh
self.s = s
self.noise = noise
self.srtype = srtype
self.seed = seed
rng = np.random.default_rng(seed)
y1 = rng.uniform(size=N)
y2 = rng.uniform(size=N)
if srtype == 'uniform':
tt = np.sqrt((b * b - a * a) * y1 + a * a)
elif srtype == 'classic':
tt = (b - a) * y1 + a
tt *= np.pi
if dim == 2:
x = np.array((tt * np.cos(tt), tt * np.sin(tt)))
elif dim == 3:
x = np.array((tt * np.cos(tt), 21 * y2, tt * np.sin(tt)))
if noise:
x += rng.normal(size=x.shape)
self.x = x
self.dim = dim
coords = utils.rescale_center(x)
dist = utils.distanz(coords)
W = np.exp(-np.power(dist, 2) / (2. * s**2))
W -= np.diag(np.diag(W))
W[W < thresh] = 0
plotting = {
'vertex_size': 60,
'limits': np.array([-1, 1, -1, 1, -1, 1]),
'elevation': 15,
'azimuth': -90,
'distance': 7,
}
super(SwissRoll, self).__init__(W, coords=coords.T,
plotting=plotting,
**kwargs)
def __init__(self, N=400, a=1, b=4, dim=3, thresh=1e-6, s=None,
noise=False, srtype='uniform', seed=None, **kwargs):
if s is None:
s = np.sqrt(2. / N)
self.a = a
self.b = b
self.dim = dim
self.thresh = thresh
self.s = s
self.noise = noise
self.srtype = srtype
self.seed = seed
rs = np.random.RandomState(seed)
y1 = rs.rand(N)
y2 = rs.rand(N)
if srtype == 'uniform':
tt = np.sqrt((b * b - a * a) * y1 + a * a)
elif srtype == 'classic':
tt = (b - a) * y1 + a
tt *= np.pi
if dim == 2:
x = np.array((tt * np.cos(tt), tt * np.sin(tt)))
elif dim == 3:
x = np.array((tt * np.cos(tt), 21 * y2, tt * np.sin(tt)))
if noise:
x += rs.randn(*x.shape)
self.x = x
self.dim = dim
coords = utils.rescale_center(x)
dist = utils.distanz(coords)
W = np.exp(-np.power(dist, 2) / (2. * s**2))
W -= np.diag(np.diag(W))
W[W < thresh] = 0
plotting = {
'vertex_size': 60,
'limits': np.array([-1, 1, -1, 1, -1, 1]),
'elevation': 15,
'azimuth': -90,
'distance': 7,
}
super(SwissRoll, self).__init__(W, coords=coords.T,
plotting=plotting,
**kwargs)
def __init__(self, N=400, a=1, b=4, dim=3, thresh=1e-6, s=None,
noise=False, srtype='uniform', seed=None, **kwargs):
if s is None:
s = np.sqrt(2. / N)
rs = np.random.RandomState(seed)
y1 = rs.rand(N)
y2 = rs.rand(N)
if srtype == 'uniform':
tt = np.sqrt((b * b - a * a) * y1 + a * a)
elif srtype == 'classic':
tt = (b - a) * y1 + a
tt *= np.pi
if dim == 2:
x = np.array((tt * np.cos(tt), tt * np.sin(tt)))
elif dim == 3:
x = np.array((tt * np.cos(tt), 21 * y2, tt * np.sin(tt)))
if noise:
x += rs.randn(*x.shape)
self.x = x
self.dim = dim
coords = utils.rescale_center(x)
dist = utils.distanz(coords)
W = np.exp(-np.power(dist, 2) / (2. * s**2))
W -= np.diag(np.diag(W))
W[W < thresh] = 0
plotting = {
'vertex_size': 60,
'limits': np.array([-1, 1, -1, 1, -1, 1]),
'elevation': 15,
'azimuth': -90,
'distance': 7,
}
gtype = 'swiss roll {}'.format(srtype)
super(SwissRoll, self).__init__(W=W, coords=coords.T,
plotting=plotting, gtype=gtype,
**kwargs)
def _create_weight_matrix(self, N, distributed, regular, param_Nc):
