def yaw_variation(is_control=True):
ci = [1,3]
#is_control = False
data, segs = get_pooled(ci)
data, segs = filter_by_treatment(data, segs, is_control)
smeans = []
for sid, d in groupby(data, 'session_id'):
yaw = toyota_yaw_rate_in_degrees(d['c_yaw'])
mean = np.mean(yaw)
std = np.std(yaw)
print sid, mean, std
smeans.append((sid, mean))
#mean = np.mean(smeans)
#std = np.std(smeans)
#print 'mean: ' + str(mean) + ' std: ' + str(std)
return smeans
def get_pooled_data(cis, is_control):
arr_desc = [('sid', int), ('bend', int), ('yaw', float), ('speed', float)]
table = np.array([], dtype=arr_desc)
#cis = [1,2,3]
for ci in cis:
data = get_angled_range_data2(CORNERING[ci][0], CORNERING[ci][1])
segs = get_pursuit_fits(CORNERING[ci][0], CORNERING[ci][1])
data, segs = filter_by_treatment(data, segs, is_control)
data.sort(order=['session_id', 'ts'])
grp, segs = zip(*segs)
groups = np.rec.fromrecords(grp, names='session_id,lap')
for (sid, lap), d in groupby_multiple(data, ('session_id', 'lap')):
#if (np.sum(d['g_direction_q'] < 0.2) > 0.25*(len(d))): continue
seg = [segs[k] for k in np.flatnonzero((groups.session_id == sid) &
(groups.lap == lap))]
if (len(seg) == 0): continue
seg = np.hstack(seg).view(np.recarray)
seg = seg[seg.n > 12]
lap_dpoints = []
for s in seg:
dt = s.t1 - s.t0
if dt == 0: continue
gspeed = -((s.d1[0]-s.d0[0]) / dt)
slice = d[(d['ts'] >= s.t0) &
(d['ts'] <= s.t1)]
ym = np.mean(toyota_yaw_rate_in_degrees(slice['c_yaw']))
lap_dpoints.append((ym, gspeed))
lap_dpoints = np.array(lap_dpoints)
if (len(lap_dpoints > 0)):
yaw = np.mean(lap_dpoints[:,0])
speed = np.median(lap_dpoints[:,1])
arr = np.array([(sid,ci,yaw,speed)], dtype=arr_desc)
table = np.append(table, arr)
return table
def pursuit_vs_yaw(range):
groups, data, grp, segs = get_pursuits_and_data_by_range(range)
pdf_out = PdfPages('/tmp/pursuit_vs_yaw.pdf')
bin_size = 0.5
means = []
slopes = []
noises = []
for sid, d in groupby(data, 'session_id'):
seg = [segs[i] for i in np.flatnonzero(groups.session_id == sid)]
seg = np.hstack(seg).view(np.recarray)
#seg = seg[(seg.t1 - seg.t0) > 0.2] # take out saccades
seg = seg[seg.n > 12]
# filter outliers from gaze landings
values = np.vstack((seg.d0[:,0], seg.d0[:,1]))
fkde = gaussian_kde(values)
x, y = np.mgrid[np.min(values[0]):np.max(values[0]):bin_size,
np.min(values[1]):np.max(values[1]):bin_size]
grid = np.vstack((x.ravel(), y.ravel()))
z = fkde(grid)
reshaped = z.reshape(x.shape)
medianhdr = histogram_hdr(z, bin_size**2, 0.8)
seg = [s for s in seg if fkde((s.d0[0], s.d0[1])) > medianhdr]
seg = np.hstack(seg).view(np.recarray)
# mean yaw within each pursuit
withinPursuitYaw = []
for s in seg:
slice = d[(d['ts'] >= s['t0']) &
(d['ts'] <= s['t1'])]
ym = np.mean(toyota_yaw_rate_in_degrees(slice['c_yaw']))
withinPursuitYaw.append(ym)
withinPursuitYaw = np.array(withinPursuitYaw)
dt = seg.t1 - seg.t0
valid = dt > 0
dt = dt[valid]
gspeeds = -((seg.d1[:,0] - seg.d0[:,0]) / dt)
noise = withinPursuitYaw / 2 - gspeeds
plt.figure()
# raw scatter
fit = np.polyfit(withinPursuitYaw, gspeeds, 1)
rng = np.linspace(10, 20, len(withinPursuitYaw))
rline = np.polyval(fit, rng)
model = np.poly1d([0.5, 0])
plt.subplot(221)
plt.plot(withinPursuitYaw, gspeeds, '.k', alpha=0.2)
plt.plot(rng, rline, '-b')
plt.plot([10,20], [np.polyval(model, 10), np.polyval(model, 20)], '-g')
plt.ylim(-50,50)
plt.xlim(5,25)
plt.xlabel('mean yaw deg/s')
plt.ylabel('gaze speed deg/s')
plt.annotate('spearman r: %f' % scipy.stats.spearmanr(withinPursuitYaw, gspeeds)[0], (10,-10))
plt.annotate('slope: %f' % fit[0], (10,-20))
plt.title(sid)
# noise kde + pdf
plt.subplot(222)
kde = gaussian_kde(noise)
rng = np.linspace(np.min(noise), np.max(noise), len(noise))
