def test_special_models(self):
# matern converges to gaussian
model1 = Matern()
model1.set_arg_bounds(nu=[0, 101])
model1.nu = 100
model2 = Gaussian(rescale=0.5)
self.assertAlmostEqual(model1.variogram(1), model2.variogram(1))
self.assertAlmostEqual(model1.spectrum(1), model2.spectrum(1), 2)
# stable model gets unstable for alpha < 0.3
with self.assertWarns(AttributeWarning):
Stable(alpha=0.2)
with self.assertWarns(AttributeWarning):
TPLStable(alpha=0.2)
# corner case for JBessel model
with self.assertWarns(AttributeWarning):
JBessel(dim=3, nu=0.5)
def test_covmodel_class(self):
model_std = Gaussian(rescale=3, var=1.1, nugget=1.2, len_scale=1.3)
model_var = Gau_var(rescale=3, var=1.1, nugget=1.2, len_scale=1.3)
model_cov = Gau_cov(rescale=3, var=1.1, nugget=1.2, len_scale=1.3)
model_cor = Gau_cor(rescale=3, var=1.1, nugget=1.2, len_scale=1.3)
var = model_std.variogram(2.5)
cov = model_std.covariance(2.5)
corr = model_std.correlation(2.5)
cor = model_std.cor(2.5)
self.assertFalse(check_bounds(bounds=[0]))
self.assertFalse(check_bounds(bounds=[1, -1]))
self.assertFalse(check_bounds(bounds=[0, 1, 2, 3]))
self.assertFalse(check_bounds(bounds=[0, 1, "kk"]))
self.assertRaises(ValueError, model_std.set_arg_bounds, wrong_arg=[1])
self.assertRaises(ValueError,
model_std.set_arg_bounds,
wrong_arg=[-1, 1])
# checking some properties
model_par = Stable()
self.assertFalse(model_par.do_rotation)
self.assertEqual(len(model_par.arg), len(model_par.arg_list))
self.assertEqual(len(model_par.iso_arg), len(model_par.iso_arg_list))
self.assertEqual(len(model_par.arg), len(model_par.iso_arg) + 2)
self.assertEqual(len(model_par.len_scale_vec), model_par.dim)
self.assertFalse(Gaussian() == Stable())
model_par.hankel_kw = {"N": 300}
self.assertEqual(model_par.hankel_kw["N"], 300)
# arg in bounds check
model_std.set_arg_bounds(var=[0.5, 1.5])
with self.assertRaises(ValueError):
model_std.var = 0.4
with self.assertRaises(ValueError):
model_std.var = 1.6
model_std.set_arg_bounds(var=[0.5, 1.5, "oo"])
with self.assertRaises(ValueError):
model_std.var = 0.5
with self.assertRaises(ValueError):
model_std.var = 1.5
with self.assertRaises(ValueError):
model_std.var_bounds = [1, -1]
with self.assertRaises(ValueError):
model_std.len_scale_bounds = [1, -1]
with self.assertRaises(ValueError):
model_std.nugget_bounds = [1, -1]
with self.assertRaises(ValueError):
model_std.anis_bounds = [1, -1]
# reset the standard model
model_std = Gaussian(rescale=3, var=1.1, nugget=1.2, len_scale=1.3)
# std value from bounds with neg. inf and finit bound
model_add = Mod_add()
model_add.set_arg_bounds(alpha=[-np.inf, 0])
self.assertAlmostEqual(model_add.alpha, -1)
# special treatment of anis check
model_std.set_arg_bounds(anis=[2, 4, "oo"])
self.assertTrue(np.all(np.isclose(model_std.anis, 3)))
# dim specific checks
with self.assertWarns(AttributeWarning):
Gau_fix(dim=1)
self.assertRaises(ValueError, Gaussian, dim=0)
self.assertRaises(ValueError, Gau_fix, latlon=True)
# check inputs
self.assertRaises(ValueError, model_std.percentile_scale, per=-1.0)
self.assertRaises(ValueError, Gaussian, anis=-1.0)
self.assertRaises(ValueError, Gaussian, len_scale=[1, -1])
self.assertRaises(ValueError, check_arg_in_bounds, model_std, "wrong")
self.assertWarns(AttributeWarning, Gaussian, wrong_arg=1.0)
with self.assertWarns(AttributeWarning):
self.assertRaises(ValueError, Gaussian, len_rescaled=1.0)
# check correct subclassing
with self.assertRaises(TypeError):
class Gau_err(CovModel):
pass
self.assertAlmostEqual(var, model_var.variogram(2.5))
self.assertAlmostEqual(var, model_cov.variogram(2.5))
self.assertAlmostEqual(var, model_cor.variogram(2.5))
self.assertAlmostEqual(cov, model_var.covariance(2.5))
self.assertAlmostEqual(cov, model_cov.covariance(2.5))
self.assertAlmostEqual(cov, model_cor.covariance(2.5))
self.assertAlmostEqual(corr, model_var.correlation(2.5))
self.assertAlmostEqual(corr, model_cov.correlation(2.5))
self.assertAlmostEqual(corr, model_cor.correlation(2.5))
self.assertAlmostEqual(cor, model_var.cor(2.5))
self.assertAlmostEqual(cor, model_cov.cor(2.5))
self.assertAlmostEqual(cor, model_cor.cor(2.5))