XCoords = np.zeros((N, 1))
YCoords = np.zeros((N, 1))
rs = np.random.RandomState(self.seed)
if distributed:
mdim = int(np.ceil(np.sqrt(N)))
for i in range(mdim):
for j in range(mdim):
if i * mdim + j < N:
XCoords[i * mdim + j] = np.array(
(i + rs.rand()) / mdim)
YCoords[i * mdim + j] = np.array(
(j + rs.rand()) / mdim)
# take random coordinates in a 1 by 1 square
else:
XCoords = rs.rand(N, 1)
YCoords = rs.rand(N, 1)
coords = np.concatenate((XCoords, YCoords), axis=1)
# Compute the distanz between all the points
target_dist_cutoff = 2 * N**(-0.5)
T = 0.6
s = np.sqrt(-target_dist_cutoff**2 / (2 * np.log(T)))
d = utils.distanz(x=coords.T)
W = np.exp(-d**2 / (2. * s**2))
W -= np.diag(np.diag(W))
if regular:
W = self._get_nc_connection(W, param_Nc)
else:
W2 = self._get_nc_connection(W, param_Nc)
W = np.where(W < T, 0, W)
W = np.where(W2 > 0, W2, W)
W = sparse.csc_matrix(W)
return W, coords
def __init__(self, N=400, a=1, b=4, dim=3, thresh=1e-6, s=None,
noise=False, srtype='uniform'):
if s is None:
s = sqrt(2./N)
y1 = np.random.rand(N)
y2 = np.random.rand(N)
if srtype == 'uniform':
tt = np.sqrt((b * b - a * a) * y1 + a * a)
elif srtype == 'classic':
tt = (b - a) * y1 + a
tt *= pi
if dim == 2:
x = np.array((tt*np.cos(tt), tt * np.sin(tt)))
elif dim == 3:
x = np.array((tt*np.cos(tt), 21 * y2, tt * np.sin(tt)))
if noise:
x += np.random.randn(*x.shape)
self.x = x
self.dim = dim
dist = distanz(coords)
W = np.exp(-np.power(dist, 2) / (2. * s**2))
W -= np.diag(np.diag(W))
W[W < thresh] = 0
coords = self.rescale_center(x)
plotting = {'limits': np.array([-1, 1, -1, 1, -1, 1])}
gtype = 'swiss roll {}'.format(srtype)
super(SwissRoll, self).__init__(W=W, coords=coords.T,
plotting=plotting, gtype=gtype)
def create_weight_matrix(self, N, param_distribute, param_regular, param_Nc):
XCoords = np.zeros((N, 1))
YCoords = np.zeros((N, 1))
if param_distribute:
mdim = int(ceil(sqrt(N)))
for i in range(mdim):
for j in range(mdim):
if i*mdim + j < N:
XCoords[i*mdim + j] = np.array((i + np.random.rand()) / mdim)
YCoords[i*mdim + j] = np.array((j + np.random.rand()) / mdim)
# take random coordinates in a 1 by 1 square
else:
XCoords = np.random.rand(N, 1)
YCoords = np.random.rand(N, 1)
coords = np.concatenate((XCoords, YCoords), axis=1)
# Compute the distanz between all the points
target_dist_cutoff = 2*N**(-0.5)
T = 0.6
s = sqrt(-target_dist_cutoff**2/(2*log(T)))
d = distanz(x=coords.T)
W = np.exp(-d**2/(2.*s**2))
W -= np.diag(np.diag(W))
if param_regular:
W = self.get_nc_connection(W, param_Nc)
else:
W2 = self.get_nc_connection(W, param_Nc)
W = np.where(W < T, 0, W)
W = np.where(W2 > 0, W2, W)
W = sparse.csc_matrix(W)
return W, coords
def test_distanz(x, y):
# TODO test with matlab to compare
self.assertEqual(utils.distanz(x, y))
def test_distanz(x, y):
# TODO test with matlab to compare
self.assertEqual(utils.distanz(x, y))