def test_sample_gaussian():
"""
Test sample generation from mixture.sample_gaussian where covariance
is diagonal, spherical and full
"""
n_features, n_samples = 2, 300
axis = 1
mu = rng.randint(10) * rng.rand(n_features)
cv = (rng.rand(n_features) + 1.0) ** 2
samples = mixture.sample_gaussian(
mu, cv, cvtype='diag', n_samples=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=1.3)
assert np.allclose(samples.var(axis), cv, atol=1.5)
# the same for spherical covariances
cv = (rng.rand() + 1.0) ** 2
samples = mixture.sample_gaussian(
mu, cv, cvtype='spherical', n_samples=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=1.5)
assert np.allclose(
samples.var(axis), np.repeat(cv, n_features), atol=1.5)
# and for full covariances
A = rng.randn(n_features, n_features)
cv = np.dot(A.T, A) + np.eye(n_features)
samples = mixture.sample_gaussian(
mu, cv, cvtype='full', n_samples=n_samples)
assert np.allclose(samples.mean(axis), mu, atol=1.3)
assert np.allclose(np.cov(samples), cv, atol=2.5)
def test_sample_gaussian():
# Test sample generation from mixture.sample_gaussian where covariance
# is diagonal, spherical and full
n_features, n_samples = 2, 300
axis = 1
mu = rng.randint(10) * rng.rand(n_features)
cv = (rng.rand(n_features) + 1.0) ** 2
samples = mixture.sample_gaussian(mu, cv, covariance_type="diag", n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(samples.var(axis), cv, atol=1.5))
# the same for spherical covariances
cv = (rng.rand() + 1.0) ** 2
samples = mixture.sample_gaussian(mu, cv, covariance_type="spherical", n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.5))
assert_true(np.allclose(samples.var(axis), np.repeat(cv, n_features), atol=1.5))
# and for full covariances
A = rng.randn(n_features, n_features)
cv = np.dot(A.T, A) + np.eye(n_features)
samples = mixture.sample_gaussian(mu, cv, covariance_type="full", n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(np.cov(samples), cv, atol=2.5))
# Numerical stability check: in SciPy 0.12.0 at least, eigh may return
# tiny negative values in its second return value.
from sklearn.mixture import sample_gaussian
x = sample_gaussian([0, 0], [[4, 3], [1, 0.1]], covariance_type="full", random_state=42)
assert_true(np.isfinite(x).all())
def test_sample_gaussian():
"""
Test sample generation from mixture.sample_gaussian where covariance
is diagonal, spherical and full
"""
n_features, n_samples = 2, 300
axis = 1
mu = rng.randint(10) * rng.rand(n_features)
cv = (rng.rand(n_features) + 1.0) ** 2
samples = mixture.sample_gaussian(
mu, cv, covariance_type='diag', n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(samples.var(axis), cv, atol=1.5))
# the same for spherical covariances
cv = (rng.rand() + 1.0) ** 2
samples = mixture.sample_gaussian(
mu, cv, covariance_type='spherical', n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.5))
assert_true(np.allclose(
samples.var(axis), np.repeat(cv, n_features), atol=1.5))
# and for full covariances
A = rng.randn(n_features, n_features)
cv = np.dot(A.T, A) + np.eye(n_features)
samples = mixture.sample_gaussian(
mu, cv, covariance_type='full', n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(np.cov(samples), cv, atol=2.5))
def test_sample_gaussian():
"""
Test sample generation from mixture.sample_gaussian where covariance
is diagonal, spherical and full
"""
n_features, n_samples = 2, 300
axis = 1
mu = rng.randint(10) * rng.rand(n_features)
cv = (rng.rand(n_features) + 1.0)**2
samples = mixture.sample_gaussian(mu,
cv,
covariance_type='diag',
n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(samples.var(axis), cv, atol=1.5))
# the same for spherical covariances
cv = (rng.rand() + 1.0)**2
samples = mixture.sample_gaussian(mu,
cv,
covariance_type='spherical',
n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.5))
assert_true(
np.allclose(samples.var(axis), np.repeat(cv, n_features), atol=1.5))
# and for full covariances
A = rng.randn(n_features, n_features)
cv = np.dot(A.T, A) + np.eye(n_features)
samples = mixture.sample_gaussian(mu,
cv,
covariance_type='full',
n_samples=n_samples)
assert_true(np.allclose(samples.mean(axis), mu, atol=1.3))
assert_true(np.allclose(np.cov(samples), cv, atol=2.5))
