def test_compute_09(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 60).reshape(15, 4)
gradients = np.random.uniform(-1, 1, 180).reshape(15, 3, 4)
weights = np.ones((15, 1)) / 15
metric = np.diag(2 * np.ones(3))
ss = KernelActiveSubspaces(dim=2,
n_features=4,
method='exact',
n_boot=49)
ss.fit(inputs=inputs,
gradients=gradients,
weights=weights,
metric=metric)
true_evects = np.array(
[[0.00126244, 0.99791389, 0.02926469, 0.05753138],
[0.04385229, -0.05833941, 0.78953331, 0.60935265],
[-0.99902507, -0.001436, 0.03167332, 0.03071887],
[0.00492877, -0.02761026, -0.61219077, 0.79021253]])
np.testing.assert_array_almost_equal(true_evects, ss.evects)
def test_plot_sufficient_summary_02(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces()
ss.compute(inputs=inputs,
gradients=gradients,
weights=weights,
method='exact',
nboot=49,
n_features=8,
feature_map=None)
ss.partition(3)
with self.assertRaises(ValueError):
ss.plot_sufficient_summary(10, 10)
def test_plot_sufficient_summary_02(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces(dim=3, n_features=8, n_boot=5)
ss.fit(inputs=inputs, gradients=gradients, weights=weights)
with self.assertRaises(ValueError):
ss.plot_sufficient_summary(10, 10)
def test_plot_eigenvalues_03(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces(dim=2, n_features=8, n_boot=5)
ss.fit(inputs=inputs, gradients=gradients, weights=weights)
with assert_plot_figures_added():
ss.plot_eigenvalues(n_evals=3, figsize=(7, 7), title='Eigenvalues')
def test_plot_sufficient_summary_03(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces()
ss.compute(inputs=inputs,
gradients=gradients,
weights=weights,
method='exact',
nboot=49,
n_features=8,
feature_map=None)
ss.partition(2)
with assert_plot_figures_added():
ss.plot_sufficient_summary(
np.random.uniform(-1, 1, 100).reshape(25, 4),
np.random.uniform(-1, 1, 25).reshape(-1, 1))
def test_cross_validation_run_02(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 10).reshape(5, 2)
outputs = np.random.uniform(0, 5, 10).reshape(5, 2)
gradients = np.random.uniform(-1, 1, 20).reshape(5, 2, 2)
fm = FeatureMap(distr='laplace',
bias=np.random.uniform(-1, 1, 3),
input_dim=2,
n_features=3,
params=np.zeros(1),
sigma_f=outputs.var())
ss = KernelActiveSubspaces(dim=1, feature_map=fm)
csv = CrossValidation(inputs=inputs,
outputs=outputs,
gradients=gradients,
folds=2,
subspace=ss)
true_value = (2.26333743325053, 0.43902733603381605)
np.testing.assert_array_almost_equal(csv.run(), true_value)
def test_plot_eigenvectors_03(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces(dim=2,
n_features=5,
method='exact',
n_boot=5)
ss.fit(inputs=inputs, gradients=gradients, weights=weights)
with assert_plot_figures_added():
ss.plot_eigenvectors(n_evects=2,
figsize=(5, 8),
labels=[r'$x$', r'$y$', 'q', r'$r$', r'$z$'])
def test_plot_sufficient_summary_03(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces(dim=2,
n_features=8,
method='exact',
n_boot=5)
ss.fit(inputs=inputs, gradients=gradients, weights=weights)
with assert_plot_figures_added():
ss.plot_sufficient_summary(
np.random.uniform(-1, 1, 100).reshape(25, 4),
np.random.uniform(-1, 1, 25).reshape(-1, 1))
def test_forward_01(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 30).reshape(15, 1, 2)
inputs = np.random.uniform(-1, 1, 30).reshape(15, 2)
weights = np.ones((15, 1)) / 15
ss = KernelActiveSubspaces()
