def pca_component_plot(component_ix: int, rotation_matrix: np.array,
stimXr: np.array, respXr: np.array,
stim_epochs: mne.Epochs, response_epochs: mne.Epochs):
'''Make a pretty plot.
Pars:
- component_ix: ID of the component to plot
- rotation_matrix
- stimXr: Rotated stimulus-locked data
- respXr: Rotated response-locked data
- stim_epochs, response_epochs: The original epoch objects
'''
fig, axes = plt.subplots(1,
3,
figsize=(18, 3),
gridspec_kw={'width_ratios': [1, 4, 4]})
topomap(rotation_matrix[:, component_ix],
response_epochs.info,
axes=axes[0])
axes[0].set_title('Component %i' % (component_ix + 1))
axes = axes[1:]
Xs = [stimXr, respXr]
timeses = [stim_epochs.times, response_epochs.times]
xlabs = ['Time from stimulus', 'Time to response']
for ax, X, times, xlab in zip(axes, Xs, timeses, xlabs):
plt.sca(ax)
eegf.plot_mean_sem(X[:, component_ix] * million, times)
plt.hlines(0, *plt.xlim(), linestyle='dashed')
plt.vlines(0, *plt.ylim(), linestyle='dashed')
plt.xlabel(xlab)
ax.set_yticklabels([])
plt.tight_layout()
return fig
def do_threeway_comparison(epochs,
ch,
by_subject=True,
agg_func=mean_by_subject,
crop=None,
baseline=None,
title=None,
ax=None,
legend=True,
neg_up=True):
# labels = ['Easy choice', 'Difficult choice', 'Guess']
if ax is not None:
plt.sca(ax)
es = get_condition_epochs(epochs, crop=crop, baseline=baseline)
labs = ['Easy', 'Difficult', 'Guess']
for e, l in zip(es, labs):
if by_subject:
X = agg_func(e)
else:
X = e.get_data()
if type(ch) == list and len(ch) == 2:
X = million * (X[:, ch[0]] - X[:, ch[1]])
else:
X = X[:, ch] * million
eegf.plot_mean_sem(X, e.times, label=l)
if neg_up:
eegf.flipy()
if legend:
plt.legend()
def do_twoway_comparison(epochs,
ch,
variable,
labels,
crop=None,
baseline=None,
title=None,
by_subject=True,
neg_up=True):
E = epochs.copy()
if crop is not None:
E = E.crop(*crop)
if baseline is not None:
E = E.apply_baseline(baseline)
e0 = E['%s == 0' % variable]
e1 = E['%s == 1' % variable]
es = [e0, e1]
for e, l in zip(es, labels):
if by_subject:
X = mean_by_subject(e)[:, ch] * million
else:
X = e.get_data()[:, ch] * million
eegf.plot_mean_sem(X, e.times, label=l)
if neg_up:
eegf.flipy()
plt.legend()
def do_rt_comparison(epochs, ch, drop_direction, bins=6, se=False):
'''drop direction = 1 for trial-locked epochs, -1 for response-locked'''
E = epochs.copy()
qs = pd.qcut(E.metadata['rt'], bins)
t0 = E.time_as_index(0)[0]
for i, q in enumerate(np.sort(qs.unique())):
ix = qs == q
e = E[ix]
lbl = '%.2f < RT < %.2f' % (q.left, q.right)
X = e.get_data()[:, ch] * million
if drop_direction == 1:
ti = e.time_as_index(q.right)[0]
if ti > 0:
X[:, ti:] = np.nan
else:
ti = e.time_as_index(-q.right)[0]
if ti > 0:
X[:, :ti] = np.nan
if se:
eegf.plot_mean_sem(X, e.times, label=lbl)
else:
plt.plot(e.times, X.mean(0), label=lbl)
plt.vlines(0, linestyle='dashed', *plt.ylim())
def do_pca_for_subject(trial_epochs_csd, response_epochs_csd, s, info):
from matplotlib.gridspec import GridSpec
print('# Participant %i' % s)
r_ep = response_epochs_csd['participant == %i' % s]
X = r_ep.copy().crop(-2., 0).get_data()[:, :32]
trialX = trial_epochs_csd['participant == %i' % s]
respX = r_ep.get_data()[:, :32]
fig = plt.figure(figsize=(20, 16))
gs = GridSpec(nrows=4, ncols=12, figure=fig)
fig.suptitle('Participant %i' % s)
## Do PCA
covariance_csd = np.array(
