def test_evaluate_gradient():
grid = StructuredGrid()
# test by setting the scalar field to the y coordinate
vector = grid.evaluate_gradient(grid.barycentre, grid.nodes[:, 1])
assert np.sum(vector - np.array([0, 1, 0])) == 0
# same test but for a bigger grid, making sure scaling for cell is ok
grid = StructuredGrid(step_vector=np.array([100, 100, 100]))
vector = grid.evaluate_gradient(grid.barycentre, grid.nodes[:, 1])
assert np.sum(vector - np.array([0, 1, 0])) == 0
def test_FDI():
xy = np.array(np.meshgrid(np.linspace(0, 1, 50),
np.linspace(0, 1, 50))).T.reshape(-1, 2)
xyz = np.hstack([xy, np.zeros((xy.shape[0], 1))])
data = pd.DataFrame(xyz, columns=['X', 'Y', 'Z'])
data['val'] = np.sin(data['X'])
data['w'] = 1
data['feature_name'] = 'strati'
randind = np.arange(0, len(data))
# np.random.shuffle(randind)
# # data.loc[randind[:int(50*50*.5)],'val'] = np.nan
# np.random.shuffle(randind)
# data.loc[randind[:int(50*50*.1)],'val'] = 0
origin = np.array([-0.1, -0.1, -0.1])
maximum = np.array([1.1, 1.1, 1.1])
nsteps = np.array([20, 20, 20])
step_vector = (maximum - origin) / nsteps
grid = StructuredGrid(origin=origin,
nsteps=nsteps,
step_vector=step_vector)
interpolator = FDI(grid)
interpolator.set_value_constraints(data[['X', 'Y', 'Z', 'val',
'w']].to_numpy())
interpolator._setup_interpolator()
interpolator.solve_system()
assert np.sum(
interpolator.evaluate_value(data[['X', 'Y', 'Z']].to_numpy()) -
data[['val']].to_numpy()) / len(data) < 0.5
def test_inequality_FDI_nodes():
try:
import osqp
except ImportError:
print('osqp not installed')
return
xy = np.array(np.meshgrid(np.linspace(0, 1, 50),
np.linspace(0, 1, 50))).T.reshape(-1, 2)
xyz = np.hstack([xy, np.zeros((xy.shape[0], 1))])
data = pd.DataFrame(xyz, columns=['X', 'Y', 'Z'])
data['val'] = np.sin(data['X'])
data['w'] = 1
data['feature_name'] = 'strati'
data['l'] = -3
data['u'] = 10
randind = np.arange(0, len(data))
origin = np.array([-0.1, -0.1, -0.1])
maximum = np.array([1.1, 1.1, 1.1])
nsteps = np.array([20, 20, 20])
step_vector = (maximum - origin) / nsteps
grid = StructuredGrid(origin=origin,
nsteps=nsteps,
step_vector=step_vector)
interpolator = FDI(grid)
interpolator.set_value_constraints(data[['X', 'Y', 'Z', 'val',
'w']].to_numpy())
interpolator.set_inequality_constraints(data[['X', 'Y', 'Z', 'l',
'u']].to_numpy())
interpolator._setup_interpolator()
interpolator.solve_system(solver='osqp')
def test_global_to_local_coordinates():
grid = StructuredGrid()
point = np.array([[1.2, 1.5, 1.7]])
lx, ly, lz = grid.position_to_local_coordinates(point)
assert (np.isclose(lx[0], .2))
assert (np.isclose(ly[0], .5))
assert (np.isclose(lz[0], .7))
def test_equality_FDI_nodes():
xy = np.array(np.meshgrid(np.linspace(0, 1, 50),
np.linspace(0, 1, 50))).T.reshape(-1, 2)
xyz = np.hstack([xy, np.zeros((xy.shape[0], 1))])
data = pd.DataFrame(xyz, columns=['X', 'Y', 'Z'])
data['val'] = np.sin(data['X'])
data['w'] = 1
data['feature_name'] = 'strati'
origin = np.array([-0.1, -0.1, -0.1])
maximum = np.array([1.1, 1.1, 1.1])
nsteps = np.array([20, 20, 20])
step_vector = (maximum - origin) / nsteps
grid = StructuredGrid(origin=origin,
nsteps=nsteps,
step_vector=step_vector)
interpolator = FDI(grid)
interpolator.set_value_constraints(data[['X', 'Y', 'Z', 'val',
