def test_non_matching_length(self):
n = 100
data = [np.random.random(size=(n, 2)) for _ in range(3)]
weights = [np.random.random(n) for _ in range(3)]
weights[0] = weights[0][:-3]
with self.assertRaises(ValueError):
covariance_lagged(data=data, weights=weights, lag=1)
with self.assertRaises(ValueError):
covariance_lagged(data, weights=weights[:2])
def test_XX(self):
cc = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.mx)
assert np.allclose(cc.C00_, self.Mxx)
assert np.allclose(cc1.mean, self.mx)
assert np.allclose(cc1.C00_, self.Mxx)
def test_XXXY_sym_withmean(self):
# many passes
cc = covariance_lagged(data=self.data, remove_data_mean=False, c0t=True, lag=self.lag, reversible=True,
bessel=False, chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.m_sym)
np.testing.assert_allclose(cc.C00_, self.Mxx_sym)
np.testing.assert_allclose(cc.C0t_, self.Mxy_sym)
def test_XXXY_weightobj_sym_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data, remove_data_mean=True, c0t=True, lag=self.lag, reversible=True,
bessel=False, weights=self.wobj, chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.m_sym_wobj)
np.testing.assert_allclose(cc.C00_, self.Mxx0_sym_wobj)
np.testing.assert_allclose(cc.C0t_, self.Mxy0_sym_wobj)
def test_XY_weighted_meanconst(self):
cc = covariance_lagged(data=self.data, remove_constant_mean=self.mean_const, c0t=True, weights=self.wobj, lag=self.lag,
bessel=False, chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_c_wobj)
np.testing.assert_allclose(cc.mean_tau, self.my_c_wobj)
np.testing.assert_allclose(cc.C00_, self.Mxx_c_wobj)
np.testing.assert_allclose(cc.C0t_, self.Mxy_c_wobj)
def test_re_estimate_weight_types(self):
# check different types are allowed and re-estimation works
x = np.random.random((100, 2))
c = covariance_lagged(lag=1)
c.estimate(x, weights=1)
c.estimate(x, weights=1.0)
c.estimate(x, weights=None)
c.estimate(x, weights=x[:,0])
def test_XXXY_weightobj_withmean(self):
# many passes
cc = covariance_lagged(data=self.data, remove_data_mean=False, c0t=True, lag=self.lag, weights=self.wobj,
bessel=False, chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_wobj)
np.testing.assert_allclose(cc.mean_tau, self.my_wobj)
np.testing.assert_allclose(cc.C00_, self.Mxx_wobj)
np.testing.assert_allclose(cc.C0t_, self.Mxy_wobj)
def test_XX_removeconstantmean(self):
cc = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
remove_constant_mean=self.mean_constant,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
remove_constant_mean=self.mean_constant,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.mx_c)
assert np.allclose(cc.C00_, self.Mxx_c)
assert np.allclose(cc1.mean, self.mx_c)
assert np.allclose(cc1.C00_, self.Mxx_c)
def test_XXXY_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data, remove_data_mean=True, c0t=True, lag=self.lag, bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx)
np.testing.assert_allclose(cc.mean_tau, self.my)
np.testing.assert_allclose(cc.C00_, self.Mxx0)
np.testing.assert_allclose(cc.C0t_, self.Mxy0)
def test_XX_meanconst(self):
cc = covariance_lagged(data=self.data,
c0t=False,
remove_constant_mean=self.mean_const,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.mx_c_lag0)
assert np.allclose(cc.cov, self.Mxx_c_lag0)
def test_XX_removedatamean(self):
cc = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
remove_data_mean=True,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
c0t=False,
lag=self.tau,
remove_data_mean=True,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.mx)
assert np.allclose(cc.cov, self.Mxx0)
assert np.allclose(cc1.mean, self.mx)
assert np.allclose(cc1.cov, self.Mxx0)
def test_XX_withmean(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=False,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_lag0)
np.testing.assert_allclose(cc.cov, self.Mxx_lag0)
def test_XX_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=True,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.mx_lag0)
assert np.allclose(cc.cov, self.Mxx0_lag0)
def test_XX_weighted_meanconst(self):
cc = covariance_lagged(data=self.data,
c0t=False,
remove_constant_mean=self.mean_const,
weights=self.wobj,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_c_wobj_lag0)
np.testing.assert_allclose(cc.cov, self.Mxx_c_wobj_lag0)
def test_XX_weightobj_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=True,
weights=self.wobj,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_wobj_lag0)
np.testing.assert_allclose(cc.cov, self.Mxx0_wobj_lag0)
def test_XY_sym(self):
cc = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
reversible=True,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
reversible=True,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.msym)
assert np.allclose(cc.C00_, self.Mxx_sym)
assert np.allclose(cc.C0t_, self.Mxy_sym)
assert np.allclose(cc1.mean, self.msym)
assert np.allclose(cc1.C00_, self.Mxx_sym)
assert np.allclose(cc1.C0t_, self.Mxy_sym)
def test_XX_weightobj_withmean(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=False,
weights=self.wobj,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.mx_wobj_lag0)
