def test_hkernel():
""" test the hrf computation
"""
tr = 2.0
h = _hrf_kernel('spm', tr)
assert_almost_equal(h[0], spm_hrf(tr))
assert len(h) == 1
h = _hrf_kernel('spm + derivative', tr)
assert_almost_equal(h[1], spm_time_derivative(tr))
assert len(h) == 2
h = _hrf_kernel('spm + derivative + dispersion', tr)
assert_almost_equal(h[2], spm_dispersion_derivative(tr))
assert len(h) == 3
h = _hrf_kernel('glover', tr)
assert_almost_equal(h[0], glover_hrf(tr))
assert len(h) == 1
h = _hrf_kernel('glover + derivative', tr)
assert_almost_equal(h[1], glover_time_derivative(tr))
assert_almost_equal(h[0], glover_hrf(tr))
assert len(h) == 2
h = _hrf_kernel('fir', tr, fir_delays=np.arange(4))
assert len(h) == 4
for dh in h:
assert_almost_equal(dh.sum(), 1.)
h = _hrf_kernel(None, tr)
assert len(h) == 1
assert_almost_equal(h[0], np.hstack((1, np.zeros(49))))
def test_hkernel():
""" test the hrf computation
"""
tr = 2.0
h = _hrf_kernel('spm', tr)
assert_almost_equal(h[0], spm_hrf(tr))
assert len(h) == 1
h = _hrf_kernel('spm + derivative', tr)
assert_almost_equal(h[1], spm_time_derivative(tr))
assert len(h) == 2
h = _hrf_kernel('spm + derivative + dispersion', tr)
assert_almost_equal(h[2], spm_dispersion_derivative(tr))
assert len(h) == 3
h = _hrf_kernel('glover', tr)
assert_almost_equal(h[0], glover_hrf(tr))
assert len(h) == 1
h = _hrf_kernel('glover + derivative', tr)
assert_almost_equal(h[1], glover_time_derivative(tr))
assert_almost_equal(h[0], glover_hrf(tr))
assert len(h) == 2
h = _hrf_kernel('glover + derivative + dispersion', tr)
assert len(h) == 3
assert_almost_equal(h[2], glover_dispersion_derivative(tr))
assert_almost_equal(h[1], glover_time_derivative(tr))
assert_almost_equal(h[0], glover_hrf(tr))
h = _hrf_kernel('fir', tr, fir_delays=np.arange(4))
assert len(h) == 4
for dh in h:
assert_almost_equal(dh.sum(), 1.)
h = _hrf_kernel(None, tr)
assert len(h) == 1
assert_almost_equal(h[0], np.hstack((1, np.zeros(49))))
with pytest.raises(ValueError,
match="Could not process custom HRF model provided."):
_hrf_kernel(lambda x: np.ones(int(x)), tr)
_hrf_kernel([lambda x, y, z: x + y + z], tr)
_hrf_kernel([lambda x: np.ones(int(x))] * 2, tr)
h = _hrf_kernel(lambda tr, ov: np.ones(int(tr * ov)), tr)
assert len(h) == 1
assert_almost_equal(h[0], np.ones(100))
h = _hrf_kernel([lambda tr, ov: np.ones(int(tr * ov))], tr)
assert len(h) == 1
assert_almost_equal(h[0], np.ones(100))
with pytest.raises(ValueError, match="is not a known hrf model."):
_hrf_kernel("foo", tr)
def test_glover_hrf():
""" test that the spm_hrf is correctly normalized and has correct length
"""
h = glover_hrf(2.0)
assert_almost_equal(h.sum(), 1)
assert len(h) == 800
h = glover_dispersion_derivative(2.0)
assert_almost_equal(h.sum(), 0)
assert len(h) == 800
def simulate_signal(onsets, conditions, TR=2, duration=None, icept=0, params_canon=None, params_tderiv=None,
params_deriv1=None, phi=None, std_noise=1, osf=100, rnd_seed=None, plot=True):
""" Simulates a somewhat realistic voxel signal with an associated design matrix. """
if rnd_seed is not None:
np.random.seed(rnd_seed)
conds = sorted(np.unique(conditions))
P = len(conds)
if duration is None:
duration = np.max(onsets) + 30
if params_canon is None:
params_canon = np.zeros(P)
# TMP FIX
params_tderiv = params_deriv1
if params_tderiv is None:
params_tderiv = np.zeros(P)
X = np.zeros((duration * osf, P))
for ons, con in zip(onsets, conditions):
X[int(ons * osf), conds.index(con)] = 1
hrf = glover_hrf(tr=1, oversampling=osf)
hrf_d = glover_time_derivative(tr=1, oversampling=osf)
t_orig = np.arange(0, duration, 1 / osf)
t_new = np.arange(0, duration, TR)
Xconv = np.zeros((t_orig.size, P * 2))
idx = 0
for i, this_hrf in enumerate([hrf, hrf_d]):
this_hrf /= this_hrf.max()
for ii in range(P):
Xconv[:, idx] = np.convolve(X[:, ii], this_hrf)[:t_orig.size]
idx += 1
Xconv = np.c_[np.ones(Xconv.shape[0]), Xconv]
params = np.r_[icept, params_canon, params_tderiv]
y = Xconv @ params
resampler = interp1d(t_orig, y)
y = resampler(t_new)
if phi is None:
noise_cov = std_noise ** 2 * np.eye(y.size)
else:
noise_cov = std_noise ** 2 * phi ** toeplitz(np.arange(y.size))
y = y + np.random.multivariate_normal(np.zeros(y.size), noise_cov)
Xconv_ds = np.zeros((t_new.size, Xconv.shape[1]))
for i in range(Xconv.shape[1]):
resampler = interp1d(t_orig, Xconv[:, i])
Xconv_ds[:, i] = resampler(t_new)
est_betas = np.linalg.lstsq(Xconv_ds, y, rcond=None)[0]
if plot:
plt.figure(figsize=(15, 5))
plt.plot(y)
plt.plot(Xconv_ds @ est_betas)
plt.xlim(0, y.size)
plt.legend(['y', 'y-hat'])
plt.show()
return y, Xconv_ds