def annotated_plot(dm):
"""
desc:
Plots mean pupil size separately for dark and bright trials, and
annotates the plot with significance markers.
arguments:
dm:
type: DataMatrix
"""
plot.new(size=(8,6))
lm = trace_lmer_simple(dm)
plt.axvline(RT, color='black', linestyle=':')
x = np.arange(3000)
for color, a in [
('black', 1.96), # p < .05
('red', 2.57), # p < .01
('orange', 2.81), # p < .005
('yellow', 3.29) # p < .001
]:
threshold = series.threshold(lm.t,
lambda t: abs(t) > a, min_length=1)
print('Alpha %.5f (%.2f)' % (a, series.reduce_(threshold)[1]))
plot.threshold(threshold[1], color=color, linewidth=1)
dm_dark = dm.type == 'ctrl'
plot.trace(dm_dark.pupil, color=grey[3],
label='Neutral (N=%d)' % len(dm_dark))
pupil.brightness_plot(dm, subplot=True)
plt.xticks(range(0, 3001, 500), np.linspace(0,3,7))
plt.xlabel('Time since word onset (s)')
plt.ylabel('Pupil size (normalized to word onset)')
plot.save('annotated_plot')
def preprocess(dm):
"""
desc:
Performs pupil preprocessing, and removes trials where pupil size was
unrealistic.
"""
dm.pupil = series.blinkreconstruct(dm.ptrace_target)
dm.pupil.depth = 3000
dm.pupil = series.smooth(dm.pupil, winlen=51)
dm.pupil = series.baseline(dm.pupil, dm.pupil, 0, 1)
# Remove all trials where pupil size has unrealistic values
dm.pupilmax = FloatColumn
dm.pupilmin = FloatColumn
for row in dm:
row.pupilmin = min(row.pupil)
row.pupilmax = max(row.pupil)
plot.new()
# plot.histogram(dm.pupilmin, bins=100, range=(0, 5), color=red[1], alpha=.5)
# plot.histogram(dm.pupilmax, bins=100, range=(0, 5), color=green[1], alpha=.5)
plt.hist(dm.pupilmin, bins=100, range=(0, 5), color=red[1], alpha=0.5)
plt.hist(dm.pupilmax, bins=100, range=(0, 5), color=green[1], alpha=0.5)
l0 = len(dm)
dm = (dm.pupilmin > PUPILRANGE[0]) & (dm.pupilmax < PUPILRANGE[1])
l1 = len(dm)
s = "%d of %d unrealistic values" % (l0 - l1, l0)
plt.title(s)
plot.save("pupil-peaks")
print(s)
return dm
def filteredplot(dm):
dm.bl = baseline.baseline(dm)
dm.y2 = baseline.correct1(dm)
dm.y4 = baseline.correct3(dm)
dm = dm.bl >= MIN_BASELINE
plot.new(size=(8, 3))
plt.subplot(121)
plotmean(dm, 'y2')
plt.subplot(122)
plotmean(dm, 'y4')
plot.save('filteredplot')
def plot_pupiltrace(dm, suffix='', interaction=False):
"""
desc:
Create a plot of the pupil trace annotated with significant regions.
arguments:
dm:
type: DataMatrix
keywords:
suffix: A suffix for the plot.
interaction: Indicates whether the log_velocity by side interaction
should be included.