assert np.allclose(cc.cov, self.Mxx_wobj_lag0)
def test_XY(self):
cc = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.mx)
assert np.allclose(cc.mean_tau, self.my)
assert np.allclose(cc.cov, self.Mxx)
assert np.allclose(cc.cov_tau, self.Mxy)
assert np.allclose(cc1.mean, self.mx)
assert np.allclose(cc1.mean_tau, self.my)
assert np.allclose(cc1.cov, self.Mxx)
assert np.allclose(cc1.cov_tau, self.Mxy)
def test_XX_meanconst(self):
cc = covariance_lagged(data=self.data,
c0t=False,
remove_constant_mean=self.mean_const,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_c_lag0)
np.testing.assert_allclose(cc.C00_, self.Mxx_c_lag0)
cc.column_selection = self.cols_2
cc.estimate(self.data)
np.testing.assert_allclose(cc.C00_, self.Mxx_c_lag0[:, self.cols_2])
def test_XY_meanconst(self):
cc = covariance_lagged(data=self.data,
remove_constant_mean=self.mean_const,
c0t=True,
lag=self.lag,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_c)
np.testing.assert_allclose(cc.mean_tau, self.my_c)
np.testing.assert_allclose(cc.cov, self.Mxx_c)
np.testing.assert_allclose(cc.cov_tau, self.Mxy_c)
def test_XY_sym_meanconst(self):
cc = covariance_lagged(data=self.data,
remove_constant_mean=self.mean_const,
c0t=True,
reversible=True,
lag=self.lag,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.m_c_sym)
assert np.allclose(cc.cov, self.Mxx_c_sym)
assert np.allclose(cc.cov_tau, self.Mxy_c_sym)
def test_XY_sym_removedatamean(self):
cc = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
reversible=True,
remove_data_mean=True,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
reversible=True,
remove_data_mean=True,
bessel=False,
weights=self.weight_object)
assert np.allclose(cc.mean, self.msym)
assert np.allclose(cc.cov, self.Mxx0_sym)
assert np.allclose(cc.cov_tau, self.Mxy0_sym)
assert np.allclose(cc1.mean, self.msym)
assert np.allclose(cc1.cov, self.Mxx0_sym)
assert np.allclose(cc1.cov_tau, self.Mxy0_sym)
def test_XX_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=True,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_lag0)
np.testing.assert_allclose(cc.C00_, self.Mxx0_lag0)
cc.column_selection = self.cols_2
cc.estimate(self.data)
np.testing.assert_allclose(cc.C00_, self.Mxx0_lag0[:, self.cols_2])
def test_XXXY_sym_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data,
remove_data_mean=True,
c0t=True,
lag=self.lag,
reversible=True,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.m_sym)
assert np.allclose(cc.cov, self.Mxx0_sym)
assert np.allclose(cc.cov_tau, self.Mxy0_sym)
def test_XXXY_withmean(self):
# many passes
cc = covariance_lagged(data=self.data,
remove_data_mean=False,
c0t=True,
lag=self.lag,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.mx)
assert np.allclose(cc.mean_tau, self.my)
assert np.allclose(cc.cov, self.Mxx)
assert np.allclose(cc.cov_tau, self.Mxy)
def test_XY_removeconstantmean(self):
cc = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
remove_constant_mean=self.mean_constant,
bessel=False,
weights="koopman")
cc1 = covariance_lagged(data=self.data,
lag=self.tau,
c0t=True,
remove_constant_mean=self.mean_constant,
bessel=False,
weights="koopman")
assert np.allclose(cc.mean, self.mx_c)
assert np.allclose(cc.mean_tau, self.my_c)
assert np.allclose(cc.C00_, self.Mxx_c)
assert np.allclose(cc.C0t_, self.Mxy_c)
assert np.allclose(cc1.mean, self.mx_c)
assert np.allclose(cc1.mean_tau, self.my_c)
assert np.allclose(cc1.C00_, self.Mxx_c)
assert np.allclose(cc1.C0t_, self.Mxy_c)
def test_XXXY_weightobj_sym_withmean(self):
# many passes
cc = covariance_lagged(data=self.data,
remove_data_mean=False,
c0t=True,
lag=self.lag,
reversible=True,
bessel=False,
weights=self.wobj,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.m_sym_wobj)
assert np.allclose(cc.cov, self.Mxx_sym_wobj)
assert np.allclose(cc.cov_tau, self.Mxy_sym_wobj)
def test_XXXY_weightobj_meanfree(self):
# many passes
cc = covariance_lagged(data=self.data,
remove_data_mean=True,
c0t=True,
lag=self.lag,
weights=self.wobj,
bessel=False,
chunksize=self.chunksize)
assert np.allclose(cc.mean, self.mx_wobj)
assert np.allclose(cc.mean_tau, self.my_wobj)
assert np.allclose(cc.cov, self.Mxx0_wobj)
assert np.allclose(cc.cov_tau, self.Mxy0_wobj)
def test_XX_weightobj_withmean(self):
# many passes
cc = covariance_lagged(data=self.data,
c0t=False,
remove_data_mean=False,
weights=self.wobj,
bessel=False,
chunksize=self.chunksize)
np.testing.assert_allclose(cc.mean, self.mx_wobj_lag0)
np.testing.assert_allclose(cc.C00_, self.Mxx_wobj_lag0)
cc.column_selection = self.cols_2
cc.estimate(self.data)
np.testing.assert_allclose(cc.C00_, self.Mxx_wobj_lag0[:, self.cols_2])
def test_weights_close_to_zero(self):
n = 1000
data = [np.random.random(size=(n, 2)) for _ in range(5)]
# create some artificial correlations
data[0][:,0] *= np.random.randint(n)
weights = [np.ones(n, dtype=np.float32) for _ in range(5)]
# omit the first trajectory by setting a weight close to zero.
weights[0][:] = 1E-44
cov = covariance_lagged(data, lag=3, weights=weights, chunksize=10)
assert np.all(cov.C00_ < 1)