ss.compute(inputs=inputs,
gradients=gradients,
weights=weights,
method='exact',
nboot=49,
n_features=4,
feature_map=None)
ss.partition(2)
active = ss.forward(np.random.uniform(-1, 1, 4).reshape(2, 2))[0]
true_active = np.array([[1.34199032, 0.02509303],
[1.55021982, -0.29461026]])
np.testing.assert_array_almost_equal(true_active, active)
def test_plot_eigenvalues_03(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 200).reshape(50, 1, 4)
inputs = np.random.uniform(-1, 1, 200).reshape(50, 4)
weights = np.ones((50, 1)) / 50
ss = KernelActiveSubspaces()
ss.compute(inputs=inputs,
gradients=gradients,
weights=weights,
method='exact',
nboot=49,
n_features=8,
feature_map=None)
with assert_plot_figures_added():
ss.plot_eigenvalues(n_evals=3, figsize=(7, 7), title='Eigenvalues')
def test_forward_02(self):
np.random.seed(42)
gradients = np.random.uniform(-1, 1, 30).reshape(15, 1, 2)
inputs = np.random.uniform(-1, 1, 30).reshape(15, 2)
weights = np.ones((15, 1)) / 15
ss = KernelActiveSubspaces()
ss.compute(inputs=inputs,
gradients=gradients,
weights=weights,
method='exact',
nboot=49,
n_features=4,
feature_map=None)
ss.partition(2)
inactive = ss.forward(np.random.uniform(-1, 1, 4).reshape(2, 2))[1]
print(inactive)
true_inactive = np.array([[-0.47449407, 0.51271165],
[-0.27475082, 0.36433068]])
np.testing.assert_array_almost_equal(true_inactive, inactive)
def test_average_rrmse_01(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 10).reshape(5, 2)
outputs = np.random.uniform(0, 5, 10).reshape(5, 2)
gradients = np.random.uniform(-1, 1, 20).reshape(5, 2, 2)
fm = FeatureMap(distr='laplace',
bias=np.random.uniform(-1, 1, 3),
input_dim=2,
n_features=3,
params=np.zeros(1),
sigma_f=outputs.var())
ss = KernelActiveSubspaces(dim=1, feature_map=fm)
csv = CrossValidation(inputs=inputs,
outputs=outputs,
gradients=gradients,
folds=2,
subspace=ss)
best = [0.1, np.zeros((3, 2))]
hyperparams = np.array([-1.])
score = average_rrmse(hyperparams, csv, best, resample=1)
np.testing.assert_equal(best[0], 0.1)
def test_brute(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 10).reshape(5, 2)
outputs = np.random.uniform(0, 5, 10).reshape(5, 2)
gradients = np.random.uniform(-1, 1, 20).reshape(5, 2, 2)
fm = FeatureMap(distr='laplace',
bias=np.random.uniform(-1, 1, 3),
input_dim=2,
n_features=3,
params=np.array([5.34265038]),
sigma_f=outputs.var())
ss = KernelActiveSubspaces(feature_map=fm)
csv = CrossValidation(inputs=inputs, outputs=outputs, gradients=gradients, folds=2, subspace=ss)
best = fm.tune_pr_matrix(func=average_rrmse,
bounds=[slice(-1, 1.2, 0.2) for i in range(1)],
args=(csv, ),
maxiter=10,
save_file=False)[1]
true = np.array([[0.03857183, -0.45825228], [-1.06057884, 0.9981594],
[1.01812996, 0.19529565]])
np.testing.assert_array_almost_equal(true, best)
def test_average_rrmse_02(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 10).reshape(5, 2)
outputs = np.random.uniform(0, 5, 10).reshape(5, 2)
gradients = np.random.uniform(-1, 1, 20).reshape(5, 2, 2)
fm = FeatureMap(distr='laplace',
bias=np.random.uniform(-1, 1, 3),
input_dim=2,
n_features=3,
params=np.zeros(1),
sigma_f=outputs.var())
ss = KernelActiveSubspaces(dim=1, feature_map=fm, method='exact')
csv = CrossValidation(inputs=inputs,
outputs=outputs,
gradients=gradients,
folds=2,
subspace=ss)
best = [0.1, np.zeros((3, 2))]
hyperparams = np.array([-1.])
