def plot_fit_combined(
data: Union[sim2d.Data2DRT, dict] = None,
pModel_cond_rt_chFlat=None, model=None,
pModel_dimRel_condDense_chFlat=None,
# --- in place of data:
pAll_cond_rt_chFlat=None,
evAll_cond_dim=None,
pTrain_cond_rt_chFlat=None,
evTrain_cond_dim=None,
pTest_cond_rt_chFlat=None,
evTest_cond_dim=None,
dt=None,
# --- optional
ev_dimRel_condDense_fr_dim_meanvar=None,
dt_model=None,
to_plot_internals=True,
to_plot_params=True,
to_plot_choice=True,
# to_group_irr=False,
group_dcond_irr=None,
to_combine_ch_irr_cond=True,
kw_plot_pred=(),
kw_plot_pred_ch=(),
kw_plot_data=(),
axs=None,
):
"""
:param data:
:param pModel_cond_rt_chFlat:
:param model:
:param pModel_dimRel_condDense_chFlat:
:param ev_dimRel_condDense_fr_dim_meanvar:
:param to_plot_internals:
:param to_plot_params:
:param to_group_irr:
:param to_combine_ch_irr_cond:
:param kw_plot_pred:
:param kw_plot_data:
:param axs:
:return:
"""
if data is None:
if pTrain_cond_rt_chFlat is None:
pTrain_cond_rt_chFlat = pAll_cond_rt_chFlat
if evTrain_cond_dim is None:
evTrain_cond_dim = evAll_cond_dim
if pTest_cond_rt_chFlat is None:
pTest_cond_rt_chFlat = pAll_cond_rt_chFlat
if evTest_cond_dim is None:
evTest_cond_dim = evAll_cond_dim
else:
_, pAll_cond_rt_chFlat, _, _, evAll_cond_dim = \
data.get_data_by_cond('all')
_, pTrain_cond_rt_chFlat, _, _, evTrain_cond_dim = data.get_data_by_cond(
'train_valid', mode_train='easiest')
_, pTest_cond_rt_chFlat, _, _, evTest_cond_dim = data.get_data_by_cond(
'test', mode_train='easiest')
dt = data.dt
hs = {}
if model is None:
assert not to_plot_internals
assert not to_plot_params
if dt_model is None:
if model is None:
dt_model = dt
else:
dt_model = model.dt
if axs is None:
if to_plot_params:
axs = plt2.GridAxes(3, 3)
else:
if to_plot_internals:
axs = plt2.GridAxes(3, 3)
else:
if to_plot_choice:
axs = plt2.GridAxes(2, 2)
else:
axs = plt2.GridAxes(1, 2) # TODO: beautify ratios
rts = []
hs['rt'] = []
for dim_rel in range(consts.N_DIM):
# --- data_pred may not have all conditions, so concatenate the rest
# of the conditions so that the color scale is correct. Then also
# concatenate p_rt_ch_data_pred1 with zeros so that nothing is
# plotted in the concatenated.
evTest_cond_dim1 = np.concatenate([
evTest_cond_dim, evAll_cond_dim
], axis=0)
pTest_cond_rt_chFlat1 = np.concatenate([
pTest_cond_rt_chFlat, np.zeros_like(pAll_cond_rt_chFlat)
], axis=0)
if ev_dimRel_condDense_fr_dim_meanvar is None:
evModel_cond_dim = evAll_cond_dim
else:
if ev_dimRel_condDense_fr_dim_meanvar.ndim == 5:
evModel_cond_dim = npy(ev_dimRel_condDense_fr_dim_meanvar[
dim_rel][:, 0, :, 0])
else:
assert ev_dimRel_condDense_fr_dim_meanvar.ndim == 4
evModel_cond_dim = npy(ev_dimRel_condDense_fr_dim_meanvar[
dim_rel][:, 0, :])
pModel_cond_rt_chFlat = npy(pModel_dimRel_condDense_chFlat[dim_rel])
if to_plot_choice:
# --- Plot choice
ax = axs[0, dim_rel]
plt.sca(ax)
if to_combine_ch_irr_cond:
ev_cond_model1, p_rt_ch_model1 = combine_irr_cond(
dim_rel, evModel_cond_dim, pModel_cond_rt_chFlat
)
sim2d.plot_p_ch_vs_ev(ev_cond_model1, p_rt_ch_model1,
dim_rel=dim_rel, style='pred',
group_dcond_irr=None,
kw_plot=kw_plot_pred_ch,
cmap=lambda n: lambda v: [0., 0., 0.],
)
else:
sim2d.plot_p_ch_vs_ev(evModel_cond_dim, pModel_cond_rt_chFlat,
dim_rel=dim_rel, style='pred',
group_dcond_irr=group_dcond_irr,
kw_plot=kw_plot_pred,
cmap=cmaps[dim_rel]
)
hs, conds_irr = sim2d.plot_p_ch_vs_ev(