[np.cov(X[i] - X[i].mean()) for i in range(X.shape[0])])
cov = covariance_csd.mean(0)
# eig_vals, eig_vecs = np.linalg.eig(cov)
eig_vals, eig_vecs = np.linalg.eigh(cov)
eig_vals = eig_vals[::-1] ## Reverse order
eig_vecs = eig_vecs[:, ::-1]
## Variance explained
ve = eig_vals / eig_vals.sum()
ax1 = fig.add_subplot(gs[0, 0:2])
plt.sca(ax1)
plt.plot(list(range(1, len(ve) + 1)), ve * 100, '-o')
plt.hlines(ve.mean() * 100, linestyle='dashed', *plt.xlim())
plt.ylabel('% variance explained')
plt.xlabel('Component')
plt.xticks(list(range(1, 32, 2)))
plt.xlim(0, 12)
## Correct signs
t0, t1 = response_epochs_csd.time_as_index([-2, 0])
respX_pca = np.stack(
[respX[i].T.dot(eig_vecs) for i in range(respX.shape[0])],
axis=0).swapaxes(2, 1)
X = respX_pca.mean(0)
comp_signs = np.sign(X[:, t1] - X[:, t0])
eig_vecs *= comp_signs
## Plot topography
n = 9
for i in range(n):
ax = fig.add_subplot(gs[0, 2 + i])
topomap(eig_vecs[:, i], info, axes=ax)
plt.title('C%i' % (i + 1))
# Get timecourse
respX_pca = np.stack(
[respX[i].T.dot(eig_vecs)
for i in range(respX.shape[0])], axis=0).swapaxes(2, 1) * 1000000
trialX = trial_epochs_csd.get_data()[:, :32]
trialX_pca = np.stack(
[trialX[i].T.dot(eig_vecs)
for i in range(trialX.shape[0])], axis=0).swapaxes(2, 1) * 1000000
## PCA timecourse 1
ax3 = fig.add_subplot(gs[1:2, :6])
plt.sca(ax3)
for i in range(9):
eegf.plot_mean_sem(trialX_pca[:, i],
trial_epochs_csd.times,
label='C %i' % (i + 1))
plt.vlines(0, linestyle='--', *plt.ylim())
plt.legend(loc='upper left', prop={'size': 10})
plt.xlim(-.5, 2)
plt.xlabel('Time from onset (s)')
## PCA timecourse 2
ax4 = fig.add_subplot(gs[1:2, 6:])
plt.sca(ax4)
for i in range(9):
eegf.plot_mean_sem(respX_pca[:, i],
response_epochs_csd.times,
label='C %i' % (i + 1))
plt.vlines(0, linestyle='--', *plt.ylim())
plt.legend(loc='upper left', prop={'size': 10})
plt.xlim(-2, .5)
plt.xlabel('Time to action (s)')
## Rotate
varimax_vectors = varimax(eig_vals[:n] * eig_vecs[:, :n],
method='varimax').T
## Correct signs
t0, t1 = response_epochs_csd.time_as_index([-2, 0])
respX_vmax = np.stack(
[respX[i].T.dot(varimax_vectors) for i in range(respX.shape[0])],
axis=0).swapaxes(2, 1)
X = respX_vmax.mean(0)
comp_signs = np.sign(X[:, t1] - X[:, t0])
varimax_vectors *= comp_signs
## Topography
for i in range(n):
ax = fig.add_subplot(gs[2, 2 + i])
topomap(varimax_vectors[:, i], info, axes=ax, show=False)
plt.title('VM%i' % (i + 1))
## Varimax timecourses
respX_vmax = np.stack(
[respX[i].T.dot(varimax_vectors) for i in range(respX.shape[0])],
axis=0).swapaxes(2, 1)
trialX_vmax = np.stack(
[trialX[i].T.dot(varimax_vectors) for i in range(trialX.shape[0])],
axis=0).swapaxes(2, 1)
## Varimax timecourse 1
ax3 = fig.add_subplot(gs[3, :6])
plt.sca(ax3)
for i in range(9):
eegf.plot_mean_sem(trialX_vmax[:, i],
trial_epochs_csd.times,
label='C %i' % (i + 1))
plt.vlines(0, linestyle='--', *plt.ylim())
plt.legend(loc='upper left', prop={'size': 10})
plt.xlim(-.5, 2)
plt.xlabel('Time from onset (s)')
## Varimax timecourse 2
ax4 = fig.add_subplot(gs[3, 6:])
plt.sca(ax4)
for i in range(9):
eegf.plot_mean_sem(respX_vmax[:, i],
response_epochs_csd.times,
label='C %i' % (i + 1))
plt.vlines(0, linestyle='--', *plt.ylim())
plt.legend(loc='upper left', prop={'size': 10})
plt.xlim(-2, .5)
plt.xlabel('Time to action (s)')
## Finish
# plt.tight_layout()
return fig