score = average_rrmse(hyperparams, best, csv, verbose=True, resample=1)
true = 2.1357048222078703
np.testing.assert_array_almost_equal(score, true)
def test_brute(self):
np.random.seed(42)
inputs = np.random.uniform(-1, 1, 10).reshape(5, 2)
outputs = np.random.uniform(0, 5, 10).reshape(5, 2)
gradients = np.random.uniform(-1, 1, 20).reshape(5, 2, 2)
fm = FeatureMap(distr='laplace',
bias=np.random.uniform(-1, 1, 3),
input_dim=2,
n_features=3,
params=np.array([5.34265038]),
sigma_f=outputs.var())
ss = KernelActiveSubspaces(dim=1, feature_map=fm)
csv = CrossValidation(inputs=inputs,
outputs=outputs,
gradients=gradients,
folds=2,
subspace=ss)
best = fm.tune_pr_matrix(func=average_rrmse,
bounds=[slice(-2, 1, 0.2) for i in range(1)],
fn_args={'csv':csv},
maxiter=10,
save_file=False)[1]
true = np.array([[0., 0.], [0., 0.], [0., 0.]])
np.testing.assert_array_almost_equal(true, best)
# this is the number of parameters to tune after
n_params = 1
# sample the bias term
b = np.random.uniform(0, 2 * np.pi, n_features)
# define the feature map
fm = FeatureMap(distr='laplace',
bias=b,
input_dim=input_dim,
n_features=n_features,
params=np.zeros(n_params),
sigma_f=f.var())
# instantiate a KernelActiveSubspaces object with associated feature map
kss = KernelActiveSubspaces(feature_map=fm, dim=1, n_features=n_features)
# number of folds for the cross-validation algorithm
folds = 3
verbose = True
# Skip if bias and projection matrix are loaded
csv = CrossValidation(inputs=xx,
outputs=f.reshape(-1, 1),
gradients=df.reshape(n_samples, 1, input_dim),
folds=folds,
subspace=kss)
best = fm.tune_pr_matrix(func=average_rrmse,
bounds=[slice(-2, 0., 0.2) for i in range(n_params)],
fn_args={
def test_compute_01(self):
ss = KernelActiveSubspaces(dim=2)
with self.assertRaises(TypeError):
ss.fit()
def test_init_pseudo_gradients(self):
ss = KernelActiveSubspaces(dim=2)
self.assertIsNone(ss.pseudo_gradients)
def test_init_features(self):
ss = KernelActiveSubspaces(dim=2)
self.assertIsNone(ss.features)
def test_init_evects(self):
ss = KernelActiveSubspaces(dim=2)
self.assertIsNone(ss.evects)
def test_init_cov_matrix(self):
ss = KernelActiveSubspaces()
self.assertIsNone(ss.cov_matrix)
def test_plot_sufficient_summary_01(self):
ss = KernelActiveSubspaces(dim=2)
with self.assertRaises(TypeError):
ss.plot_sufficient_summary(10, 10)
def test_init_feature_map(self):
ss = KernelActiveSubspaces()
self.assertIsNone(ss.feature_map)
def test_init_dim(self):
ss = KernelActiveSubspaces(dim=2)
self.assertEqual(ss.dim, 2)
def test_plot_eigenvectors_01(self):
ss = KernelActiveSubspaces(dim=2)
with self.assertRaises(TypeError):
ss.plot_eigenvectors(figsize=(7, 7), title='Eigenvalues')
def test_compute_01(self):
ss = KernelActiveSubspaces()
with self.assertRaises(ValueError):
ss.compute()
def test_init_subs_br(self):
ss = KernelActiveSubspaces(dim=2)
self.assertIsNone(ss.subs_br)
def test_init_evals_br(self):
ss = KernelActiveSubspaces()
self.assertIsNone(ss.evals_br)
def test_init_dim(self):
ss = KernelActiveSubspaces()
self.assertIsNone(ss.dim)
def test_init_W2(self):
ss = KernelActiveSubspaces(dim=2)
self.assertIsNone(ss.W2)