evTest_cond_dim1, pTest_cond_rt_chFlat1,
dim_rel=dim_rel, style='data_pred',
group_dcond_irr=group_dcond_irr,
cmap=cmaps[dim_rel],
kw_plot=kw_plot_data,
)
hs1 = [h[0] for h in hs]
odim = 1 - dim_rel
odim_name = consts.DIM_NAMES_LONG[odim]
legend_odim(conds_irr, hs1, odim_name)
sim2d.plot_p_ch_vs_ev(evTrain_cond_dim, pTrain_cond_rt_chFlat,
dim_rel=dim_rel, style='data_fit',
group_dcond_irr=group_dcond_irr,
cmap=cmaps[dim_rel],
kw_plot=kw_plot_data
)
plt2.detach_axis('x', np.amin(evTrain_cond_dim[:, dim_rel]),
np.amax(evTrain_cond_dim[:, dim_rel]))
ax.set_xlabel('')
ax.set_xticklabels([])
if dim_rel != 0:
plt2.box_off(['left'])
plt.yticks([])
ax.set_ylabel('P(%s choice)' % consts.CH_NAMES[dim_rel][1])
# --- Plot RT
ax = axs[int(to_plot_choice) + 0, dim_rel]
plt.sca(ax)
hs1, rts1 = sim2d.plot_rt_vs_ev(
evModel_cond_dim,
pModel_cond_rt_chFlat,
dim_rel=dim_rel, style='pred',
group_dcond_irr=group_dcond_irr,
dt=dt_model,
kw_plot=kw_plot_pred,
cmap=cmaps[dim_rel]
)
hs['rt'].append(hs1)
rts.append(rts1)
sim2d.plot_rt_vs_ev(evTest_cond_dim1, pTest_cond_rt_chFlat1,
dim_rel=dim_rel, style='data_pred',
group_dcond_irr=group_dcond_irr,
dt=dt,
cmap=cmaps[dim_rel],
kw_plot=kw_plot_data
)
sim2d.plot_rt_vs_ev(evTrain_cond_dim, pTrain_cond_rt_chFlat,
dim_rel=dim_rel, style='data_fit',
group_dcond_irr=group_dcond_irr,
dt=dt,
cmap=cmaps[dim_rel],
kw_plot=kw_plot_data
)
plt2.detach_axis('x', np.amin(evTrain_cond_dim[:, dim_rel]),
np.amax(evTrain_cond_dim[:, dim_rel]))
if dim_rel != 0:
ax.set_ylabel('')
plt2.box_off(['left'])
plt.yticks([])
ax.set_xlabel(consts.DIM_NAMES_LONG[dim_rel].lower() + ' strength')
if dim_rel == 0:
ax.set_ylabel('RT (s)')
if to_plot_internals:
for ch1 in range(consts.N_CH):
ch0 = dim_rel
ax = axs[3 + ch1, dim_rel]
plt.sca(ax)
ch_flat = consts.ch_by_dim2ch_flat(np.array([ch0, ch1]))
model.tnds[ch_flat].plot_p_tnd()
ax.set_xlabel('')
ax.set_xticklabels([])
ax.set_yticks([0, 1])
if ch0 > 0:
ax.set_yticklabels([])
ax.set_ylabel(r"$\mathrm{P}(T^\mathrm{n} \mid"
" \mathbf{z}=[%d,%d])$"
% (ch0, ch1))
ax = axs[5, dim_rel]
plt.sca(ax)
if hasattr(model.dtb, 'dtb1ds'):
model.dtb.dtb1ds[dim_rel].plot_bound(color='k')
plt2.sameaxes(axs[-1, :consts.N_DIM], xy='y')
if to_plot_params:
ax = axs[0, -1]
plt.sca(ax)
model.plot_params()
return axs, rts, hs
def plot_rt_distrib(
n_cond_rt_ch: np.ndarray,
ev_cond_dim: np.ndarray,
abs_cond=True,
lump_wrong=True,
dt=consts.DT,
colors=None,
alpha=1.,
alpha_face=0.5,
smooth_sigma_sec=0.05,
to_normalize_max=False,
to_cumsum=False,
to_exclude_last_frame=True,
to_skip_zero_trials=False,
label='',
# to_exclude_bins_wo_trials=10,
kw_plot=(),
fig=None,
axs=None,
to_use_sameaxes=True,
):
"""
:param n_cond_rt_ch:
:param ev_cond_dim:
:param abs_cond:
:param lump_wrong:
:param dt:
:param gs:
:param colors:
:param alpha:
:param smooth_sigma_sec:
:param kw_plot:
:param axs:
:return: axs, p_cond01__rt_ch01, p_cond01__rt_ch01_sm, hs
"""
if colors is None:
colors = ['red', 'blue']
elif type(colors) is str:
colors = [colors] * 2
else:
assert len(colors) == 2
nt = n_cond_rt_ch.shape[1]
t_all = np.arange(nt) * dt + dt
out = np.meshgrid(np.unique(ev_cond_dim[:, 0]),
np.unique(ev_cond_dim[:, 1]),
np.arange(nt),
np.arange(2),
np.arange(2),
indexing='ij')
cond0, cond1, fr, ch0, ch1 = [v.flatten() for v in out]
from copy import deepcopy
n0 = deepcopy(n_cond_rt_ch)
if to_exclude_last_frame:
n0[:, -1, :] = 0.
