def krig(vario, grid, **kwargs):
# get the grid
xx, yy = grid
ok = OrdinaryKriging(vario, **kwargs)
field = ok.transform(xx.flatten(), yy.flatten())
sigma = ok.sigma
return field.reshape(xx.shape), sigma.reshape(xx.shape)
def ordinary_kriging(x, y, z, grid, model='spherical', estimator='matheron',
n_lags=15, maxlag='median', min_points=5, max_points=15,
mode='exact', precision=1000, **settings):
# build coordinates
coords = __point_array(x, y)
# fit a Variogram
V = Variogram(coords, z, model=model, estimator=estimator, n_lags=n_lags,
maxlag=maxlag, normalize=False)
# get the shape and build the Kriging
shape = grid[0].shape
ok = OrdinaryKriging(V, min_points=min_points, max_points=max_points,
mode=mode, precision=precision)
# apply
return ok.transform(grid[0].flatten(), grid[1].flatten()).reshape(shape)
def setUp(self):
# Generate some random but spatially correlated data
# with a range of ~20
np.random.seed(42)
c = np.random.sample((50, 2)) * 60
np.random.seed(42)
v = np.random.normal(10, 4, 50)
V = Variogram(c, v).describe()
V["effective_range"] = 20
OK = OrdinaryKriging(V, coordinates=c, values=v)
self.c = np.random.sample((500, 2)) * 60
self.v = OK.transform(self.c)
self.c = self.c[~np.isnan(self.v), :]
self.v = self.v[~np.isnan(self.v)]
def test_ordinary(self):
if not GSTOOLS_AVAILABLE: # pragma: no cover
return True
x = np.array([self.c[0][0]])
y = np.array([self.c[0][1]])
# run ordinary kriging with skgstat
ok = OrdinaryKriging(self.V, min_points=3)
sk_res = ok.transform(x, y)
# get the gstools Krige class
krige = self.V.to_gs_krige()
gs_res, _ = krige.structured([x, y])
# test
assert_array_almost_equal(sk_res.flatten(),
gs_res.flatten(),
decimal=1)
class TestPerformance(unittest.TestCase):
"""
The TestPerformance class is not a real unittest. It will always be true.
It does apply some benchmarking, which could be included into the testing
framework, as soon as the OrdinaryKriging class is finalized. From that
point on, new code should not harm the performance significantly.
"""
def setUp(self):
# define the target field
def func(x, y):
return np.sin(0.02 * np.pi * y) * np.cos(0.02 * np.pi * x)
# create a grid
self.grid_x, self.grid_y = np.mgrid[0:100:100j, 0:100:100j]
# sample the field
np.random.seed(42)
self.x = np.random.randint(100, size=300)
np.random.seed(1337)
self.y = np.random.randint(100, size=300)
self.z = func(self.x, self.y)
# build the Variogram and Kriging class
self.V = Variogram(list(zip(self.x, self.y)),
self.z,
model='gaussian',
n_lags=15,
maxlag=0.4)
self.ok = OrdinaryKriging(self.V,
min_points=2,
max_points=5,
perf=True)
def _run_benchmark(self, points):
xi = self.grid_x.flatten()[:points]
yi = self.grid_y.flatten()[:points]
# run
res = self.ok.transform(xi, yi)
self.ok.perf_dist *= 1000
self.ok.perf_mat *= 1000
self.ok.perf_solv *= 1000
print('Benchmarking OrdinaryKriging...')
print('-------------------------------')
print('Points:', points)
print('Solver:', self.ok.solver)
print('Mode:', self.ok.mode)
print('Build distance matrix: %.1f ms (%.4f ms each)' %
(np.sum(self.ok.perf_dist), np.std(self.ok.perf_dist)))
print('Build variogram matrix: %.1f ms (%.4f ms each)' %
(np.sum(self.ok.perf_mat), np.std(self.ok.perf_mat)))
print('Solve kriging matrix: %.1f ms (%.4f ms each)' %
(np.sum(self.ok.perf_solv), np.std(self.ok.perf_solv)))
print('---------------------------------------------')
def test_20points_exact(self):
self.ok.mode = 'exact'
self.ok.solver = 'inv'
self._run_benchmark(points=20)
def test_100points_exact(self):
self.ok.mode = 'exact'
self.ok.solver = 'inv'
self._run_benchmark(points=100)
def test_20points_estimate(self):
self.ok.mode = 'estimate'
self.ok.solver = 'inv'
self._run_benchmark(points=20)
def test_100points_estimate(self):
self.ok.mode = 'estimate'
self.ok.solver = 'inv'
self._run_benchmark(points=100)