plt.plot(rng, kde(rng), '-b', label='kde')
plt.plot(rng, scipy.stats.norm.pdf(rng, loc=np.mean(noise), scale=np.std(noise)), '-r', label='pdf')
plt.hist(noise, bins=30, normed=True, color='green')
plt.xlabel('mean yaw / 2 - gaze speed')
plt.legend(loc='upper right')
# pursuit landings scatter & density map
plt.subplot(223)
plt.contour(x,y,reshaped, [medianhdr])
plt.plot(seg.d0[:,0], seg.d0[:,1], ',k')
plt.xlim(-60,60)
plt.ylim(-20,20)
pdf_out.savefig()
plt.close()
print '%i \nspearman r: %f' % (sid, scipy.stats.spearmanr(withinPursuitYaw, gspeeds)[0])
# how many fixations are from right to left (74-90 degrees)
#validspeed = gspeeds < 0
#print '%f of fixations from right to left' % (1.0 * len(gspeeds[validspeed]) / len(gspeeds))
print 'mean of yaws / gaze speeds %f' % (np.mean(gspeeds / withinPursuitYaw))#, scipy.stats.mode(np.around(withinPursuitYaw / gspeeds, 1))[0])
print 'mean of mean yaw / 2 - gaze speed: %f' % (np.mean(withinPursuitYaw / 2 - gspeeds))
print 'noise shapiro-wilk w: %f p: %f' % scipy.stats.shapiro(noise)
#print 'median hdr %f' % medianhdr
#print 'mean of yaws: %f' % np.mean(withinPursuitYaw)
#print 'mean of gspeeds: %f' % np.mean(gspeeds)
means.append((np.mean(withinPursuitYaw), np.mean(gspeeds)))
slopes.append(fit)
noises.append(noise)
pdf_out.close()
return (means, slopes, noises)
def pursuit_yaw_means():
pdf_out = PdfPages('/tmp/pursuit_vs_yaw_pooled.pdf')
cis = [1,2,3]
all_data = []
all_segs = []
for ci in cis:
data = get_angled_range_data2(CORNERING[ci][0], CORNERING[ci][1])
segs = get_pursuit_fits(CORNERING[ci][0], CORNERING[ci][1])
all_data.append(data)
all_segs.append(segs)
data = np.hstack(all_data)
segs = np.vstack(all_segs)
data.sort(order=['session_id', 'ts'])
grp, segs = zip(*segs)
groups = np.rec.fromrecords(grp, names='session_id,lap')
all_dpoints = []
xrange = [12,18]
for sid, i in groupby_i(data, 'session_id'):
d = data[i]
print sid
seg = [segs[k] for k in np.flatnonzero(groups.session_id == sid)]
seg = np.hstack(seg).view(np.recarray)
seg = seg[seg.n > 12]
dpoints = []
for s in seg:
dt = s.t1 - s.t0
if dt == 0: continue
gspeed = -((s.d1[0]-s.d0[0]) / dt)
slice = d[(d['ts'] >= s.t0) &
(d['ts'] <= s.t1)]
ym = np.mean(toyota_yaw_rate_in_degrees(slice['c_yaw']))
naksu_x = pxx2heading(slice['naksu_x'])
naksu_mean = (naksu_x[1] - naksu_x[0]) / dt
dpoints.append((ym, gspeed, naksu_mean))
all_dpoints.append(dpoints)
dpoints = np.array(dpoints)
print scipy.stats.pearsonr(dpoints[:,0], dpoints[:,1])
plt.figure()
plt.title(sid)
plt.hist(dpoints[:,0], bins=50)
#plt.xlim(xrange)
pdf_out.savefig()
plt.close()
all_dpoints = np.array(all_dpoints)
means = np.array([np.median(x,0) for x in all_dpoints])
ymeans = means[:,0]
gmeans = means[:,1]
nmeans = means[:,2]
#from utils import fitLine
#fitLine(ymeans,gmeans)
fit = np.polyfit(ymeans, gmeans, 1)
lsr = np.poly1d(fit)
model = np.poly1d([0.5, 0])
plt.figure()
plt.plot(ymeans,gmeans,'.k')
plt.plot(xrange, lsr(xrange), '-b')
plt.plot(xrange, model(xrange), '-g')
plt.xlabel('mean yaw rate (deg/s)')
plt.ylabel('mean gaze speed (deg/s)')
pdf_out.savefig()
plt.close()
nfit = np.polyfit(ymeans, nmeans, 1)
nlsr = np.poly1d(nfit)
plt.figure()
plt.plot(ymeans,nmeans,'.k')
plt.plot(xrange, nlsr(xrange), '-b')
plt.xlabel('mean yaw rate (deg/s)')
plt.ylabel('mean change in horizontal TP coordinate during pursuit (deg/s)')
pdf_out.savefig()
plt.close()
pdf_out.close()
print '\nspearman r: %f, p %f' % scipy.stats.spearmanr(ymeans, gmeans)
print 'pearson r: %f, p %f' % scipy.stats.pearsonr(ymeans,gmeans)
print 'least-squares fit: %f %f' % (fit[0], fit[1])
print 'yaw/TP location least-squares fit: %f %f' % (nfit[0], nfit[1])