n0 = n0.flatten()
def sign_cond(v):
v1 = np.sign(v)
v1[v == 0] = 1
return v1
if abs_cond:
ch0 = consts.ch_bool2sign(ch0)
ch1 = consts.ch_bool2sign(ch1)
# 1 = correct, -1 = wrong
ch0 = sign_cond(cond0) * ch0
ch1 = sign_cond(cond1) * ch1
cond0 = np.abs(cond0)
cond1 = np.abs(cond1)
ch0 = consts.ch_sign2bool(ch0).astype(np.int)
ch1 = consts.ch_sign2bool(ch1).astype(np.int)
else:
raise ValueError()
if lump_wrong:
# treat all choices as correct when cond == 0
ch00 = ch0 | (cond0 == 0)
ch10 = ch1 | (cond1 == 0)
ch0 = (ch00 & ch10)
ch1 = np.ones_like(ch00, dtype=np.int)
cond_dim = np.stack([cond0, cond1], -1)
conds = []
dcond_dim = []
for cond in cond_dim.T:
conds1, dcond1 = np.unique(cond, return_inverse=True)
conds.append(conds1)
dcond_dim.append(dcond1)
dcond_dim = np.stack(dcond_dim)
n_cond01_rt_ch01 = npg.aggregate(
[*dcond_dim, fr, ch0, ch1], n0, 'sum',
[*(np.amax(dcond_dim, 1) + 1), nt, consts.N_CH, consts.N_CH])
p_cond01__rt_ch01 = np2.nan2v(n_cond01_rt_ch01 / n_cond01_rt_ch01.sum(
(2, 3, 4), keepdims=True))
n_conds = p_cond01__rt_ch01.shape[:2]
if axs is None:
axs = plt2.GridAxes(
n_conds[1],
n_conds[0],
left=0.6,
right=0.3,
bottom=0.45,
top=0.74,
widths=[1],
heights=[1],
wspace=0.04,
hspace=0.04,
)
kw_label = {
'fontsize': 12,
}
pad = 8
axs[0, 0].set_title('strong\nmotion', pad=pad, **kw_label)
axs[0, -1].set_title('weak\nmotion', pad=pad, **kw_label)
axs[0, 0].set_ylabel('strong\ncolor', labelpad=pad, **kw_label)
axs[-1, 0].set_ylabel('weak\ncolor', labelpad=pad, **kw_label)
if smooth_sigma_sec > 0:
from scipy import signal, stats
sigma_fr = smooth_sigma_sec / dt
width = np.ceil(sigma_fr * 2.5).astype(np.int)
kernel = stats.norm.pdf(np.arange(-width, width + 1), 0, sigma_fr)
kernel = np2.vec_on(kernel, 2, 5)
p_cond01__rt_ch01_sm = signal.convolve(p_cond01__rt_ch01,
kernel,
mode='same')
else:
p_cond01__rt_ch01_sm = p_cond01__rt_ch01.copy()
if to_cumsum:
p_cond01__rt_ch01_sm = np.cumsum(p_cond01__rt_ch01_sm, axis=2)
if to_normalize_max:
p_cond01__rt_ch01_sm = np2.nan2v(
p_cond01__rt_ch01_sm /
np.amax(np.abs(p_cond01__rt_ch01_sm), (2, 3, 4), keepdims=True))
n_row = n_conds[1]
n_col = n_conds[0]
for dcond0 in range(n_conds[0]):
for dcond1 in range(n_conds[1]):
row = n_row - 1 - dcond1
col = n_col - 1 - dcond0
ax = axs[row, col] # type: plt.Axes
for ch0 in [0, 1]:
for ch1 in [0, 1]:
if lump_wrong and ch1 == 0:
continue
p1 = p_cond01__rt_ch01_sm[dcond0, dcond1, :, ch0, ch1]
kw = {
'linewidth': 1,
'color': colors[ch1],
'alpha': alpha,
'zorder': 1,
**dict(kw_plot)
}
y = p1 * consts.ch_bool2sign(ch0)
p_cond01__rt_ch01_sm[dcond0, dcond1, :, ch0, ch1] = y
if to_skip_zero_trials and np.sum(np.abs(y)) < 1e-2:
h = None
else:
h = ax.plot(
t_all,
y,
label=label if ch0 == 1 and ch1 == 1 else None,
**kw)
ax.fill_between(t_all,
0,
y,
ec='None',
fc=kw['color'],
alpha=alpha_face,
zorder=-1)
plt2.box_off(ax=ax)
ax.axhline(0, color='k', linewidth=0.5)
ax.set_yticklabels([])
if row < n_row - 1 or col > 0:
ax.set_xticklabels([])
ax.set_xticks([])
plt2.box_off(['bottom'], ax=ax)
else:
ax.set_xlabel('RT (s)')
# if col > 0:
ax.set_yticks([])
ax.set_yticklabels([])
plt2.box_off(['left'], ax=ax)
plt2.detach_axis('x', 0, 5, ax=ax)
if to_use_sameaxes:
plt2.sameaxes(axs)
axs[-1, 0].set_xlabel('RT (s)')
return axs, p_cond01__rt_ch01, p_cond01__rt_ch01_sm, h
def main_plot_models_on_columns(i_subjs=(0, 1), axs=None, coef='slope'):
if coef not in ['slope', 'bias']:
raise ValueError()
n_subj = len(i_subjs)
# ---- Load goodness-of-fit
def load_gof(i_subj, buffix_dur):
fix = fix0_pre + [('buffix', buffix_dur)] + fix0_post
dict_cache, subdir = main_fit(i_subj,
fit_mode='dict_cache_only',
fix=fix)
file = locfile.get_file('tab',
'best_loss',
dict_cache,
ext='.csv',
subdir=subdir)
import csv, os
rows = None
if not os.path.exists(file):
print('File absent - returning NaN: %s' % file)
gof = np.nan
return gof
with open(file, 'r') as csvfile:
reader = csv.DictReader(csvfile)
