def test_sample_gaussian():
"""
Test sample generation from gmm.sample_gaussian where covariance
is diagonal, spherical and full
"""
n_dim, n_samples = 2, 300
axis = 1
mu = np.random.randint(10) * np.random.rand(n_dim)
cv = (np.random.rand(n_dim) + 1.0) ** 2
samples = gmm.sample_gaussian(mu, cv, cvtype='diag', n=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=0.3)
assert np.allclose(samples.var(axis), cv, atol=0.5)
# the same for spherical covariances
cv = (np.random.rand() + 1.0) ** 2
samples = gmm.sample_gaussian(mu, cv, cvtype='spherical', n=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=0.3)
assert np.allclose(samples.var(axis), np.repeat(cv, n_dim), atol=0.5)
# and for full covariances
A = np.random.randn(n_dim, n_dim)
cv = np.dot(A.T, A) + np.eye(n_dim)
samples = gmm.sample_gaussian(mu, cv, cvtype='full', n=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=0.3)
assert np.allclose(np.cov(samples), cv, atol=0.7)
def _test_sample_gaussian_diag(self, ndim, n=10000):
mu = np.random.randint(10) * np.random.rand(ndim)
cv = (np.random.rand(ndim) + 1.0) ** 2
samples = gmm.sample_gaussian(mu, cv, cvtype="diag", n=n)
if ndim > 1:
axis = 1
else:
axis = None
assert_array_almost_equal(samples.mean(axis), mu, decimal=1)
assert_array_almost_equal(samples.var(axis), cv, decimal=1)
def _test_sample_gaussian_full(self, ndim, n=10000):
mu = np.random.randint(10) * np.random.rand(ndim)
cv = _generate_random_spd_matrix(ndim)
samples = gmm.sample_gaussian(mu, cv, cvtype="full", n=n)
if ndim > 1:
axis = 1
else:
axis = None
assert_array_almost_equal(samples.mean(axis), mu, decimal=1)
assert_array_almost_equal(np.cov(samples), cv, decimal=1)