def load_and_plot_scores_regressions(
sub_path="Intercept_surprise_100_Surprisenp1_RepeatAlter_RepeatAlternp1",
filter_name='Stand',
scores_name='scores--remapped_gtmbaselined_clean.npy'):
from jr.plot import pretty_decod
results_path = op.join(config.result_path, 'linear_models', filter_name,
sub_path)
scores_all = []
for subject in config.subjects_list:
subject_path = op.join(results_path, subject)
score = np.load(op.join(subject_path, scores_name))
scores_all.append(score)
scores_all = np.asarray(scores_all)
pretty_decod(np.mean(scores_all, 1),
np.linspace(-0.048, 0.65, scores_all.shape[-1]),
chance=0)
plt.title("Score " + filter_name)
plt.show()
gat.estimators_ = [gat.estimators_[t] for t in toi_]
gat.train_times_['times'] = [gat.train_times_['times'][t] for t in toi_]
gat.train_times_['slices'] = [gat.train_times_['slices'][t] for t in toi_]
# predict all trials, including absent to keep cv scheme
gat.predict(epochs[sel_gat])
# subscore on present only
gat.scores_ = subscore(gat, sel, y[sel])
scores.append(gat.scores_)
# Save classifier results
save([gat, analysis, sel_gat, events],
'decod',
subject=subject,
analysis=analysis['name'] + '_probe_to_target')
times = epochs.times[:-5]
score_diag = np.array([np.diag(np.array(score)[4:, :]) for score in scores])
p_val = stats(score_diag)
pretty_decod(score_diag,
times=times,
sig=p_val < .05,
color=analysis['color'],
fill=True)
# toi_ = np.where((times >= toi[0]) & (times < toi[1]))[0]
toi_ = np.where(p_val < .05)[0]
score = np.mean(score_diag[:, toi_], axis=1)
print(np.mean(score, axis=0),
np.std(score, axis=0) / np.sqrt(len(score_diag)), np.min(p_val[toi_]),
times[toi_])
scores = -np.array(results['bias'][:, ii, jj, ...])
p_val = results['bias_pval'][ii, jj, :, :]
pretty_gat(scores.mean(0), times=times, ax=axes[ii, jj],
colorbar=False, clim=[-.1, .1], sig=p_val < .05)
axes[ii, jj].axvline(.800, color='k')
axes[ii, jj].axhline(.800, color='k')
pretty_axes(axes, ylabelpad=-15, xticks=np.linspace(-.100, 1.400, 13),
xticklabels=['', 0] + [''] * 13 + [1400, ''])
pretty_colorbar(cax=fig.add_axes([.88, .25, .02, .55]), ax=axes[0, 0])
report.add_figs_to_section(fig, 'gat', 'bias')
# Plot bias of TG diagonal (i.e. decoding bias)
fig, ax = plt.subplots(1, figsize=[7., 2.])
scores = np.array([np.diag(s) for s in results['bias'][:, 0, 1, ...]])
p_val = np.diag(results['bias_pval'][0, 1, :, :])
pretty_decod(-scores, ax=ax, times=times, color=cmap(1.), sig=p_val < .05,
fill=True)
ax.axvline(.800, color='k')
ax.set_xlabel('Times', labelpad=-10)
fig.tight_layout()
report.add_figs_to_section(fig, 'diagonal', 'bias')
def quick_stats(x, ax=None):
"""Report significant bias in each toi for unseen and seen"""
pvals = [wilcoxon(ii)[1] for ii in x.T]
sig = [['', '*'][p < .05] for p in pvals]
m = np.nanmean(x, axis=0)
s = np.nanstd(x, axis=0)
print(m, s, pvals)
if ax is not None:
for x_, (y_, sig_) in enumerate(zip(m / 2., sig)):
fill_color=analysis['color'])
for ax in axs:
ax.axvline(.800, color='k')
if analysis['typ'] == 'regress':
ax.set_ylabel('R', labelpad=-15)
elif analysis['typ'] == 'categorize':
ax.set_ylabel('AUC', labelpad=-15)
else:
ax.set_ylabel('rad.', labelpad=-15)
ax.set_yticklabels(['', '', '%.2f' % ax.get_yticks()[2]])
ax.set_xlabel('Times', labelpad=-10)
report.add_figs_to_section(fig_offdiag, 'slices', analysis['name'])
# Decod
ax_diag = fig_alldiag.add_subplot(ax_diag)
pretty_decod(scores_diag, times=times, sig=p_values_diag < alpha,
chance=chance, color=analysis['color'], fill=True, ax=ax_diag)
xlim, ylim = ax_diag.get_xlim(), np.array(ax_diag.get_ylim())