for row in reader:
if rows is None:
rows = {k: [row[k]] for k in row.keys()}
else:
for k in row.keys():
rows[k].append(row[k])
gof_kind = 'loss_NLL_test'
igof = rows['name'].index(gof_kind)
gof = float(rows[' value'][igof][3:])
return gof
bufdurs = [0., bufdur_best, 1.2]
nbufdurs = len(bufdurs)
gofs = np.zeros([nbufdurs, n_subj])
for i_subj in range(n_subj):
for idur, bufdur in enumerate(bufdurs):
gofs[idur, i_subj] = load_gof(i_subj, bufdur)
gofs = gofs - np.nanmin(gofs, 0, keepdims=True)
# ---- Load coefs
fixs = [
('buffer+serial', [
('buffix', bufdur_best),
]),
('parallel', [
('buffix', 1.2),
]),
('serial', [
('buffix', 0.),
]),
]
model_names = [f[0] for f in fixs]
n_models = len(model_names)
models = np.empty([n_models, n_subj], dtype=np.object)
dict_caches = np.empty([n_models, n_subj], dtype=np.object)
ds = np.empty([n_models, n_subj], dtype=np.object)
datas = np.empty([n_subj], dtype=np.object)
subdir = ','.join(fix0) + '+buffix=' + ','.join(
[('%1.2f' % v) for v in [0., bufdur_best, 1.2]])
for i_subj in range(n_subj):
for i_model, (name, fix1) in enumerate(fixs):
fix = fix0_pre + fix1 + fix0_post
model, data, dict_cache, d, _ = main_fit(i_subj,
fit_mode='load',
fix=fix)
models[i_model, i_subj] = model
datas[i_subj] = data
dict_caches[i_model, i_subj] = dict_cache
ds[i_model, i_subj] = d
# ---- Plot coefs
kw_plots = {
'serial': {
'linestyle': '-'
},
'buffer+serial': {
'linestyle': '-'
},
'parallel': {
'linestyle': '-'
},
}
def beautify(row, col, axs):
plt.sca(axs[row, col])
plt.xlim(xmax=1.2 * 1.05)
plt2.detach_axis('x', 0, 1.2)
plt2.box_off()
beautify_ticks(axs[row, col],
add_ticklabel=(row == n_row - 1) and (col == 0))
if col > 0:
plt.ylabel('')
import matplotlib as mpl
for i_subj in range(n_subj):
fig = plt.figure(figsize=[2.25 * n_models, 5.25])
n_row = 3
n_col = n_models
n_col = len(fixs)
n_row = 2
axs = plt2.GridAxes(n_row,
n_col,
top=.7,
left=1.15,
right=0.1,
bottom=.5,
widths=[2],
wspace=0.35,
heights=[1.5],
hspace=[0.2])
# ---- Plot slopes
title_model = {
'buffer+serial': 'parallel followed\nby serial',
'parallel': 'strictly parallel',
'serial': 'strictly serial'
}
for ii_model, model_name in enumerate(
# ['buffer+serial', 'serial']):
['buffer+serial', 'parallel', 'serial']):
i_model = model_names.index(model_name)
fix1 = fixs[i_model]
model = models[i_model, i_subj]
data = datas[i_subj]
name = model_names[i_model]
plot_coefs_dur_odif_pred_data(
data,
model,
axs=axs[:2, :][:, [ii_model]],
to_plot_data=True,
kw_plot_model=kw_plots[model_name],
coefs_to_plot=[1 if coef == 'slope' else 0],
add_rowtitle=False)
for row in range(2):
plt.sca(axs[row, ii_model])
plt2.sameaxes(axs[row, :], xy='y')
beautify(row, ii_model, axs)
if ii_model > 0:
plt.gca().set_yticklabels([])
if row == 0:
plt.title(title_model[model_name])
plt.xlabel('')
else:
plt.title('')
if ii_model == 0:
plt.xlabel('stimulus duration (s)')
else:
plt.xlabel('')
bnd = axs[0, 1].get_position().bounds
subj = consts.SUBJS['VD'][i_subj]
plt.suptitle(subj, x=bnd[0] + bnd[2] / 2)
rowtitles = [
r'Motion sensitivity ($\beta$)',
r'Color sensitivity ($\beta$)',
] if coef == 'slope' else [
r'Motion bias ($\beta$)',
r'Color bias ($\beta$)',
]
for row, title in enumerate(rowtitles):
plt.sca(axs[row, 0])
plt.ylabel(title)
dict_cache = dict_caches[0, 0]
if 'fix' in dict_cache:
dict_cache.pop('fix')
dict_cache['sbj'] = subj
dict_cache['coef'] = coef
for ext in ['.png', '.pdf']:
file = locfile.get_file_fig('coefs_dur_odif',
dict_cache,
subdir=subdir,
ext=ext)
plt.savefig(file, dpi=300)
print('Saved to %s' % file)
print('--')
return models, data
def main_fit(
subjs=None,
skip_fit_if_absent=False,
bufdur=None,
bufdur_best=0.12,
bufdurs_sim=None,
bufdurs_fit=None,
loss_kind='NLL',
plot_kind='line_sim_fit',
fix_post=None,
seed_sims=(0, ),
err_kind='std',
base=10,
):
# if torch.cuda.is_available():
# torch.set_default_tensor_type(torch.cuda.FloatTensor)
torch.set_num_threads(1)