"""
plot.new(size=(12, 6))
# Plot the three velocities as separate lines
for color, velocity in ( (orange[1], 30), (blue[1], 3), (green[1], .3) ):
dm_ = dm.velocity == velocity
plot.trace(dm_.pupil, label='%s cm/s (N=%d)' % (velocity, len(dm_)),
color=color)
# Run statistical model, and annotate the figure with three alpha levels:
# .05, .01, and .005
if interaction:
model = 'pupil ~ log_velocity*side + (1+log_velocity*side|subject_nr)'
else:
model = 'pupil ~ log_velocity + (1+log_velocity|subject_nr)'
lm = lme4.lmer_series(dm, formula=model, winlen=3,
cacheid='lmer%s' % suffix)
for y, color1, color2, alpha in [
(.94, grey[2], blue[0], .05),
(.942, grey[3], blue[1], .01),
(.944, grey[4], blue[2], .005),
(.946, grey[5], blue[2], .001)
]:
a = series.threshold(lm.p, lambda p: p > 0 and p < alpha,
min_length=1)
plot.threshold(a[1], y=y, color=color1, linewidth=4)
if interaction:
plot.threshold(a[3], y=y+.01, color=color2, linewidth=4)
# Mark the baseline period
plt.axvspan(50, 90, color='black', alpha=.2)
# Mark the cue onset
plt.axvline(90, color='black')
# Adjust the x axis such that 0 is the cue onset, and the units are time in
# seconds (assuming 33 ms per sample)
plt.xticks(np.arange(40, 540, 50), np.arange(-50, 450, 50)*.033)
plt.xlim(0, 540)
plt.xlabel('Time since auditory cue (s)')
plt.ylabel('Pupil size (relative to baseline)')
plt.legend(frameon=False)
plot.save('pupiltrace'+suffix)
def mainplots(dm):
plot.new()
dm.bl = baseline.baseline(dm)
dm.y2 = baseline.correct1(dm)
dm.y4 = baseline.correct3(dm)
plotall(dm, 'y', 0, title='No baseline correction')
plotall(dm, 'y2', 1, title='Divide by baseline')
plotall(dm, 'y4', 2, title='Subtrace baseline')
if '--sim' in sys.argv:
plot.save('main-sim')
elif '--real' in sys.argv:
plot.save('main-real')
def model_comparison(dm):
"""TODO"""
plot.new(size=(8,6))
colors = brightcolors[:]
for iv in ['rating_brightness', 'rating_valence', 'rating_intensity']:
lm = trace_lmer_simple(dm, iv)
plt.plot(np.abs(lm.t[1]), color=colors.pop(), label=iv)
plt.legend()
plt.xticks(range(0, 3001, 500), np.linspace(0,3,7))
plt.xlabel('Time from word onset (s)')
plt.ylabel('|t|')
plot.save('model-comparison')
def word_summary(dm):
"""
desc:
Plots the mean pupil size for dark and bright words as a bar plot. The
time window is indicated by the PEAKWIN constant. This data is also
written to a .csv file.
arguments:
dm:
type: DataMatrix
"""
dm = (dm.type == "light") | (dm.type == "dark")
x = np.arange(dm.pupil.depth)
sm = DataMatrix(length=len(dm.word.unique))
sm.word = 0
sm.type = 0
sm.pupil_win = FloatColumn
sm.pupil_win_se = FloatColumn
sm.pupil_full = FloatColumn
sm.pupil_full_se = FloatColumn
for i, w in enumerate(dm.word.unique):
_dm = dm.word == w
sm.word[i] = w
sm.type[i] = (dm.word == w).type[0]
sm.pupil_win[i], sm.pupil_win_se[i] = size_se(_dm, PEAKWIN[0], PEAKWIN[1])
sm.pupil_full[i], sm.pupil_full_se[i] = size_se(_dm)
sm = operations.sort(sm, sm.pupil_win)
io.writetxt(sm, "%s/word_summary.csv" % OUTPUT_FOLDER)
plot.new(size=(4, 3))
dx = 0
for color, type_ in ((orange[1], "light"), (blue[1], "dark")):
sm_ = sm.type == type_
x = np.arange(len(sm_))
plt.plot(sm_.pupil_win, "o-", color=color)
if type_ == "dark":
yerr = (np.zeros(len(sm_)), sm_.pupil_win_se)
else:
yerr = (sm_.pupil_win_se, np.zeros(len(sm_)))
plt.errorbar(x, sm_.pupil_win, yerr=yerr, linestyle="", color=color, capsize=0)
plt.xlim(-1, 33)
plt.ylabel("Pupil size (normalized)")
plt.xlabel("Word")
plt.xticks([])
plot.save("word_summary")
def histogram(dm):
dm.bl = baseline.baseline(dm)
plot.new(size=(8,4))
y, x = np.histogram(dm.bl, bins=50, range=(0,4000))
x =.5*x[1:]+.5*x[:-1]
plt.subplot(121)
plt.xlabel('Baseline pupil size')
plt.ylabel('N')
plt.fill_between(x, y)
plt.xlim(0, 3000)
plt.axvline(MIN_BASELINE, color='black', linestyle=':')
plt.subplot(122)
plt.xlabel('Baseline pupil size')
plt.ylabel('log10(N)')
plt.fill_between(x, np.log10(y))
plt.xlim(0, 3000)
plt.axvline(MIN_BASELINE, color='black', linestyle=':')
plot.save('baseline-hist')
def subject_diff_traces(dm):
"""
desc:
Plots the difference in pupil size for dark and bright trials over time,
separately for each participant.