# ylim[1] = np.ceil(ylim[1] * 10) / 10.
sem = scores_diag.std(0) / np.sqrt(len(scores_diag))
ylim = [np.min(scores_diag.mean(0) - sem),
np.max(scores_diag.mean(0) + sem)]
ax_diag.set_ylim(ylim)
ax_diag.axvline(.800, color='k')
ax_diag.set_xticklabels([int(x) if x in np.linspace(0., 1000., 11) else ''
for x in np.round(1e3 * ax_diag.get_xticks())])
if ax_diag != axes_alldiag[-1]:
# bottom suplot
ax_diag.set_xlabel('')
ax_diag.set_xticklabels([])
ax_diag.xaxis.set_visible(False)
elif ax_diag == axes_alldiag[0]:
for roi in rois:
idx = np.where([roi in this_roi for this_roi in labels])[0]
scores.append(np.mean(data[:, idx, :], axis=1))
p_values = stats(np.transpose(scores, [1, 2, 0]) - chance, n_jobs=-1)
fig, axes = plt.subplots(len(rois), 1, sharex=True, sharey=True,
figsize=[10, 40])
cmap = plt.get_cmap('rainbow')
colors = cmap(np.linspace(0., 1., len(rois)))
ylim = [2, -2]
# Plot
details_toi, details_roi = list(), list()
for ii, (score, p_val, ax, color, roi) in enumerate(zip(
scores, p_values.T, axes, colors, rois)):
pretty_decod(score, sig=p_val < .05, times=times-np.min(times),
color=color, ax=ax, fill=True, chance=chance)
xlim = ax.get_xlim()
ylim[0] = min([ylim[0], ax.get_ylim()[0]])
ylim[1] = max([ylim[1], ax.get_ylim()[1]])
ax.set_ylim(ylim)
ax.set_yticks(ylim)
ax.set_yticklabels(['%.3f' % y for y in ylim])
ax.set_xticklabels([int(x) if x in np.linspace(0., 1200., 11) else ''
for x in np.round(1e3 * ax.get_xticks())])
if ax != axes[-1]:
ax.set_xlabel('')
ax.set_xticklabels([])
elif ax == axes[0]:
ax.text(0, ylim[1], 'Target', backgroundcolor='w', ha='center',
va='top')
all_pval_tois[vis, t, :] = stats(score_toi - analysis['chance'])
save([all_score_tois, all_pval_tois, times], 'score', analysis=ana_name)
return [all_score_tois, all_pval_tois, times]
# Main plotting
colors = dict(visibility=plt.get_cmap('bwr')(np.linspace(0, 1, 4.)),
contrast=plt.get_cmap('hot_r')([.5, .75, 1.]))
# Loop across visibility and orientation analyses
for analysis in analyses:
# Plot correlation of decoding score with visibility and contrast
scores, R, times = _analyze_continuous(analysis)
fig, (ax1, ax2) = plt.subplots(2, 1, figsize=[20, 10])
sig = stats(R['visibility']) < .05
pretty_decod(-R['visibility'], times=times, sig=sig, ax=ax1,
color='purple', fill=True)
sig = stats(R['contrast']) < .05
pretty_decod(-R['contrast'], times=times, sig=sig, ax=ax2,
color='orange', fill=True)
report.add_figs_to_section([fig], ['continuous regress'], analysis['name'])
# Plot decoding score for each visibility level
all_scores, score_pvals, times = _subscore_pipeline(analysis)
if 'circAngle' in analysis['name']:
all_scores /= 2. # from circle to half circle
figs, axes = list(), list()
for vis in range(4):
fig, ax = plt.subplots(1, figsize=[14, 11])
scores = all_scores[:, vis, ...]
