def test_eigenvectors_left_stats(bhmm_fixture):
samples = np.array(
[m.transition_model.eigenvectors_left() for m in bhmm_fixture.bhmm])
# mean
mean = samples.mean(axis=0)
# test shape and consistency
assert np.array_equal(mean.shape,
(bhmm_fixture.n_states, bhmm_fixture.n_states))
assert np.sign(mean[0, 0]) == np.sign(mean[0, 1])
assert np.sign(mean[1, 0]) != np.sign(mean[1, 1])
# std
std = samples.std(axis=0)
# test shape
assert np.array_equal(std.shape,
(bhmm_fixture.n_states, bhmm_fixture.n_states))
# conf
L, R = confidence_interval(samples)
# test shape
assert np.array_equal(L.shape,
(bhmm_fixture.n_states, bhmm_fixture.n_states))
assert np.array_equal(R.shape,
(bhmm_fixture.n_states, bhmm_fixture.n_states))
# test consistency
tol = 1e-12
assert np.all(L - tol <= mean)
assert np.all(R + tol >= mean)
def _eigenvectors_right_stats(self, msm, tol=1e-12):
samples = np.array([s.eigenvectors_right() for s in msm.samples])
# mean
mean = samples.mean(axis=0)
# test shape and consistency
np.testing.assert_equal(mean.shape, (self.n_states, self.n_states))
assert np.sign(mean[0, 0]) == np.sign(mean[1, 0])
assert np.sign(mean[0, 1]) != np.sign(mean[1, 1])
# std
std = samples.std(axis=0)
# test shape
assert np.array_equal(std.shape, (self.n_states, self.n_states))
# conf
L, R = confidence_interval(samples)
# test shape
assert np.array_equal(L.shape, (self.n_states, self.n_states))
assert np.array_equal(R.shape, (self.n_states, self.n_states))
# test consistency
assert np.all(L - tol <= mean)
assert np.all(R + tol >= mean)
def _eigenvalues_stats(self, msm, tol=1e-12):
# mean
samples = np.array([s.eigenvalues() for s in msm.samples])
mean = samples.mean(axis=0)
# test shape and consistency
assert np.array_equal(mean.shape, (self.n_states,))
assert np.isclose(mean[0], 1)
assert np.all(mean[1:] < 1.0)
# std
std = samples.std(axis=0)
# test shape
assert np.array_equal(std.shape, (self.n_states,))
# conf
L, R = confidence_interval(samples)
# test shape
assert np.array_equal(L.shape, (self.n_states,))
assert np.array_equal(R.shape, (self.n_states,))
# test consistency
assert np.all(L - tol <= mean)
assert np.all(R + tol >= mean)
def _transition_matrix_stats(self, msm):
import deeptime.markov.tools.analysis as msmana
# mean
Ps = np.array([s.transition_matrix for s in msm.samples])
Pmean = Ps.mean(axis=0)
# test shape and consistency
assert np.array_equal(Pmean.shape, (self.n_states, self.n_states))
assert msmana.is_transition_matrix(Pmean)
# std
Pstd = Ps.std(axis=0)
# test shape
assert np.array_equal(Pstd.shape, (self.n_states, self.n_states))
# conf
L, R = confidence_interval(Ps)
# test shape
assert np.array_equal(L.shape, (self.n_states, self.n_states))
assert np.array_equal(R.shape, (self.n_states, self.n_states))
# test consistency
assert np.all(L <= Pmean)
assert np.all(R >= Pmean)
def _stationary_distribution_stats(self, msm, tol=1e-12):
samples = np.array([s.stationary_distribution for s in msm.samples])
# mean
mean = samples.mean(axis=0)
# test shape and consistency
assert np.array_equal(mean.shape, (self.n_states,))
assert np.isclose(mean.sum(), 1.0)
assert np.all(mean > 0.0)
assert np.max(np.abs(mean[0] - mean[1])) < 0.05
# std
std = samples.std(axis=0)
# test shape
assert np.array_equal(std.shape, (self.n_states,))
# conf
L, R = confidence_interval(samples)
# test shape
assert np.array_equal(L.shape, (self.n_states,))
assert np.array_equal(R.shape, (self.n_states,))
# test consistency
assert np.all(L - tol <= mean)
assert np.all(R + tol >= mean)
def test_stationary_distribution_stats(self):
samples = np.array([m.transition_model.stationary_distribution for m in self.bhmm])
tol = 1e-12
# mean
mean = samples.mean(axis=0)
# test shape and consistency
assert np.array_equal(mean.shape, (self.n_states,))
assert np.isclose(mean.sum(), 1.0)
assert np.all(mean > 0.0)
assert np.max(np.abs(mean[0] - mean[1])) < 0.05
# std
std = samples.std(axis=0)
# test shape
assert np.array_equal(std.shape, (self.n_states,))
# conf
L, R = confidence_interval(samples)
# test shape
assert np.array_equal(L.shape, (self.n_states,))
assert np.array_equal(R.shape, (self.n_states,))
# test consistency
assert np.all(L - tol <= mean)
assert np.all(R + tol >= mean)
def test_confidence_interval(conf):
g = np.random.default_rng(seed=42)
data = g.normal(size=(300000, ))
l, r = confidence_interval(data, conf=conf)
data_inside = data[(data >= l) & (data <= r)]
np.testing.assert_almost_equal(data_inside.size / data.size, conf)