torch.set_default_dtype(torch.double)
dict_res_add = {}
parad = 'VD'
seed_sims = np.array(seed_sims)
if subjs is None:
subjs = consts.SUBJS_VD
if bufdur is not None:
bufdurs_sim = list({bufdur_best}.union({bufdur}))
bufdurs_fit = list({bufdur_best}.union({bufdur}))
else:
if bufdurs_sim is None:
bufdurs_sim = bufdurs0 #
if bufdurs_fit is None:
bufdurs_fit = bufdurs0
else:
if bufdurs_fit is None:
bufdurs_fit0 = [0., bufdur_best, 1.2]
if bufdurs_sim[0] in bufdurs_fit0:
bufdurs_fit = copy(bufdurs_fit0)
else:
bufdurs_fit = bufdurs_fit0[:2] + bufdurs_sim + [1.2]
n_dur_sim = len(bufdurs_sim)
n_dur_fit = len(bufdurs_fit)
# DEF: losses[seed, subj, simSerDurPar, fitDurs]
size_all = [len(seed_sims), len(subjs), n_dur_sim, n_dur_fit]
losses0 = np.zeros(size_all) + np.nan
ns = np.zeros(size_all) + np.nan
ks0 = np.zeros(size_all) + np.nan
for i_seed, seed_sim in enumerate(seed_sims):
for i_subj, subj in enumerate(subjs):
for i_fit, bufdur_fit in enumerate(bufdurs_fit):
for i_sim, bufdur_sim in enumerate(bufdurs_sim):
d, dict_fit_sim, dict_subdir_sim = get_fit_sim(
subj,
seed_sim,
bufdur_sim,
bufdur_fit,
parad=parad,
skip_fit_if_absent=skip_fit_if_absent,
fix_post=fix_post,
)
if d is not None:
losses0[i_seed, i_subj, i_sim,
i_fit] = d['loss_NLL_test']
ns[i_seed, i_subj, i_sim, i_fit] = d['loss_ndata_test']
ks0[i_seed, i_subj, i_sim, i_fit] = d['loss_nparam']
ks = copy(ks0)
if loss_kind == 'BIC':
losses = ks * np.log(ns) + 2 * losses0
# REF: Kass, Raftery 1995 https://doi.org/10.2307%2F2291091
# https://en.wikipedia.org/wiki/Bayesian_information_criterion#Gaussian_special_case
thres_strong = 10. / np.log(base)
elif loss_kind == 'NLL':
losses = copy(losses0)
thres_strong = np.log(100) / np.log(base)
else:
raise ValueError('Unsupported loss_kind: %s' % loss_kind)
#%%
losses = losses / np.log(base)
if plot_kind == 'loss_serpar':
n_row = len(subjs)
n_col = 1
axs = plt2.GridAxes(n_row, n_col, left=1, widths=2, bottom=0.75)
for i_subj, subj in enumerate(subjs):
plt.sca(axs[i_subj, 0])
loss_dur_sim_ser_fit = losses[:, i_subj, :, 0].mean(0)
loss_dur_sim_par_fit = losses[:, i_subj, :, -1].mean(0)
dloss_serpar = loss_dur_sim_par_fit - loss_dur_sim_ser_fit
plt.bar(bufdurs_sim, dloss_serpar, color='k')
plt.axhline(0, color='k', linestyle='-', linewidth=0.5)
if i_subj < len(subjs) - 1:
plt2.box_off(['top', 'right', 'bottom'])
plt.xticks([])
else:
plt2.box_off()
plt.xticks(np.arange(0, 1.4, 0.2), ['0\nser'] + [''] * 2 +
['0.6'] + [''] * 2 + ['1.2\npar'])
plt2.detach_axis('x', 0, 1.2)
vdfit.patch_chance_level()
plt.sca(axs[0, 0])
plt.title('Support for Serial\n'
'($\mathcal{L}_\mathrm{ser} - \mathcal{L}_\mathrm{par}$)')
plt.sca(axs[-1, 0])
plt.xlabel('true buffer duration (s)')
plt2.rowtitle(consts.SUBJS_VD, axs)
elif plot_kind == 'imshow_ser_buf_par':
n_row = len(subjs)
n_col = 1
axs = plt2.GridAxes(n_row,
n_col,
left=1,
widths=2,
heights=2,
bottom=0.75)
for i_subj, subj in enumerate(subjs):
plt.sca(axs[i_subj, 0])
dloss = losses[:, i_subj, :, :].mean(0)
dloss = dloss - np.diag(dloss)[:, None]
plt.imshow(dloss, cmap='bwr')
cmax = np.amax(np.abs(dloss))
plt.clim(-cmax, +cmax)
plt.sca(axs[0, 0])
plt.sca(axs[-1, 0])
plt.xlabel('true buffer duration (s)')
plt.ylabel('model buffer duration (s)')
plt2.rowtitle(consts.SUBJS_VD, axs)
elif plot_kind == 'line_sim_fit':
bufdur_bests = np.array([get_bufdur_best(subj) for subj in subjs])
i_bests = np.array([
list(bufdurs_fit).index(bufdur_best)
for bufdur_best in bufdur_bests
])
i_subjs = np.arange(len(subjs))
loss_sim_best_fit_best = losses[:, i_subjs, i_bests, i_bests]
loss_sim_best_fit_rest = (losses[:, i_subjs, i_bests, :] -
loss_sim_best_fit_best[:, :, None])
mean_loss_sim_best_fit_rest = np.mean(loss_sim_best_fit_rest, 0)
if err_kind == 'std':
err_loss_sim_best_fit_rest = np.std(loss_sim_best_fit_rest, 0)
elif err_kind == 'sem':
err_loss_sim_best_fit_rest = np2.sem(loss_sim_best_fit_rest, 0)