p_val = score_pvals[vis]
pretty_gat(np.nanmean(scores, axis=0), times=times,
gat, _, sel_gat, _ = load('decod', subject=subject, analysis='probe_phase')
# only keep estimators after probe onset
times = gat.train_times_['times']
toi_ = np.where((times >= (.800 + toi[0])) & (times < (.800 + toi[1])))[0]
gat.estimators_ = [gat.estimators_[t] for t in toi_]
gat.train_times_['times'] = [gat.train_times_['times'][t] for t in toi_]
gat.train_times_['slices'] = [gat.train_times_['slices'][t] for t in toi_]
# predict all trials, including absent to keep cv scheme
gat.predict(epochs[sel_gat])
# subscore on present only
gat.scores_ = subscore(gat, sel, y[sel])
scores.append(gat.scores_)
# Save classifier results
save([gat, analysis, sel_gat, events], 'decod', subject=subject,
analysis=analysis['name'] + '_probe_to_target')
times = epochs.times[:-5]
score_diag = np.array([np.diag(np.array(score)[4:, :]) for score in scores])
p_val = stats(score_diag)
pretty_decod(score_diag, times=times, sig=p_val < .05, color=analysis['color'],
fill=True)
# toi_ = np.where((times >= toi[0]) & (times < toi[1]))[0]
toi_ = np.where(p_val < .05)[0]
score = np.mean(score_diag[:, toi_], axis=1)
print(np.mean(score, axis=0),
np.std(score, axis=0) / np.sqrt(len(score_diag)),
np.min(p_val[toi_]),
times[toi_])
axes[ii, jj].axvline(.800, color='k')
axes[ii, jj].axhline(.800, color='k')
pretty_axes(axes,
ylabelpad=-15,
xticks=np.linspace(-.100, 1.400, 13),
xticklabels=['', 0] + [''] * 13 + [1400, ''])
pretty_colorbar(cax=fig.add_axes([.88, .25, .02, .55]), ax=axes[0, 0])
report.add_figs_to_section(fig, 'gat', 'bias')
# Plot bias of TG diagonal (i.e. decoding bias)
fig, ax = plt.subplots(1, figsize=[7., 2.])
scores = np.array([np.diag(s) for s in results['bias'][:, 0, 1, ...]])
p_val = np.diag(results['bias_pval'][0, 1, :, :])
pretty_decod(-scores,
ax=ax,
times=times,
color=cmap(1.),
sig=p_val < .05,
fill=True)
ax.axvline(.800, color='k')
ax.set_xlabel('Times', labelpad=-10)
fig.tight_layout()
report.add_figs_to_section(fig, 'diagonal', 'bias')
def quick_stats(x, ax=None):
"""Report significant bias in each toi for unseen and seen"""
pvals = [wilcoxon(ii)[1] for ii in x.T]
sig = [['', '*'][p < .05] for p in pvals]
m = np.nanmean(x, axis=0)
s = np.nanstd(x, axis=0)
print(m, s, pvals)
0.176, 0.18 , 0.184, 0.188, 0.192, 0.196, 0.2 , 0.204,
0.208, 0.212, 0.216, 0.22 , 0.224, 0.228, 0.232, 0.236,
0.24 , 0.244, 0.248, 0.252, 0.256, 0.26 , 0.264, 0.268,
0.272, 0.276, 0.28 , 0.284, 0.288, 0.292, 0.296, 0.3 ,
0.304, 0.308, 0.312, 0.316, 0.32 , 0.324, 0.328, 0.332,
0.336, 0.34 , 0.344, 0.348, 0.352, 0.356, 0.36 , 0.364,
0.368, 0.372, 0.376, 0.38 , 0.384, 0.388, 0.392, 0.396,
0.4 , 0.404, 0.408, 0.412, 0.416, 0.42 , 0.424, 0.428,
0.432, 0.436, 0.44 , 0.444, 0.448, 0.452, 0.456, 0.46 ,
0.464, 0.468, 0.472, 0.476, 0.48 , 0.484, 0.488, 0.492,
0.496, 0.5 , 0.504, 0.508, 0.512, 0.516, 0.52 , 0.524,
0.528, 0.532, 0.536, 0.54 , 0.544, 0.548, 0.552, 0.556,
0.56 , 0.564, 0.568, 0.572, 0.576, 0.58 , 0.584, 0.588,
0.592, 0.596, 0.6 ]
pretty_decod(scores_regression_surprise,times)
plt.plot(times,scores_regression_surprise.T)
plt.title('r2 score CV 4 surprise')
plt.xlabel('time (s)')
plt.xticks([0,0.25,0.5])
plt.axvline([0])
plt.axvline([0.25])
plt.axvline([0.5])
plt.show()
betas_dict = {'intercept':[],'RepeatAlter':[],'RepeatAlternp1':[],'surpriseN':[],'surpriseNp1':[]}
mean_evoked = {'intercept':[],'RepeatAlter':[],'RepeatAlternp1':[],'surpriseN':[],'surpriseNp1':[]}
for key in betas_dict:
for subj in config.subjects_list:
base_path = config.result_path + '/linear_models/reg_repeataltern_surpriseOmegainfinity/' + subj +'/_cvbeta_'+key+'-ave.fif'
evo = mne.read_evokeds(base_path)
if analysis['typ'] == 'regress':
ax.set_ylabel('R', labelpad=-15)
elif analysis['typ'] == 'categorize':
ax.set_ylabel('AUC', labelpad=-15)
else:
ax.set_ylabel('rad.', labelpad=-15)
ax.set_yticklabels(['', '', '%.2f' % ax.get_yticks()[2]])
ax.set_xlabel('Times', labelpad=-10)
report.add_figs_to_section(fig_offdiag, 'slices', analysis['name'])
# Decod
ax_diag = fig_alldiag.add_subplot(ax_diag)
pretty_decod(scores_diag,
times=times,
sig=p_values_diag < alpha,
chance=chance,
color=analysis['color'],
fill=True,
ax=ax_diag)
xlim, ylim = ax_diag.get_xlim(), np.array(ax_diag.get_ylim())