arguments:
dm:
type: DataMatrix
"""
plot.new()
x = np.arange(dm.pupil.depth)
for i, s in enumerate(dm.subject_nr.unique):
print(s)
_dm = dm.subject_nr == s
plt.subplot(6, 5, i + 1)
plt.title("Subject %d" % i)
brightness_plot(_dm, subplot=True)
plot.save("subject_diff_trace")
def brightness_plot(dm, subplot=False):
"""
desc:
Plots mean pupil size separately for dark and bright trials over time.
arguments:
dm:
type: DataMatrix
keywords:
subplot: Indicates whether a new plot should be created, or not.
"""
if not subplot:
plot.new()
dm_bright = dm.type == "light"
dm_dark = dm.type == "dark"
plot.trace(dm_bright.pupil, color=orange[1], label="Bright (N=%d)" % len(dm_bright))
plot.trace(dm_dark.pupil, color=blue[1], label="Dark (N=%d)" % len(dm_dark))
plt.legend(frameon=False)
if not subplot:
plot.save("brightness_plot")
def brightness_intensity(dm):
"""TODO"""
lm = lme4.lmer_series((dm.type == 'light') | (dm.type == 'dark'),
'pupil ~ type + rating_intensity + (1+type|subject_nr)',
winlen=WINLEN, cacheid='lmer_series_type_intensity')
threshold = series.threshold(lm.t, lambda t: abs(t) > 1.96, min_length=200)
print('threshold', threshold > 0)
plot.new(size=(8,6))
plt.plot(threshold)
plt.plot(lm.t[1], label='Type')
plt.plot(lm.t[2], label='Intensity')
plot.save('model-type-intensity')
lm = lme4.lmer_series(dm.type != 'animal',
'pupil ~ rating_brightness + rating_intensity + (1+rating_brightness|subject_nr)',
winlen=WINLEN, cacheid='lmer_series_brightness_intensity')
plot.new(size=(8,6))
plt.plot(lm.t[1], label='Brightness')
plt.plot(lm.t[2], label='Intensity')
plot.save('model-brightness-intensity')
def subject_summary(dm):
"""
desc:
Plots the mean difference in pupil size between dark and bright trials
for each participant as a bar plot. The time window is indicated by the
PEAKWIN constant. This data is also written to a .csv file.
arguments:
dm:
type: DataMatrix
"""
x = np.arange(len(dm.subject_nr.unique))
sm = DataMatrix(length=len(dm.subject_nr.unique))
sm.subject_nr = 0
sm.effect_win = FloatColumn
sm.effect_win_se = FloatColumn
sm.effect_full = FloatColumn
sm.effect_full_se = FloatColumn
for i, s in enumerate(dm.subject_nr.unique):
_dm = dm.subject_nr == s
sm.subject_nr[i] = s
sm.effect_win[i], sm.effect_win_se[i] = effect_se(_dm, PEAKWIN[0], PEAKWIN[1])
sm.effect_full[i], sm.effect_full_se[i] = effect_se(_dm)
sm = operations.sort(sm, by=sm.effect_win)
plot.new(size=(4, 3))
plt.axhline(0, color="black")
plt.plot(sm.effect_win, "o-", color=green[-1])
plt.errorbar(x, sm.effect_win, yerr=sm.effect_win_se, linestyle="", color=green[-1], capsize=0)
plt.xlim(-1, 30)
plt.ylabel("Pupil-size difference (normalized)")
plt.xlabel("Participant")
plt.xticks([])
plot.save("subject_summary")
io.writetxt(sm, "%s/subject_summary.csv" % OUTPUT_FOLDER)
def valence_plot(dm, subplot=False):
"""
desc:
Plots mean pupil size separately for positive and negative trials over
time.
arguments:
dm:
type: DataMatrix
keywords:
subplot: Indicates whether a new plot should be created, or not.
