def test_get_parameter():
cop = GaussianCopula(3)
cop[:] = gen_corr()
assert isinstance(cop[:], np.ndarray)
assert isinstance(cop[:, 0], np.ndarray)
assert isinstance(cop[0, 0], (float, int))
def test_set_parameter_value_error(category, value):
cop = GaussianCopula(3)
with pytest.raises(ValueError):
if category == 'full':
cop[:] = value
elif category == 'slice':
cop[:, 0] = value
def get_data(N, P, rng):
# Set seed
np.random.seed(rng + 100)
# Generate X with a little bit of correlation b/w the continuous variables (i.e. 0.3 Pearson coeffincient c.a.)
X = np.zeros((N, P))
# Generate Gaussian copula correlated uniforms
g_cop = GaussianCopula(dim=P)
mycova = np.ones((P, P))
for i in range(P):
for j in range(P):
mycova[i, j] = 0.3**(np.abs(i - j)) + np.where(i == j, 0, 0.1)
g_cop[:] = mycova
rv = g_cop.random(N)
# Generate correlated covariates
X[:, 0:5] = np.asarray([sts.norm.ppf(rv[:, i]) for i in range(5)]).T
X[:,
5:P] = np.asarray([sts.binom.ppf(rv[:, i], 1, 0.3)
for i in range(5, P)]).T
# Generate Z
und_lin = -0.5 + 0.2 * X[:,
0] + 0.1 * X[:,
1] + 0.4 * X[:,
20] + sts.uniform.rvs(
size=N) / 10
pscore = sts.norm.cdf(und_lin)
Z = sts.binom.rvs(1, pscore)
# Generate Y
mu = 3 + 1.5 * np.sin(np.pi * X[:, 0]) + 0.5 * (X[:, 1] - 0.5)**2 + 1.5 * (
2 - np.abs(X[:, 2])) + 1.5 * X[:, 3] * (X[:, 20] + 1)
ITE = 0.1 + 1 * np.abs(X[:, 0] - 1) * (X[:, 20] + 2)
sigma = np.std(mu) / 2
Y = mu + ITE * Z + sts.norm.rvs(0, sigma, N)
return Y, X, Z, ITE
def test_get_parameter_index_error():
cop = GaussianCopula(3)
cop[:] = gen_corr()
with pytest.raises(IndexError):
cop[0, 0, 0]
with pytest.raises(IndexError):
cop[{}]
def test_set_parameter():
corr = gen_corr()
cop = GaussianCopula(3)
cop[:] = corr
cop[:, 0] = corr[:, 0]
cop[1, :] = corr[1, :]
cop[:, :] = corr
cop[1, 2] = 0.5
cop[0] = 0.3
def test_reshape_data_decorator():
copula = GaussianCopula(2)
# test that scalar goes through fine
assert isinstance(array_io(lambda cop, x: x)(copula, 0.5), float)
assert isinstance(array_io(lambda cop, x: x, dim=1)(copula, 0.5), float)
# test that vector goes through okay
assert array_io(lambda cop, x: x, dim=1)(copula, [0.5, 0.2]).ndim == 1
# test that 1D array gets converted to 2D array
res = array_io(lambda cop, x: x, dim=2)(copula, np.zeros(2))
assert res.ndim == 2
# Non-optional input
with pytest.raises(AssertionError):
array_io(lambda cop, x: x, optional=False)(copula, None)
def test_reshape_data_decorator():
copula = GaussianCopula(2)
# test that array output with 1 element becomes scalar and is float
x1 = array_io(lambda cop, x: [1])(copula, [0.5, 0.5])
assert x1 == 1
assert isinstance(x1, float)
# test that vector output gets squeezed if one of the dimension is 1
for output in (
[[1, 2]],
[[1], [2]]
):
x2 = array_io(lambda cop, x: output)(copula, [0.5, 0.5])
assert x2.ndim == 1, "vector output must have squeezed"
# test that 2D output stays 2D
x3 = array_io(lambda cop, x: [[1, 2], [3, 4]])(copula, [0.5, 0.5])
assert x3.ndim == 2, "2D output should remain 2D"
# Non-optional input
with pytest.raises(ValueError):
array_io(lambda cop, x: x, optional=False)(copula, None)
def fitted_gaussian(residual_data):
dim = residual_data.shape[1]
cop = GaussianCopula(dim)
cop.fit(residual_data)
return cop
def copula(residual_data: np.ndarray):
dim = residual_data.shape[1]
cop = GaussianCopula(dim)
cop.params = target_params
return cop
def test_set_parameter_index_error(index):
cop = GaussianCopula(3)
with pytest.raises(IndexError):
cop[index] = 0.3
def test_fitted_parameters_match_target(residual_data: np.ndarray):
dim = residual_data.shape[1]
cop = GaussianCopula(dim)
cop.fit(residual_data)
assert_array_almost_equal(cop.params, target_params, 4)
def copula(residual_data: np.ndarray):
dim = residual_data.shape[1]
cop = GaussianCopula(dim)
cop.params = target_params
return cop