n_dur = len(bufdurs_fit)
loss_sim_rest_fit_best = np.swapaxes(
np.stack([
losses[:, i_subj, np.arange(n_dur), i_best] -
losses[:, i_subj,
np.arange(n_dur),
np.arange(n_dur)]
for i_subj, i_best in zip(i_subjs, i_bests)
]), 0, 1)
mean_loss_sim_rest_fit_best = np.mean(loss_sim_rest_fit_best, 0)
if err_kind == 'std':
err_loss_sim_rest_fit_best = np.std(loss_sim_rest_fit_best, 0)
elif err_kind == 'sem':
err_loss_sim_rest_fit_best = np2.sem(loss_sim_rest_fit_best, 0)
dict_res_add['err'] = err_kind
n_row = 2
n_col = len(subjs)
axs = plt2.GridAxes(n_row,
n_col,
left=1.15,
right=0.1,
widths=2,
heights=1.5,
wspace=0.35,
hspace=0.5,
top=0.25,
bottom=0.6)
for row, (m, s, xlabel, ylabel) in enumerate([
(mean_loss_sim_best_fit_rest, err_loss_sim_best_fit_rest,
'model buffer capacity (s)',
(r'$-\mathrm{log}_{%g}\mathrm{BF}$ given simulated' +
'\nbest duration data') % base),
(mean_loss_sim_rest_fit_best, err_loss_sim_rest_fit_best,
'simulated data buffer capacity (s)',
(r'$-\mathrm{log}_{%g}\mathrm{BF}$ of the' +
'\nbest duration model') % base)
]):
for i_subj, subj in enumerate(subjs):
ax = axs[row, i_subj]
plt.sca(ax)
gs1 = axs.gs[row * 2 + 1, i_subj * 2 + 1]
bax = vdfit.breakaxis(gs1)
bax.axs[1].errorbar(bufdurs0,
m[i_subj, :],
yerr=s[i_subj, :],
color='k',
marker='.',
linewidth=0.75,
elinewidth=0.5,
markersize=3)
m1 = copy(m[i_subj, :])
m1[3:] = np.nan
s1 = copy(s[i_subj, :])
s1[3:] = np.nan
bax.axs[0].errorbar(bufdurs0,
m1,
yerr=s1,
color='k',
marker='.',
linewidth=0.75,
elinewidth=0.5,
markersize=3)
ax1 = bax.axs[1] # type: plt.Axes
plt.sca(ax1)
vdfit.patch_chance_level(level=thres_strong, signs=[-1, 1])
plt.axhline(0, color='k', linestyle='--', linewidth=0.5)
vdfit.beautify_ticks(ax1, )
vdfit.beautify(ax1)
ax1.set_yticks([0, 20])
if i_subj > 0:
ax1.set_yticklabels([])
if row == 0:
ax1.set_xticklabels([])
ax1 = bax.axs[0] # type: plt.Axes
plt.sca(ax1)
ax1.set_yticks([40, 200])
if i_subj > 0:
ax1.set_yticklabels([])
plt.sca(ax)
plt2.box_off('all')
if row == 0:
plt.title(consts.SUBJS['VD'][i_subj])
if i_subj == 0:
plt.xlabel(xlabel, labelpad=8 if row == 0 else 20)
plt.ylabel(ylabel, labelpad=30)
plt2.sameaxes(bax.axs, xy='x')
elif plot_kind == 'bar_sim_fit':
i_best = bufdurs_fit.index(bufdur_best)
loss_sim_best_fit_best = losses[0, :, i_best, i_best]
loss_sim_best_fit_rest = np2.nan2v(losses[0, :, i_best, :] -
loss_sim_best_fit_best[:, None])
n_dur = len(bufdurs_fit)
loss_sim_rest_fit_best = np2.nan2v(
losses[0, :, np.arange(n_dur), i_best] -
losses[0, :, np.arange(n_dur),
np.arange(n_dur)]).T
n_row = 2
n_col = len(subjs)
axs = plt2.GridAxes(n_row,
n_col,
left=1,
widths=3,
right=0.25,
heights=1,
wspace=0.3,
hspace=0.6,
top=0.25,
bottom=0.6)
for i_subj, subj in enumerate(subjs):
def plot_bars(gs1, bufdurs, losses1, add_xticklabel=True):
i_break = np.amax(np.nonzero(np.array(bufdurs) < 0.3)[0])
bax = brokenaxes(
subplot_spec=gs1,
xlims=((-1, i_break + 0.5), (i_break + 0.5,
len(bufdurs) - 0.5)),
ylims=((-3, 20), (20, 1250 / 5)),
height_ratios=(50 / 100, 500 / (1250 - 100)),
hspace=0.15,
wspace=0.075,
d=0.005,
)
bax.bar(np.arange(len(bufdurs)), losses1[i_subj, :], color='k')
ax11 = bax.axs[3] # type: plt.Axes
ax11.set_xticks([bufdurs.index(0.6), bufdurs.index(1.2)])
if i_subj == 0 and add_xticklabel:
ax11.set_xticklabels(['0.6', '1.2'])
else:
ax11.set_xticklabels([])
ax00 = bax.axs[0] # type: plt.Axes
ax00.set_yticks([500, 1000])
ax10 = bax.axs[2] # type: plt.Axes
ax10.set_yticks([0, 50])
plt.sca(ax10)
plt2.detach_axis('x', amin=-0.4, amax=i_break + 0.5)
for ax in [ax10, ax11]:
plt.sca(ax)
plt.axhline(0, linewidth=0.5, color='k', linestyle='--')
for sign in [-1, 1]:
plt.axhline(sign * thres_strong,
linewidth=0.5,
color='silver',
linestyle='--')
ax10.set_xticks([bufdurs.index(0.), bufdurs.index(0.2)])
if i_subj == 0:
if add_xticklabel:
ax10.set_xticklabels(['0', '0.2'])
else:
ax10.set_xticklabels([])
else:
