def generate_report(): econ_density = EconDensity() X, Y = econ_density.simulate(n_samples=1000) nke = NeighborKernelDensityEstimation() nke.fit_by_cv(X, Y) n_samples = 500 X_test = np.asarray([1 for _ in range(n_samples)]) Y_test = np.linspace(0, 8, num=n_samples) Z = nke.pdf(X_test, Y_test)
def test_NKDE_with_4d_gaussian(self):
mu = 5
std = 2.0
X = np.random.normal(loc=mu, scale=std, size=(4000, 2))
Y = np.random.normal(loc=mu, scale=std, size=(4000, 2))
model = NeighborKernelDensityEstimation(epsilon=0.3)
model.fit(X, Y)
y = np.random.uniform(low=[1.0, 1.0], high=[9.0, 9.0], size=(500, 2))
x = np.ones(shape=(500, 2)) * mu
p_est = model.pdf(x, y)
p_true = stats.multivariate_normal.pdf(y,
mean=np.ones(2) * mu,
cov=std**2)
self.assertLessEqual(np.mean(np.abs(p_true - p_est)), 0.1)
def test_NKDE_log_pdf(self):
X, Y = np.random.normal(size=(500, 2)), np.random.normal(size=(500, 2))
model = NeighborKernelDensityEstimation()
model.fit(X, Y)
x, y = np.random.normal(size=(100, 2)), np.random.normal(size=(100, 2))
prob = model.pdf(x, y)
log_prob = model.log_pdf(x, y)
self.assertLessEqual(np.mean(np.abs(prob - np.exp(log_prob))), 0.001)
def test_NKDE_loo_log_likelihood(self):
mu = 1
std = 1
X = np.random.normal(loc=mu, scale=std, size=(500, 2))
Y = np.random.normal(loc=mu, scale=std, size=(500, 2))
model = NeighborKernelDensityEstimation(epsilon=0.3)
model.fit(X, Y)
bw = np.array([0.5])
epsilon = 0.3
ll1 = model.loo_likelihood(bandwidth=bw, epsilon=epsilon)
bw = np.array([0.05])
epsilon = 0.1
ll2 = model.loo_likelihood(bandwidth=bw, epsilon=epsilon)
self.assertGreater(ll1, ll2)
def test_NKDE_with_2d_gaussian(self):
X = np.random.uniform(-1, 1, size=4000)
Y = (2 + X) * np.random.normal(size=4000) + 2 * X
for weighted in [True, False]:
model = NeighborKernelDensityEstimation(epsilon=0.3,
weighted=weighted)
model.fit(X, Y)
y = np.linspace(-5, 5, num=100)
x = np.ones(100) * 0
p_est = model.pdf(x, y)
p_true = norm.pdf(y, loc=0, scale=2)
self.assertLessEqual(np.mean(np.abs(p_true - p_est)), 0.1)
y = np.linspace(-5, 5, num=100)
x = -np.ones(100) * 0.5
p_est = model.pdf(x, y)
p_true = norm.pdf(y, loc=-1, scale=1.5)
self.assertLessEqual(np.mean(np.abs(p_true - p_est)), 0.1)
def test_NKDE_param_selection(self):
mu = 5
std = 2
X = np.random.normal(loc=mu, scale=std, size=(500, 2))
Y = np.random.normal(loc=mu, scale=std, size=(500, 2))
model1 = NeighborKernelDensityEstimation('NKDE',
2,
2,
epsilon=0.1,
bandwidth=0.3,
param_selection=None)
model1.fit(X, Y)
self.assertAlmostEqual(model1.epsilon, 0.1)
model2 = NeighborKernelDensityEstimation(
'NKDE',
2,
2,
epsilon=0.1,
bandwidth=0.3,
param_selection='normal_reference')
model2.fit(X, Y)
self.assertAlmostEqual(model1.epsilon, 0.1)
self.assertGreaterEqual(np.mean(model2.bandwidth - model1.bandwidth),
0.0)
model3 = NeighborKernelDensityEstimation('NKDE',
2,
2,
epsilon=0.1,
bandwidth=0.3,
param_selection='cv_ml')
model3.fit(X, Y)
self.assertNotEqual(model3.epsilon, 0.1)
X_test, Y_test = np.random.normal(loc=mu, scale=std,
size=(2000, 2)), np.random.normal(
loc=mu,
scale=std,
size=(2000, 2))
score1 = model1.score(X_test, Y_test)
score2 = model2.score(X_test, Y_test)
score3 = model3.score(X_test, Y_test)
print(score1, score2, score3)
self.assertGreaterEqual(score2, score1)
self.assertGreaterEqual(score3, score2 - 0.1)