def test_local_linear_gradients_02(self):
np.random.seed(42)
inputs = np.random.uniform(-1.0, 1.0, size=(200, 2))
outputs = 2 - 5 * inputs[:, 0] + 4 * inputs[:, 1]
gradients = local_linear_gradients(inputs, outputs, n_neighbors=8)[0]
M = gradients.shape[0]
np.testing.assert_array_almost_equal(gradients,
np.tile(np.array([-5.0, 4.0]),
(M, 1)),
decimal=9)
def test_local_linear_gradients_05(self):
np.random.seed(42)
inputs = np.random.uniform(-1.0, 1.0, size=(10, 2))
outputs = 2 - 5 * inputs[:, 0] + 4 * inputs[:, 1]
with self.assertRaises(ValueError):
local_linear_gradients(inputs, outputs, n_neighbors=15)
asub1_.plot_eigenvectors(figsize=(6, 4), title=title)
asub1_.plot_sufficient_summary(x, f, figsize=(6, 4), title=title)
asub2_ = ActiveSubspaces(dim=2, method='exact', n_boot=100)
asub2_.fit(gradients=df_exact)
asub2_.plot_sufficient_summary(x, f, figsize=(6, 4), title=title)
# Error analysis
samples_lis = [2**i for i in range(6, 12)]
err_abs, err_rel = [], []
for n_samples in samples_lis:
X = inputs_uniform(n_samples, input_dim, lb, ub)
nor = Normalizer(lb, ub)
x = nor.fit_transform(X)
f = func(x)
ll_gradients, new_inputs = local_linear_gradients(inputs=x, outputs=f)
df = egrad(func)(new_inputs)
absdiff = (1 / n_samples) * np.sum(
np.linalg.norm(df - ll_gradients, axis=1))
err_abs += [absdiff]
err_rel += [
absdiff / ((1 / n_samples) * np.sum(np.linalg.norm(df, axis=1)))
]
fig, axes = plt.subplots(2, 1, sharex=True, figsize=(9, 7))
fig.suptitle('Error Analysis')
axes[0].scatter(samples_lis, err_abs)
axes[0].plot(samples_lis, err_abs)
axes[0].set_ylabel('Absolute error')
axes[1].scatter(samples_lis, err_rel)
axes[1].plot(samples_lis, err_rel)