def plot_size_tuning(arr, colors=None):
usize = np.array((0, 5, 8, 13, 22, 36))
ucontrast = np.array((0, 6, 12, 25, 50, 100))
arr_sz = gen_size_tuning(arr)
arr_sz = arr_sz / arr_sz.max(1).max(1)[:, np.newaxis, np.newaxis]
lb, ub = ut.bootstrap(arr_sz, np.mean, pct=(2.5, 97.5))
to_plot = arr_sz.mean(0)
ut.plot_bootstrapped_errorbars_hillel(usize,
arr_sz[:, :, 1::].transpose(
(0, 2, 1)),
colors=colors)
plt.ylim(to_plot.min() - 0.1, to_plot.max() + 0.1)
# plt.title('%d%% contrast' % ucontrast[ic])
plt.xlabel('size ($^o$)')
# plt.subplot(1,2,1)
plt.tight_layout()
def plot_size_tuning_by_contrast(arr):
usize = np.array((0, 5, 8, 13, 22, 36))
ucontrast = np.array((0, 6, 12, 25, 50, 100))
arr_sz = gen_size_tuning(arr)
arr_sz = arr_sz / arr_sz.max(1).max(1)[:, np.newaxis, np.newaxis]
lb, ub = ut.bootstrap(arr_sz, np.mean, pct=(2.5, 97.5))
to_plot = arr_sz.mean(0)
for ic in range(1, 6):
plt.subplot(1, 5, ic)
ut.plot_bootstrapped_errorbars_hillel(usize,
arr_sz[:, :, ic].transpose(
(0, 2, 1)),
colors=['k', 'r'],
markersize=5)
# plt.scatter((0,),to_plot[:,0,0].mean(0))
# plt.scatter((0,),to_plot[:,0,1].mean(0))
plt.ylim(0.5 * to_plot.min(), 1.2 * to_plot.max())
plt.title('%d%% contrast' % ucontrast[ic])
plt.xlabel('size ($^o$)')
plt.subplot(1, 5, 1)
plt.ylabel('event rate / max event rate')
plt.tight_layout()
def plot_interp_size_tuning(ams=None,
data=None,
theta=None,
these_contrasts=[1, 3, 5],
ninterp=101,
usize=np.array((5, 8, 13, 22, 36, 60)),
ucontrast=np.array((0, 6, 12, 25, 50, 100)) / 100,
colors=None,
error_type='bs',
deriv=False,
deriv_axis=1,
sub=False,
two_n=False):
usize_interp = np.linspace(0, usize[-1], ninterp)
this_ncontrast = len(these_contrasts)
this_usize = usize_interp
this_ucontrast = ucontrast
if ams is None:
#assert(not data is None and not theta is None)
assert (not data is None)
ams = ayaz_model(data,
usize=usize,
ucontrast=ucontrast,
theta=theta,
sub=sub,
two_n=two_n)
this_ucontrast = np.array([ams.ucontrast[i] for i in these_contrasts])
this_theta = ams.theta
fn = ams.fn
usize0 = np.concatenate(((0, ), usize))
this_data = sim_utils.gen_size_tuning(ams.data)
if deriv:
if deriv_axis == 1:
this_data = sca.compute_slope_avg(usize0,
this_data,
axis=deriv_axis)
elif deriv_axis == 2:
this_data = sca.compute_slope_avg(ucontrast,
this_data,
axis=deriv_axis)
this_modeled = np.zeros((ninterp, this_ncontrast))
for icontrast in range(this_ncontrast):
if deriv_axis == 1:
srf = lambda d: ams.fn(np.array(
(this_ucontrast[icontrast], )), np.array((d, )))[0, 0]
sslope = np.array([grad(srf)(ss) for ss in this_usize])
this_modeled[:, icontrast] = sslope
elif deriv_axis == 2:
for isize, ss in enumerate(this_usize):
crf = lambda c: ams.fn(np.array((c, )), np.array(
(ss, )))[0, 0]
sslope = grad(crf)(this_ucontrast[icontrast])
this_modeled[isize, icontrast] = sslope
else:
this_modeled = ams.fn(this_ucontrast, this_usize)
if True:
if error_type == 'bs':
ut.plot_bootstrapped_errorbars_hillel(
usize0,
this_data[:, :, these_contrasts].transpose((0, 2, 1)),
linewidth=0,
colors=colors)
elif error_type == 'pct':
ut.plot_pct_errorbars_hillel(usize0,
this_data[:, :,
these_contrasts].transpose(
(0, 2, 1)),
linewidth=0,
colors=colors,
pct=(16, 84))
for icontrast in range(this_ncontrast):
plt.plot(usize_interp,
this_modeled[:, icontrast],
c=colors[icontrast])
#plt.xticks(cinds,(100*ucontrast).astype('int'))
ut.erase_top_right()
plt.xlabel('size ($^o$)')
plt.ylabel('event rate/mean')
if not deriv:
plt.gca().set_ylim(bottom=0)
plt.tight_layout()
def plot_interp_contrast_tuning(ams=None,
data=None,
theta=None,
these_sizes=[0, 2, 4],
ninterp=101,
usize=np.array((5, 8, 13, 22, 36, 60)),
ucontrast=np.array(
(0, 6, 12, 25, 50, 100)) / 100,
colors=None,
deriv=False,
deriv_axis=2):
ucontrast_interp = np.linspace(0, 1, ninterp)
this_nsize = len(these_sizes)
this_ucontrast = ucontrast_interp
if ams is None:
#assert(not data is None and not theta is None)
assert (not data is None)
ams = ayaz_model(data, usize=usize, ucontrast=ucontrast, theta=theta)
this_usize = np.array([ams.usize[i] for i in these_sizes])
this_theta = ams.theta
fn = ams.fn
this_data = ams.data
ind0 = 2
cinds = np.concatenate(
((5 + np.log2(ucontrast_interp[ind0]), ), np.arange(1, 6)))
if deriv:
if deriv_axis == 1:
this_data = sca.compute_slope_avg(usize,
this_data,
axis=deriv_axis)
elif deriv_axis == 2:
this_data = sca.compute_slope_avg(ucontrast,
this_data,
axis=deriv_axis)
this_modeled = np.zeros((this_nsize, ninterp))
for isize in range(this_nsize):
if deriv_axis == 2:
crf = lambda c: ams.fn(np.array(
(c, )), np.array((this_usize[isize], )))[0, 0]
cslope = np.array([grad(crf)(cc) for cc in this_ucontrast])
this_modeled[isize] = cslope
#print(cslope)
elif deriv_axis == 1:
for icontrast, cc in enumerate(this_ucontrast):
srf = lambda d: ams.fn(np.array((cc, )), np.array(
(d, )))[0, 0]
sslope = grad(srf)(this_usize[isize])
this_modeled[isize, icontrast] = sslope
else:
#this_modeled = nub_utils.ayaz_like_theta(this_ucontrast,this_usize,this_theta,fn=fn)
this_modeled = ams.fn(this_ucontrast, this_usize)
ut.plot_bootstrapped_errorbars_hillel(cinds,
this_data[:,
these_sizes, :].transpose(
(0, 1, 2)),
linewidth=0,
colors=colors)
for isize in range(this_nsize):
plt.plot(5 + np.log2(ucontrast_interp[ind0:]),
this_modeled[isize, ind0:],
c=colors[isize])
plt.xticks(cinds, (100 * ucontrast).astype('int'))
ut.erase_top_right()
plt.xlabel('contrast (%)')
plt.ylabel('event rate/mean')
if not deriv:
plt.gca().set_ylim(bottom=0)
plt.tight_layout()