def test_ajdc_expl_var(rndstate):
# Test get_src_expl_var
n_subjects, n_conditions, n_channels, n_times = 2, 2, 3, 256
X = rndstate.randn(n_subjects, n_conditions, n_channels, n_times)
ajdc = AJDC().fit(X)
n_matrices = 4
X_new = rndstate.randn(n_matrices, n_channels, n_times)
v = ajdc.get_src_expl_var(X_new)
assert v.shape == (n_matrices, ajdc.n_sources_)
with pytest.raises(ValueError): # not 3 dims
ajdc.get_src_expl_var(X_new[0])
with pytest.raises(ValueError): # unequal # of chans
ajdc.get_src_expl_var(rndstate.randn(n_matrices, n_channels + 1, 1))
def test_ajdc_transform_error(rndstate):
n_subjects, n_conditions, n_channels, n_times = 2, 2, 3, 256
X = rndstate.randn(n_subjects, n_conditions, n_channels, n_times)
ajdc = AJDC().fit(X)
n_matrices = 4
X_new = rndstate.randn(n_matrices, n_channels, n_times)
with pytest.raises(ValueError): # not 3 dims
ajdc.transform(X_new[0])
with pytest.raises(ValueError): # unequal # of chans
ajdc.transform(rndstate.randn(n_matrices, n_channels + 1, 1))
def test_ajdc_fit_error(rndstate):
n_conditions, n_channels, n_times = 3, 8, 512
ajdc = AJDC()
with pytest.raises(ValueError): # unequal # of conditions
ajdc.fit([
rndstate.randn(n_conditions, n_channels, n_times),
rndstate.randn(n_conditions + 1, n_channels, n_times),
])
with pytest.raises(ValueError): # unequal # of channels
ajdc.fit([
rndstate.randn(n_conditions, n_channels, n_times),
rndstate.randn(n_conditions, n_channels + 1, n_times),
])
def test_ajdc_fit_variable_input(rndstate):
n_subjects, n_cond, n_chan, n_times = 2, 2, 3, 256
X = rndstate.randn(n_subjects, n_cond, n_chan, n_times)
ajdc = AJDC()
# 3 subjects, same # conditions and channels, different # of times
X = [
rndstate.randn(n_cond, n_chan, n_times + rndstate.randint(500))
for _ in range(3)
]
ajdc.fit(X)
# 2 subjects, 2 conditions, same # channels, different # of times
X = [
[
rndstate.randn(n_chan, n_times + rndstate.randint(500))
for _ in range(2)
],
[
rndstate.randn(n_chan, n_times + rndstate.randint(500))
for _ in range(2)
],
]
ajdc.fit(X)
def test_ajdc_inverse_transform(rndstate):
n_subjects, n_conditions, n_channels, n_times = 2, 2, 3, 256
X = rndstate.randn(n_subjects, n_conditions, n_channels, n_times)
ajdc = AJDC().fit(X)
n_matrices = 4
X_new = rndstate.randn(n_matrices, n_channels, n_times)
Xt = ajdc.transform(X_new)
Xit = ajdc.inverse_transform(Xt)
assert_array_equal(Xit.shape, [n_matrices, n_channels, n_times])
with pytest.raises(ValueError): # not 3 dims
ajdc.inverse_transform(Xt[0])
with pytest.raises(ValueError): # unequal # of sources
shape = (n_matrices, ajdc.n_sources_ + 1, 1)
ajdc.inverse_transform(rndstate.randn(*shape))
Xit = ajdc.inverse_transform(Xt, supp=[ajdc.n_sources_ - 1])
assert Xit.shape == (n_matrices, n_channels, n_times)
with pytest.raises(ValueError): # not a list
ajdc.inverse_transform(Xt, supp=1)
def test_ajdc_fit(rndstate):
n_subjects, n_conditions, n_channels, n_times = 5, 3, 8, 512
X = rndstate.randn(n_subjects, n_conditions, n_channels, n_times)
ajdc = AJDC().fit(X)
assert ajdc.forward_filters_.shape == (ajdc.n_sources_, n_channels)
assert ajdc.backward_filters_.shape == (n_channels, ajdc.n_sources_)
def test_ajdc_init():
ajdc = AJDC(fmin=1, fmax=32, fs=64)
assert ajdc.window == 128
assert ajdc.overlap == 0.5
assert ajdc.dim_red is None
assert ajdc.verbose
sfreq=sfreq)
ch_info.set_montage('standard_1020')
signal = RawArray(signal_raw, ch_info, verbose=False)
signal.plot(duration=duration, start=0, n_channels=ch_count,
scalings={'eeg': 3e1}, color={'eeg': 'steelblue'},
title='Original EEG signal', show_scalebars=False)
###############################################################################
# AJDC: Second-Order Statistics (SOS)-based BSS, diagonalizing cospectra
# ----------------------------------------------------------------------
# Compute and diagonalize Fourier cospectral matrices between 1 and 32 Hz
window, overlap = sfreq, 0.5
fmin, fmax = 1, 32
ajdc = AJDC(window=window, overlap=overlap, fmin=fmin, fmax=fmax, fs=sfreq,
expl_var=0.99)
ajdc.fit(signal_raw[np.newaxis, np.newaxis, ...])