"""
if not subplot:
plot.new()
dm_pos = dm.category == "positive"
dm_neg = dm.category == "negative"
plot.trace(dm_pos.pupil, color=green[1], label="Positive (N=%d)" % len(dm_pos))
plot.trace(dm_neg.pupil, color=red[1], label="Negative (N=%d)" % len(dm_neg))
plt.legend(frameon=False)
if not subplot:
plot.save("valence_plot")
rm = DataMatrix(length=101)
rm.word = ''
rm._type = ''
rm.rating_brightness = FloatColumn
rm.rating_valence = FloatColumn
rm.rating_intensity = FloatColumn
for row, word in zip(rm, dm.word.unique):
dm_ = dm.word == word
dme = dm_.rating_type == 'emotion'
dmb = dm_.rating_type == 'brightness'
# Pénombre was accidentally rated twice.
assert(len(dmb)==30 or word == 'pénombre')
assert(len(dme)==30 or word == 'pénombre')
row.word = word
row._type = dme.type[0]
row.rating_brightness = dmb.rating.mean
row.rating_valence = dme.rating.mean
# The intensity is just the deviation from the middle score (2). In the
# initial analysis, the deviation from the mean valence was taken. But
# taking the per-trial deviation from the middle score, and then averaging
# that seems to make more sense.
row.rating_intensity = np.abs(dme.rating-2).mean()
io.writetxt(rm, 'ratings.csv')
# Determine the correlations
print(plot.regress(rm.rating_brightness, rm.rating_valence))
plot.save('regress.brightness.valence')
print(plot.regress(rm.rating_brightness, rm.rating_intensity))
plot.save('regress.brightness.intensity')
def powersummary(dummy):
plot.new()
for i, blinksinbaseline in enumerate((True, False)):
l1 = []
l2 = []
l4 = []
ls1 = []
ls2 = []
ls4 = []
effectsizes = range(0, 501, 50)
for effectsize in effectsizes:
kwargs = {
'effectsize': effectsize,
'blfilter': False,
'blinksinbaseline': blinksinbaseline
}
cid = 'power-%(effectsize)s-%(blfilter)s-%(blinksinbaseline)s' % kwargs
p1, p2, p4, s1, s2, s4 = power(None, cacheid=cid, **kwargs)
l1.append(p1)
l2.append(p2)
l4.append(p4)
ls1.append(s1)
ls2.append(s2)
ls4.append(s4)
plt.subplot(2, 2, i * 2 + 1)
if i == 0:
plt.title('With blinks during baseline')
else:
plt.title('No blinks during baseline')
plt.axhline(.025, color='black', linestyle=':')
plt.xlim(effectsizes[0] - 10, effectsizes[-1] + 10)
plt.ylim(-.05, 1.05)
plt.xlabel('True effect size')
plt.ylabel('Proportion correct detections')
plt.plot(effectsizes,
l1,
'o-',
label='No baseline correction',
color=COLORNOCORRECT)
plt.plot(effectsizes,
l2,
'o-',
label='Divide by baseline',
color=COLORDIVIDE)
plt.plot(effectsizes,
l4,
'o-',
label='Subtract baseline',
color=COLORSUBTRACT)
plt.subplot(2, 2, i * 2 + 2)
plt.title('Spurious')
plt.axhline(.025, color='black', linestyle=':')
plt.xlim(effectsizes[0] - 10, effectsizes[-1] + 10)
plt.ylim(-.05, 1.05)
plt.xlabel('True effect size')
plt.ylabel('Proportion spurious detections')
plt.plot(effectsizes,
ls1,
'o-',
label='No baseline correction',
color=COLORNOCORRECT)
plt.plot(effectsizes,
ls2,
'o-',
label='Divide by baseline',
color=COLORDIVIDE)
plt.plot(effectsizes,
ls4,
'o-',
label='Subtract baseline',
color=COLORSUBTRACT)
if i == 3:
plt.legend(frameon=False, loc='lower right')
plot.save('powersummary')
plot.new(size=(16, 5))
plt.subplot(1, 3, 1)
plt.xlim(-5, 5)
plt.title('a) VC ~ pupil size')
roi_hist(ldm, 't_bold_pupil')
plt.subplot(1, 3, 2)
plt.xlim(-5, 5)
plt.title('b) VC ~ luminance')
roi_hist(ldm, 't_bold_luminance')
plt.subplot(1, 3, 3)
plt.xlim(-10, 10)
plt.title('c) VC ~ visual change')
roi_hist(ldm, 't_bold_change')
plot.save('histograms', show=True)
# </codecell>
# <markdowncell>
"""
## Correlations between pupil size, luminance, and visual change
__Conclusion:__
- Pupil size and luminance are strongly negatively correlated
- Pupil size and visual change are negatively correlated
- Visual change and luminance are positively correlated
"""
# </markdowncell>
