def test_log_density_cens_norm_at_bounds(self):
mu = 4
sigma = 2
a = 3
b = 6
data = np.array([3, 6])
dist = CensoredNormalDistribution(mu=mu, sigma=sigma, lower=a, upper=b)
actual = dist.log_density(data)
assert all(i > 1 for i in actual)
def test_log_density_cens_norm(self):
mu = 4
sigma = 2
a = 3
b = 6
data = np.array([2, 7])
dist = CensoredNormalDistribution(mu=mu, sigma=sigma, lower=a, upper=b)
actual = dist.log_density(data)
expected = np.array([-9999, -9999])
assert (actual == expected).all()
def test_compare_censored_to_normal(self):
mu = 4
sigma = 2
a = 3
b = 6
data = np.array([3.5, 4, 5, 5.5])
dist = CensoredNormalDistribution(mu=mu, sigma=sigma, lower=a, upper=b)
norm = s.norm(loc=mu, scale=sigma)
actual = dist.log_density(data)
expected = norm.logpdf(data)
assert (actual == expected).all()
def test_estimate_parameters_cens_norm_neg_mean(self):
mu = -2
sigma = 1
a = -3
b = -1
data = np.random.normal(loc=mu, scale=sigma, size=1000)
data = np.array(list(filter(lambda x: x > a and x < b, data)))
dist = CensoredNormalDistribution(mu=mu, sigma=sigma, lower=a, upper=b)
actual = dist.estimate_parameters(data, np.ones((data.shape)))
new_mu = dist.mu
new_sigma = dist.sigma
assert isclose(mu, new_mu, abs_tol=0.5)
assert new_sigma > 0
def test_estimate_parameters_cens_norm(self):
mu = 4
sigma = 2
a = 3
b = 6
data = np.random.normal(loc=mu, scale=sigma, size=1000)
data = np.where(data <= a, a, data)
data = np.where(data >= b, b, data)
dist = CensoredNormalDistribution(mu=mu, sigma=sigma, lower=a, upper=b)
actual = dist.estimate_parameters(data, np.ones((data.shape)))
new_mu = dist.mu
new_sigma = dist.sigma
assert isclose(mu, new_mu, abs_tol=0.5)
assert new_sigma > 0
def test_censored_and_normal_p_values(self):
data1 = np.random.normal(loc=0, scale=1, size=10000)
data1 = np.where(data1 <= 0, 0, data1)
data1 = np.where(data1 >= 1, 1, data1)
data2 = np.random.normal(loc=3, scale=1, size=10000)
dist1 = CensoredNormalDistribution(mu=0.5, sigma=1, lower=0, upper=1)
dist2 = NormalDistribution(mu=3, sigma=1)
data = np.concatenate((data1, data2))
dist_list = [dist1, dist2]
mixture = Data_Rep(data, dist_list)
dist_weight = mixture.weights[0]
observed = []
for i in [-1, 1.5, 5]:
observed.append(mixture.get_p_value(i))
expected = [
0 + s.norm.sf(-4), 1 - (dist_weight + s.norm.cdf(-1.5)),
0 + s.norm.sf(2)
] #manually get p value here
for i in range(len(expected)):
assert isclose(expected[i], observed[i], abs_tol=0.05)
def test_censored_p_values(self):
data1 = np.random.normal(loc=0, scale=1, size=1000)
data1 = np.where(data1 <= 0, 0, data1)
data1 = np.where(data1 >= 1, 1, data1)
data2 = np.random.normal(loc=3, scale=1, size=1000)
data2 = np.where(data2 <= 2, 2, data2)
data2 = np.where(data2 >= 4, 4, data2)
dist1 = CensoredNormalDistribution(mu=0.5, sigma=1, lower=0, upper=1)
dist2 = CensoredNormalDistribution(mu=1.5, sigma=1, lower=2, upper=4)
# after dataRep, make sure that params are close to expected
data = np.concatenate((data1, data2))
dist_list = [dist1, dist2]
mixture = Data_Rep(data, dist_list)
observed = []
for i in [-1, 1.5, 5]:
observed.append(mixture.get_p_value(i))
expected = [1.0, 0.5, 0.0]
assert observed == expected