freqs = ajdc.freqs_
# Plot cospectra in channel space, after trace-normalization by frequency: each
# cospectrum, associated to a frequency, is a covariance matrix
plot_cospectra(ajdc._cosp_channels, freqs, ylabels=ch_names,
title='Cospectra, in channel space')
###############################################################################
# Plot diagonalized cospectra in source space
sr_count = ajdc.n_sources_
sr_names = ['S' + str(s).zfill(2) for s in range(sr_count)]
plot_cospectra(ajdc._cosp_sources, freqs, ylabels=sr_names,
n_channels=ch_count,
scalings={'eeg': 3e1},
color={'eeg': 'steelblue'},
title='Original EEG signal',
show_scalebars=False)
###############################################################################
# AJDC: Second-Order Statistics (SOS)-based BSS, diagonalizing cospectra
# ----------------------------------------------------------------------
# Compute and diagonalize Fourier cospectral matrices between 1 and 32 Hz
window, overlap = sfreq, 0.5
fmin, fmax = 1, 32
ajdc = AJDC(window=window,
overlap=overlap,
fmin=fmin,
fmax=fmax,
fs=sfreq,
dim_red={'max_cond': 100})
ajdc.fit(signal_raw[np.newaxis, np.newaxis, ...])
freqs = ajdc.freqs_
# Plot cospectra in channel space, after trace-normalization by frequency: each
# cospectrum, associated to a frequency, is a covariance matrix
plot_cospectra(ajdc._cosp_channels,
freqs,
ylabels=ch_names,
title='Cospectra, in channel space')
###############################################################################
# Plot diagonalized cospectra in source space
def test_AJDC():
"""Test AJDC"""
rs = np.random.RandomState(33)
n_subjects, n_conditions, n_channels, n_samples = 5, 3, 8, 512
X = rs.randn(n_subjects, n_conditions, n_channels, n_samples)
# Test Init
with pytest.raises(ValueError): # value out of bounds
AJDC(expl_var=0)
with pytest.raises(ValueError): # value out of bounds
AJDC(expl_var=1.1)
ajdc = AJDC(fmin=1, fmax=32, fs=64)
assert ajdc.window == 128
assert ajdc.overlap == 0.5
assert ajdc.expl_var == 0.999
assert ajdc.verbose
# Test fit
ajdc.fit(X)
assert ajdc.n_channels_ == n_channels
assert ajdc.n_sources_ <= n_channels
assert_array_equal(ajdc.forward_filters_.shape,
[ajdc.n_sources_, n_channels])
assert_array_equal(ajdc.backward_filters_.shape,
[n_channels, ajdc.n_sources_])
with pytest.raises(ValueError): # unequal # of conditions
ajdc.fit([rs.randn(n_conditions, n_channels, n_samples),
rs.randn(n_conditions + 1, n_channels, n_samples)])
with pytest.raises(ValueError): # unequal # of channels
ajdc.fit([rs.randn(n_conditions, n_channels, n_samples),
rs.randn(n_conditions, n_channels + 1, n_samples)])
# 3 subjects, same # conditions and channels, different # of samples
X = [rs.randn(n_conditions, n_channels, n_samples),
rs.randn(n_conditions, n_channels, n_samples + 200),
rs.randn(n_conditions, n_channels, n_samples + 500)]
ajdc.fit(X)
# 2 subjects, 2 conditions, same # channels, different # of samples
X = [[rs.randn(n_channels, n_samples),
rs.randn(n_channels, n_samples + 200)],
[rs.randn(n_channels, n_samples + 500),
rs.randn(n_channels, n_samples + 100)]]
ajdc.fit(X)
# Test transform
n_trials = 4
X = rs.randn(n_trials, n_channels, n_samples)
Xt = ajdc.transform(X)
assert_array_equal(Xt.shape, [n_trials, ajdc.n_sources_, n_samples])
with pytest.raises(ValueError): # not 3 dims
ajdc.transform(X[0])
with pytest.raises(ValueError): # unequal # of chans
ajdc.transform(rs.randn(n_trials, n_channels + 1, 1))
# Test inverse_transform
Xtb = ajdc.inverse_transform(Xt)
assert_array_equal(Xtb.shape, [n_trials, n_channels, n_samples])
with pytest.raises(ValueError): # not 3 dims
ajdc.inverse_transform(Xt[0])
with pytest.raises(ValueError): # unequal # of sources
ajdc.inverse_transform(rs.randn(n_trials, ajdc.n_sources_ + 1, 1))
Xtb = ajdc.inverse_transform(Xt, supp=[ajdc.n_sources_ - 1])
assert_array_equal(Xtb.shape, [n_trials, n_channels, n_samples])
with pytest.raises(ValueError): # not a list
ajdc.inverse_transform(Xt, supp=1)
# Test get_src_expl_var
v = ajdc.get_src_expl_var(X)
assert_array_equal(v.shape, [n_trials, ajdc.n_sources_])
with pytest.raises(ValueError): # not 3 dims
ajdc.get_src_expl_var(X[0])
with pytest.raises(ValueError): # unequal # of chans
ajdc.get_src_expl_var(rs.randn(n_trials, n_channels + 1, 1))