ax10.set_yticklabels([])
ax10.set_xticklabels([])
ax00.set_yticklabels([])
return bax
bax = plot_bars(axs.gs[1, i_subj * 2 + 1],
bufdurs_fit,
loss_sim_best_fit_rest,
add_xticklabel=False)
ax = axs[0, i_subj]
plt.sca(ax)
plt2.box_off('all')
plt.title(consts.SUBJS['VD'][consts.SUBJS['VD'].index(subj)])
if i_subj == 0:
ax.set_ylabel(
'misfit to simulated\nbest duration data\n'
r'($\Delta$BIC)',
labelpad=35)
ax.set_xlabel('model buffer duration (s)', labelpad=8)
plot_bars(axs.gs[3, i_subj * 2 + 1],
bufdurs_sim,
loss_sim_rest_fit_best,
add_xticklabel=True)
ax = axs[1, i_subj]
plt.sca(ax)
plt2.box_off('all')
if i_subj == 0:
ax.set_ylabel(
'misfit of\nbest duration model\n'
r'($\Delta$BIC)',
labelpad=35)
ax.set_xlabel('simulated data buffer duration (s)',
labelpad=20)
elif plot_kind == 'bar_ser_buf_par':
n_row = len(subjs)
n_col = 1
axs = plt2.GridAxes(n_row * n_dur_sim,
n_col,
left=1.25,
widths=1.5,
right=0.25,
heights=0.4,
hspace=[0.15] * (n_dur_sim - 1) + [0.5] + [0.15] *
(n_dur_sim - 1),
top=0.6,
bottom=0.5)
row = -1
for i_subj, subj in enumerate(subjs):
for i_sim in range(n_dur_sim):
row += 1
plt.sca(axs[row, 0])
loss1 = losses[:, i_subj, i_sim, :].mean(0)
dloss = loss1 - loss1[i_sim]
x = np.arange(n_dur_fit)
for x1, dloss1 in zip(x, dloss):
plt.bar(x1, dloss1, color='r' if dloss1 > 0 else 'b')
plt.axhline(0, color='k', linewidth=0.5)
vdfit.patch_chance_level(6.)
plt.ylim([-100, 100])
plt.yticks([-100, 0, 100], [''] * 2)
plt.ylabel('%g' % bufdurs_sim[i_sim],
rotation=0,
va='center',
ha='right')
if i_sim == 0:
plt.title(subj)
if i_sim < n_dur_sim - 1 or i_subj < len(subjs) - 1:
plt2.box_off(['top', 'bottom', 'right'])
plt.xticks([])
else:
plt2.box_off(['top', 'right'])
plt.sca(axs[-1, 0])
plt.xticks(np.arange(n_dur_sim), ['%g' % v for v in bufdurs_sim])
plt2.detach_axis('x', 0, n_dur_sim - 1)
plt.xlabel('model buffer duration (s)', fontsize=10)
c = axs[-1, 0].get_position().corners()
plt.figtext(x=0.15,
y=np.mean(c[:2, 1]),
fontsize=10,
s='true\nbuffer\nduration\n(s)',
rotation=0,
va='center',
ha='center')
plt.figtext(
x=(1.25 + 1.5 / 2) / (1.25 + 1.5 + 0.25),
y=0.98,
s=r'$\mathrm{BIC} - \mathrm{BIC}_\mathrm{true}$',
ha='center',
va='top',
fontsize=12,
)
if plot_kind != 'None':
dict_res = deepcopy(dict_fit_sim) # noqa
for k in ['sbj', 'prd']:
dict_res.pop(k)
dict_res = {**dict_res, 'los': loss_kind}
dict_res.update(dict_res_add)
for ext in ['.png', '.pdf']:
file = locfile.get_file_fig(plot_kind,
dict_res,
subdir=dict_subdir_sim,
ext=ext) # noqa
plt.savefig(file, dpi=300)
print('Saved to %s' % file)
#%%
print('--')
return losses
def main_plot_across_models(
i_subjs=i_subjs,
axs=None,
models_incl=('buffer+serial', ),
base=10,
**kwargs,
):
n_subj = len(i_subjs)
# ---- Load goodness-of-fit
def load_gof(i_subj, buffix_dur):
fix = fix0_pre + [('buffix', buffix_dur)] + fix0_post
dict_cache, subdir = main_fit(i_subj,
fit_mode='dict_cache_only',
fix=fix,
**kwargs)
file = locfile.get_file('tab',
'best_loss',
dict_cache,
ext='.csv',
subdir=subdir)
import csv, os
rows = None
if not os.path.exists(file):
print('File absent - returning NaN: %s' % file)
gof = np.nan
return gof
with open(file, 'r') as csvfile:
reader = csv.DictReader(csvfile)
for row in reader:
if rows is None:
rows = {k: [row[k]] for k in row.keys()}
else:
for k in row.keys():
rows[k].append(row[k])
gof_kind = 'loss_NLL_test'
igof = rows['name'].index(gof_kind)
gof = float(rows[' value'][igof][3:])
return gof
nbufdurs = len(bufdurs0)
gofs = np.zeros([nbufdurs, n_subj])
for i_subj in range(n_subj):
for idur, bufdur in enumerate(bufdurs0):
gofs[idur, i_subj] = load_gof(i_subj, bufdur)
gofs = gofs - np.nanmin(gofs, 0, keepdims=True)
gofs = gofs / np.log(base)
# ---- Load slopes
fixs = [
('buffer+serial', [
('buffix', bufdur_best),
]),
# (
# 'parallel', [
# ('buffix', 1.2),
# ]),
# (
# 'serial', [
# ('buffix', 0.),
# ]),
]
model_names = [f[0] for f in fixs]
n_models = len(model_names)