# ylim[1] = np.ceil(ylim[1] * 10) / 10.
sem = scores_diag.std(0) / np.sqrt(len(scores_diag))
ylim = [
np.min(scores_diag.mean(0) - sem),
np.max(scores_diag.mean(0) + sem)
]
ax_diag.set_ylim(ylim)
ax_diag.axvline(.800, color='k')
ax_diag.set_xticklabels([
int(x) if x in np.linspace(0., 1000., 11) else ''
for x in np.round(1e3 * ax_diag.get_xticks())
print('There is something wrong with the filename')
inds = np.where(filter_times)
scores_to_avg_filter = scores_to_avg[:, inds[0], inds[0]]
signif = np.zeros((scores_to_avg.shape[1], 1))
p_vals = stats.stats_cluster_based_permutation_test(scores_to_avg_filter -
chance[k])
sig = (p_vals < 0.05)
for l, ll in enumerate(inds[0]):
signif[ll] = sig[l]
pretty_decod(np.mean(scores_to_avg, axis=0),
times=time_list,
chance=chance[k],
sfreq=100,
sig=signif,
fill=True)
fig = plt.gcf()
fig.savefig(fig_path + 'diagonal_' + names[k])
plt.close('all')
# ============= fancy plot ============
# COMP
# hemifield, location-within-hemifield, retinotopic, decade symbolic, unit symbolic, quantity
conditions = [
['/hemifield/scores_*_standard_-400_1000_comp.pkl'],
[
'/location/scores_*_standard_-400_1000_comp_left_hemifield.pkl',
return [all_score_tois, all_pval_tois, times]
# Main plotting
colors = dict(visibility=plt.get_cmap('bwr')(np.linspace(0, 1, 4.)),
contrast=plt.get_cmap('hot_r')([.5, .75, 1.]))
# Loop across visibility and orientation analyses
for analysis in analyses:
# Plot correlation of decoding score with visibility and contrast
scores, R, times = _analyze_continuous(analysis)
fig, (ax1, ax2) = plt.subplots(2, 1, figsize=[20, 10])
sig = stats(R['visibility']) < .05
pretty_decod(-R['visibility'],
times=times,
sig=sig,
ax=ax1,
color='purple',
fill=True)
sig = stats(R['contrast']) < .05
pretty_decod(-R['contrast'],
times=times,
sig=sig,
ax=ax2,
color='orange',
fill=True)
report.add_figs_to_section([fig], ['continuous regress'], analysis['name'])
# Plot decoding score for each visibility level
all_scores, score_pvals, times = _subscore_pipeline(analysis)
if 'circAngle' in analysis['name']:
all_scores /= 2. # from circle to half circle
# get some time informations (TODO stats)
fname = op.join(data_path, 's%i_%s_fit.pkl' % (subject, contrast))
with open(fname, 'rb') as f:
gat = pickle.load(f)
times = gat.train_times_['times']
# plot scores
fig_diag, axes_diag = plt.subplots(len(contrasts), 1, figsize=[2, 5])
cmap = plt.get_cmap('rainbow')
colors = cmap(np.linspace(0., 1., len(contrasts)))
for contrast, color, ax in zip(contrasts, colors, axes_diag):
scores = all_scores[contrast]
# plot diagonal
scores_diag = np.array([np.diag(x) for x in scores])
pretty_decod(scores_diag, times=times, color=color, ax=ax, chance=.5)
ax.set_title(contrast)
if ax != axes_diag[-1]:
ax.set_xticklabels([])
ax.set_xlabel('')
# plot gat
fig, ax = plt.subplots(1, figsize=[5, 5])
pretty_gat(np.mean(scores, axis=0), times=times, chance=.5)
report.add_figs_to_section(fig, 'GAT %s' % contrast, contrast)
# plot slices
tois = np.arange(0., .600, .100)
fig, axes = plt.subplots(len(tois), 1, figsize=[5, 5])
pretty_slices(scores, tois=tois, times=times, chance=.5, axes=axes)
report.add_figs_to_section(fig, 'Slices %s' % contrast, contrast)
fig_diag.tight_layout()
report.add_figs_to_section(fig_diag, 'Diagonal', 'All')