'w']].to_numpy())
node_idx = np.arange(
0, interpolator.nx)[interpolator.support.nodes[:, 2] > .9]
interpolator.add_equality_constraints(node_idx,
np.ones(node_idx.shape[0]),
name='top')
interpolator._setup_interpolator()
interpolator.solve_system(solver='cg')
def test_outside_box():
grid = StructuredGrid()
# test by setting the scalar field to the y coordinate
inside = grid.inside(grid.barycentre + 5)
assert np.all(
~inside == np.any((grid.barycentre + 5) > grid.maximum, axis=1))
inside = grid.inside(grid.barycentre - 5)
assert np.all(
~inside == np.any((grid.barycentre - 5) < grid.origin, axis=1))
cix, ciy, ciz = grid.position_to_cell_index(grid.barycentre - 5)
assert np.all(cix[inside] < grid.nsteps_cells[0])
assert np.all(ciy[inside] < grid.nsteps_cells[1])
assert np.all(ciz[inside] < grid.nsteps_cells[2])
cornersx, cornersy, cornersz = grid.cell_corner_indexes(cix, ciy, ciz)
assert np.all(cornersx[inside] < grid.nsteps[0])
assert np.all(cornersy[inside] < grid.nsteps[1])
assert np.all(cornersz[inside] < grid.nsteps[2])
globalidx = grid.global_indicies(
np.dstack([cornersx, cornersy, cornersz]).T)
# print(globalidx[inside],grid.n_nodes,inside)
assert np.all(globalidx[inside] < grid.n_nodes)
inside = grid.inside(grid.barycentre - 5)
inside, grid.position_to_cell_corners(grid.barycentre - 5)
vector = grid.evaluate_gradient(grid.barycentre - 5, grid.nodes[:, 1])
assert np.sum(np.mean(vector[inside, :], axis=0) -
np.array([0, 1, 0])) == 0
vector = grid.evaluate_gradient(grid.nodes, grid.nodes[:, 1])
def test_create_structured_grid_origin_nsteps():
grid = StructuredGrid(
origin=np.zeros(3),
nsteps=np.array([10, 10, 10]),
step_vector=np.array([0.1, 0.1, 0.1]),
)
assert np.sum(grid.step_vector - np.array([0.1, 0.1, 0.1])) == 0
assert np.sum(grid.maximum - np.ones(3)) == 0
def test_change_maximum_and_origin():
grid = StructuredGrid(origin=np.zeros(3), nsteps=np.array([5, 5, 5]))
grid.origin = np.array([-1.0, -1.0, -1.0])
assert np.all(np.isclose(grid.origin, np.array([-1, -1, -1])))
assert np.all(np.isclose(grid.nsteps, np.array([6, 6, 6])))
assert np.all(np.isclose(grid.step_vector, np.ones(3)))
grid.maximum = np.array([7.0, 7.0, 7.0])
assert np.all(np.isclose(grid.nsteps, np.array([9, 9, 9])))
assert np.all(np.isclose(grid.maximum, np.array([7.0, 7.0, 7.0])))
assert np.all(np.isclose(grid.step_vector, np.ones(3)))
def test_evaluate_gradient2():
# this test is the same as above but we will use a random vector
np.random.seed(0)
for i in range(10):
step = np.random.uniform(0, 100)
grid = StructuredGrid(step_vector=np.array([step, step, step]))
# define random vector
n = np.random.random(3)
n /= np.linalg.norm(n)
distance = (n[0] * grid.nodes[:, 0] + n[1] * grid.nodes[:, 1] +
n[2] * grid.nodes[:, 2])
vector = grid.evaluate_gradient(np.random.uniform(1, 8, size=(100, 3)),
distance)
assert np.all(np.isclose(np.sum(vector - n[None, :], axis=1),
0)) == True
assert i == 9
def test_inequality_FDI():
try:
import osqp
except ImportError:
print('osqp not installed')
return
xy = np.array(np.meshgrid(np.linspace(0, 1, 50),
np.linspace(0, 1, 50))).T.reshape(-1, 2)
xyz = np.hstack([xy, np.zeros((xy.shape[0], 1))])
data = pd.DataFrame(xyz, columns=['X', 'Y', 'Z'])
data['val'] = np.sin(data['X'])
data['w'] = 1