models = np.empty([n_models, n_subj], dtype=np.object)
dict_caches = np.empty([n_models, n_subj], dtype=np.object)
ds = np.empty([n_models, n_subj], dtype=np.object)
datas = np.empty([n_subj], dtype=np.object)
kw_plots = {
'serial': {
'linestyle': '--'
},
'buffer+serial': {
'linestyle': '-'
},
'parallel': {
'linestyle': ':'
},
}
subdir = ','.join(fix0) + '+buffix=' + ','.join(
[('%1.2f' % v) for v in [0., bufdur_best, 1.2]])
for i_subj in range(n_subj):
for i_model, (name, fix1) in enumerate(fixs):
fix = fix0_pre + fix1 + fix0_post
# try:
model, data, dict_cache, d, _ = main_fit(i_subj,
fit_mode='load',
fix=fix,
**kwargs)
# except RuntimeError:
# model = None
# data = None
# dict_cache = None
# d = None
models[i_model, i_subj] = model
datas[i_subj] = data
dict_caches[i_model, i_subj] = dict_cache
ds[i_model, i_subj] = d
# ---- Plot goodness-of-fit
if axs is None:
n_row = 2 + len(models_incl)
n_row_gs = n_row + 2
n_col = n_subj
axs = plt2.GridAxes(3,
2,
top=.3,
left=1.15,
right=0.1,
bottom=.5,
widths=[2],
wspace=0.35,
heights=[1.5, 1.5, 1.5],
hspace=[0.2, 0.9])
for i_subj in range(n_subj):
ax = axs[-1, i_subj]
plt.sca(ax)
plt2.box_off('all')
gs1 = axs.gs[-2, i_subj * 2 + 1]
bax = breakaxis(gs1)
ax0 = bax.axs[0] # type: plt.Axes
ax1 = bax.axs[1] # type: plt.Axes
ax0.plot(bufdurs0[:3],
gofs[:3, i_subj],
'k.-',
linewidth=0.75,
markersize=4.5)
ax1.plot(bufdurs0,
gofs[:, i_subj],
'k.-',
linewidth=0.75,
markersize=4.5)
plt.sca(ax1)
patch_chance_level(level=np.log(100.) / np.log(base), signs=[-1, 1])
plt.axhline(0, color='k', linestyle='--', linewidth=0.5)
beautify(ax1)
beautify_ticks(ax1, add_ticklabel=True) # i_subj == 0)
plt2.sameaxes([ax0, ax1], xy='x')
ax1.set_yticks([0, 20])
ax0.set_yticks([40, 200])
if i_subj == 0:
plt.sca(ax1)
plt.xlabel('buffer capacity (s)')
plt.sca(ax)
plt.ylabel(r'$-\mathrm{log}_{10}\mathrm{BF}$', labelpad=27)
else:
ax0.set_yticklabels([])
ax1.set_yticklabels([])
plt.sca(axs[0, i_subj])
beautify_ticks(axs[0, i_subj], add_ticklabel=False)
beautify_ticks(axs[1, i_subj], add_ticklabel=True)
plt.sca(axs[-2, 0])
plt.xlabel('stimulus duration (s)')
# ---- Plot slopes
for i_subj in range(n_subj):
hss = []
for model_name in models_incl:
i_model = model_names.index(model_name)
fix1 = fixs[i_model]
model = models[i_model, i_subj]
data = datas[i_subj]
name = model_names[i_model]
if model is not None:
_, hs = plot_coefs_dur_odif_pred_data(
data,
model,
axs=axs[:2, [i_subj]],
# to_plot_data=True,
to_plot_data=True,
kw_plot_model=kw_plots[model_name],
coefs_to_plot=[1],
add_rowtitle=False)
hss.append(hs)
if i_subj == 0:
# hss[0] = hs['pred'|'data'][coef, dim, dif]
hs1 = hss[0]['data'] # type: np.ndarray
for dim in range(hs1.shape[1] - 1, -1, -1):
hs = []
for dif in range(hs1.shape[2]):
hs.append(hs1[0, dim, dif])
odim = 1 - dim
odim_name = consts.DIM_NAMES_LONG[odim]
plt.sca(axs[dim, 0])
plt.legend(hs, [
'weak ' + odim_name.lower(), 'strong ' + odim_name.lower()
],
bbox_to_anchor=[0., 0., 1., 1.],
handletextpad=0.35,
handlelength=0.5,
labelspacing=0.3,
borderpad=0.,
frameon=False,
loc='upper left')
for row in range(axs.shape[0]):
beautify(axs[row, i_subj])
if i_subj > 0:
plt.ylabel('')
plt2.sameaxes(axs[-1, :])
ax = axs[-1, -1]
ax.set_yticklabels([])
for i_subj in range(n_subj):
plt.sca(axs[0, i_subj])
plt.title(consts.SUBJS['VD'][i_subj])
for row in range(1, n_row):
plt.sca(axs[row, i_subj])
plt.title('')
for row, title in enumerate([
r'Motion sensitivity ($\beta$)',
r'Color sensitivity ($\beta$)',
]):
plt.sca(axs[row, 0])
plt.ylabel(title)
dict_cache = dict_caches[0, 0]
for k in ['fix', 'sbj']:
if k in dict_cache:
dict_cache.pop(k)
for ext in ['.pdf', '.png']:
file = locfile.get_file_fig('coefs_dur_odif_sbjs',
dict_cache,
subdir=subdir,
ext=ext)
from lib.pylabyk.cacheutil import mkdir4file
mkdir4file(file)
plt.savefig(file, dpi=300)
print('Saved to %s' % file)
print('--')
return gofs, models, data