# Numerical stability check: in SciPy 0.12.0 at least, eigh may return
# tiny negative values in its second return value.
from sklearn.mixture import sample_gaussian
x = sample_gaussian([0, 0], [[4, 3], [1, .1]],
covariance_type='full',
random_state=42)
print(x)
assert_true(np.isfinite(x).all())
def _generate_sample_from_state(self, state, random_state=None):
if self._covariance_type == 'tied':
cv = self._covars_
else:
cv = self._covars_[state]
return sample_gaussian(self._means_[state], cv, self._covariance_type,
random_state=random_state)
def execute(self, activation):
self.activation = self.activation_fun(activation)
# self.activation = max(0.01, self.activation)
if self.t >= self.t_last_voc + 0 and self.activation > rand():
self.m = sample_gaussian(self.mean, self.covar)
self.t_last_voc = deepcopy(self.t)
else:
self.m = None
self.t += 1
return self.m
def _generate_sample_from_state(self, state, random_state=None):
if random_state is None:
random_state = self.random_state
random_state = check_random_state(random_state)
cur_means = self.means_[state]
cur_covs = self.covars_[state]
cur_weights = self.weights_[state]
i_gauss = random_state.choice(self.n_mix, p=cur_weights)
mean = cur_means[i_gauss]
if self.covariance_type == 'tied':
cov = cur_covs
else:
cov = cur_covs[i_gauss]
return sample_gaussian(mean, cov, self.covariance_type,
random_state=random_state)
def _generate_sample_from_state(self, state, random_state=None):
if random_state is None:
random_state = self.random_state
random_state = check_random_state(random_state)
cur_means = self.means_[state]
cur_covs = self.covars_[state]
cur_weights = self.weights_[state]
i_gauss = random_state.choice(self.n_mix, p=cur_weights)
mean = cur_means[i_gauss]
if self.covariance_type == 'tied':
cov = cur_covs
else:
cov = cur_covs[i_gauss]
return sample_gaussian(mean, cov, self.covariance_type,
random_state=random_state)
def infer(self, in_dims, out_dims, x):
if self.t < n_neighbors:
raise ExplautoBootstrapError
if in_dims == self.m_dims and out_dims == self.s_dims: # forward
dists, indexes = self.dataset.nn_x(x, k=1)
return self.dataset.get_y(indexes[0])
elif in_dims == self.s_dims and out_dims == self.m_dims: # inverse
if self.mode == 'explore':
if not self.to_explore:
self.current_goal = x
dists, indexes = self.dataset.nn_y(x, k=1)
self.mean_explore = self.dataset.get_x(indexes[0])
self.to_explore = self.n_explore
self.to_explore -= 1
return sample_gaussian(self.mean_explore, self.sigma_expl ** 2)
else: # exploit'
dists, indexes = self.dataset.nn_y(x, k=1)
return self.dataset.get_x(indexes[0])
else:
raise NotImplementedError("NearestNeighbor only implements forward"
"(M -> S) and inverse (S -> M) model, "
"not general prediction")
def infer(self, in_dims, out_dims, x):
if self.t < n_neighbors:
raise ExplautoBootstrapError
if in_dims == self.m_dims and out_dims == self.s_dims: # forward
dists, indexes = self.dataset.nn_x(x, k=1)
return self.dataset.get_y(indexes[0])
elif in_dims == self.s_dims and out_dims == self.m_dims: # inverse
if self.mode == 'explore':
if not self.to_explore:
self.current_goal = x
dists, indexes = self.dataset.nn_y(x, k=1)
self.mean_explore = self.dataset.get_x(indexes[0])
self.to_explore = self.n_explore
self.to_explore -= 1
return sample_gaussian(self.mean_explore, self.sigma_expl**2)
else: # exploit'
dists, indexes = self.dataset.nn_y(x, k=1)
return self.dataset.get_x(indexes[0])
else:
raise NotImplementedError("NearestNeighbor only implements forward"
"(M -> S) and inverse (S -> M) model, "
"not general prediction")
def sample(self):
return sample_gaussian(self.mean, self.covar)