data['feature_name'] = 'strati'
data['l'] = -3
data['u'] = 10
randind = np.arange(0, len(data))
# np.random.shuffle(randind)
# # data.loc[randind[:int(50*50*.5)],'val'] = np.nan
# np.random.shuffle(randind)
# data.loc[randind[:int(50*50*.1)],'val'] = 0
origin = np.array([-0.1, -0.1, -0.1])
maximum = np.array([1.1, 1.1, 1.1])
nsteps = np.array([20, 20, 20])
step_vector = (maximum - origin) / nsteps
grid = StructuredGrid(origin=origin,
nsteps=nsteps,
step_vector=step_vector)
interpolator = FDI(grid)
interpolator.set_value_constraints(data[['X', 'Y', 'Z', 'val',
'w']].to_numpy())
interpolator.set_inequality_constraints(data[['X', 'Y', 'Z', 'l',
'u']].to_numpy())
interpolator._setup_interpolator()
# col = np.arange(0,interpolator.nx,dtype=int)
# col = np.tile(col, (interpolator.nx, 1)).T
# interpolator.add_inequality_constraints_to_matrix(np.eye(interpolator.nx),
# np.zeros(interpolator.nx)-4,
# np.zeros(interpolator.nx)+np.inf,
# col
# )
interpolator.solve_system(solver='osqp')
def __init__(self,
values,
step_vector,
nsteps,
origin=np.zeros(3),
name="scalar field"):
"""[summary]
Parameters
----------
values : numpy array
nodes values of regular grid
step_vector : numpy array
step vector for x,y,z step
nsteps : numpy array
number of steps
name : string, optional
name of the feature for the visualisation
"""
self.values = values
self.grid = StructuredGrid(origin, nsteps, step_vector)
self.name = name
def test_create_structured_grid():
grid = StructuredGrid()
def test_get_element_outside():
grid = StructuredGrid()
point = np.array([grid.origin - np.ones(3)])
idc, inside = grid.position_to_cell_corners(point)
assert inside[0] == False
def test_change_origin():
grid = StructuredGrid(origin=np.zeros(3), nsteps=np.array([5, 5, 5]))
grid.origin = np.array([-1, -1, -1])
assert np.all(grid.origin == np.array([-1, -1, -1]))
assert np.all(grid.nsteps == np.array([6, 6, 6]))
assert np.all(grid.step_vector == np.ones(3))
def test_get_element():
grid = StructuredGrid()
point = grid.barycentre()[[0], :]
idc, inside = grid.position_to_cell_corners(point)
bary = np.mean(grid.nodes[idc, :], axis=0)
assert np.sum(point - bary) == 0
def test_evaluate_gradient():
grid = StructuredGrid()
grid.update_property('Y', grid.nodes[:, 1])
vector = np.mean(grid.evaluate_gradient(grid.barycentre(), 'Y'), axis=0)
# vector/=np.linalg.norm(vector)
assert np.sum(vector - np.array([0, grid.step_vector[1], 0])) == 0
def test_evaluate_value():
grid = StructuredGrid()
grid.update_property('X', grid.nodes[:, 0])
assert np.sum(grid.barycentre()[:, 0] -
grid.evaluate_value(grid.barycentre(), 'X')) == 0
def test_create_structured_grid_origin_nsteps():
grid = StructuredGrid(origin=np.zeros(3), nsteps=np.array([5, 5, 5]))
assert grid.n_nodes == 5 * 5 * 5
assert np.sum(grid.maximum - np.ones(3) * 5) == 0
def test_change_maximum():
grid = StructuredGrid(origin=np.zeros(3), nsteps=np.array([5, 5, 5]))
grid.maximum = np.array([7, 7, 7])
assert np.all(np.isclose(grid.nsteps, np.array([8, 8, 8])))
assert np.all(np.isclose(grid.maximum, np.array([7, 7, 7])))
assert np.all(np.isclose(grid.step_vector, np.ones(3)))
def test_evaluate_value():
grid = StructuredGrid()
assert (
np.sum(grid.barycentre[:, 0] -
grid.evaluate_value(grid.barycentre, grid.nodes[:, 0])) == 0)