def test_draw_gaussian_samples(self):
""" Tests for draw gaussian samples. """
self.report(
'draw_gaussian_samples. Probabilistic test, could fail at times')
num_samples = 10000
num_pts = 3
mu = list(range(num_pts))
K = np.random.normal(size=(num_pts, num_pts))
K = K.dot(K.T)
samples = general_utils.draw_gaussian_samples(num_samples, mu, K)
sample_mean = samples.mean(axis=0)
sample_centralised = samples - sample_mean
sample_covar = sample_centralised.T.dot(
sample_centralised) / num_samples
mean_tol = 4 * np.linalg.norm(mu) / np.sqrt(num_samples)
covar_tol = 4 * np.linalg.norm(K) / np.sqrt(num_samples)
mean_err = np.linalg.norm(mu - sample_mean)
covar_err = np.linalg.norm(K - sample_covar)
self.report(
'Mean error (tol): ' + str(mean_err) + ' (' + str(mean_tol) + ')',
'test_result')
self.report(
'Cov error (tol): ' + str(covar_err) + ' (' + str(covar_tol) + ')',
'test_result')
assert mean_err < mean_tol
assert covar_err < covar_tol
def draw_samples(self,
num_samples,
X_test=None,
mean_vals=None,
covar=None):
""" Draws num_samples samples at returns their values at X_test. """
if X_test is not None:
mean_vals, covar = self.eval(X_test, 'covar')
return draw_gaussian_samples(num_samples, mean_vals, covar)
def draw_samples_with_hallucinated_observations(self, num_samples, X_test,
X_halluc):
""" Draws samples with hallucinated observations. """
mean_vals, aug_covar = self.eval_with_hallucinated_observations(X_test,
X_halluc, uncert_form='covar')
return draw_gaussian_samples(num_samples, mean_vals, aug_covar)