def plot_weights(ax, Coefs, prds=None, xlim=None, xlab=tlab,
ylab='unit coefficient', title='', ytitle=1.04):
"""Plot decoding weights."""
# Unstack dataframe with results.
lCoefs = pd.DataFrame(Coefs.unstack().unstack(), columns=['coef'])
lCoefs['time'] = lCoefs.index.get_level_values(0)
lCoefs['value'] = lCoefs.index.get_level_values(1)
lCoefs['uid'] = lCoefs.index.get_level_values(2)
lCoefs.index = np.arange(len(lCoefs.index))
# Plot time series.
sns.tsplot(lCoefs, time='time', value='coef', unit='value',
condition='uid', ax=ax)
# Add chance level line and stimulus periods.
putil.add_chance_level(ax=ax, ylevel=0)
putil.plot_periods(prds, ax=ax)
# Set axis limits.
xlim = xlim if xlim is not None else tlim
putil.set_limits(ax, xlim)
# Format plot.
putil.set_labels(ax, xlab, ylab, title, ytitle)
putil.hide_legend(ax)
def plot_prd_spike_count(spk_cnt, prd, sps, fig):
"""Plot spike count results for period."""
# Init axes.
ax_total, ax_per_tr, ax_mean = putil.sps_add_axes(fig, sps, 1, 3)
# Plot total spike count histogram.
plot_spike_count_hist(spk_cnt.unstack(), prd + ', all trials', ax_total)
# Per trial.
cols = sns.color_palette('Spectral', len(spk_cnt.index))
for (itr, nspk), col in zip(spk_cnt.iterrows(), cols):
plot_spike_count_hist(nspk, prd + ', per trial', ax_per_tr,
False, 1.0, col, 1)
# Mean over time.
x = spk_cnt.index
y = spk_cnt.median(1)
pplot.lines(x, y, xlim=[x.min(), x.max()], xlab='trial index',
ylab='mean spike counts', title='median over time', ax=ax_mean)
# Format figures.
for ax in (ax_total, ax_per_tr):
putil.set_limits(ax, xlim=[-0.5, None], ylim=[0, None])
return ax_total, ax_per_tr, ax_mean
def plot_mean_rates(mRates,
aa_res_dir,
tasks=None,
task_lbls=None,
baseline=None,
xlim=None,
ylim=None,
ci=68,
ffig=None,
figsize=(6, 4)):
"""Plot mean rates across tasks."""
# Init.
if tasks is None:
tasks = mRates.keys()
# Plot mean activity.
lRates = []
for task in tasks:
lrates = pd.DataFrame(mRates[task].unstack(), columns=['rate'])
lrates['task'] = task
lRates.append(lrates)
lRates = pd.concat(lRates)
lRates['time'] = lRates.index.get_level_values(0)
lRates['unit'] = lRates.index.get_level_values(1)
if task_lbls is not None:
lRates.task.replace(task_lbls, inplace=True)
# Plot as time series.
#putil.set_style('notebook', 'white')
fig = putil.figure(figsize=figsize)
ax = putil.axes()
sns.tsplot(lRates,
time='time',
value='rate',
unit='unit',
condition='task',
ci=ci,
ax=ax)
# Add periods and baseline.
putil.plot_periods(ax=ax)
if baseline is not None:
putil.add_baseline(baseline, ax=ax)
# Format plot.
sns.despine(ax=ax)
putil.set_labels(ax, xlab='time since S1 onset', ylab='rate (sp/s)')
putil.set_limits(ax, xlim, ylim)
putil.hide_legend_title(ax)
# Save plot.
putil.save_fig(ffig, fig)
return fig, ax
def raster(spk_trains,
t_unit=ms,
prds=None,
c='b',
xlim=None,
title=None,
xlab=None,
ylab=None,
ffig=None,
ax=None):
"""Plot rasterplot."""
# Init.
ax = putil.axes(ax)
putil.plot_periods(prds, ax=ax)
putil.set_limits(ax, xlim)
# There's nothing to plot.
if not len(spk_trains):
return ax
# Plot raster.
for i, spk_tr in enumerate(spk_trains):
x = np.array(spk_tr.rescale(t_unit))
y = (i + 1) * np.ones_like(x)
# Spike markers are plotted in absolute size (figure coordinates).
# ax.scatter(x, y, c=c, s=1.8, edgecolor=c, marker='|')
# Spike markers are plotted in relative size (axis coordinates)
patches = [
Rectangle((xi - wsp / 2, yi - hsp / 2), wsp, hsp)
for xi, yi in zip(x, y)
]
collection = PatchCollection(patches, facecolor=c, edgecolor=c)
ax.add_collection(collection)
# Format plot.
ylim = [0.5, len(spk_trains) + 0.5] if len(spk_trains) else [0, 1]
if xlab is not None:
xlab = putil.t_lbl.format(xlab)
putil.format_plot(ax, xlim, ylim, xlab, ylab, title)
putil.hide_axes(ax, show_x=True)
putil.hide_spines(ax)
# Order trials from top to bottom, only after setting axis limits.
ax.invert_yaxis()
# Save and return plot.
putil.save_fig(ffig)
return ax
def plot_RF_results(RF_res, stims, fdir, sup_title):
"""Plot receptive field results."""
# Plot distribution of coverage values.
putil.set_style('poster', 'white')
fig = putil.figure()
ax = putil.axes()
sns.distplot(RF_res.S1_cover, bins=np.arange(0, 1.01, 0.1),
kde=False, rug=True, ax=ax)
putil.set_limits(ax, [0, 1])
fst = util.format_to_fname(sup_title)
ffig = fdir + fst + '_S1_coverage.png'
putil.save_fig(ffig, fig, sup_title)
# Plot RF coverage and rate during S1 on regression plot for each
# recording and task.
tasks = RF_res.index.get_level_values(-1).unique()
for vname, ylim in [('mean_rate', [0, None]), ('max_rate', [0, None]),
('mDSI', [0, 1])]:
fig, gs, axes = putil.get_gs_subplots(nrow=len(stims), ncol=len(tasks),
subw=4, subh=4, ax_kws_list=None,
create_axes=True)
colors = sns.color_palette('muted', len(tasks))
for istim, stim in enumerate(stims):
for itask, task in enumerate(tasks):
# Plot regression plot.
ax = axes[istim, itask]
scov, sval = [stim + '_' + name for name in ('cover', vname)]
df = RF_res.xs(task, level=-1)
sns.regplot(scov, sval, df, color=colors[itask], ax=ax)
# Add unit labels.
uids = df.index.droplevel(0)
putil.add_unit_labels(ax, uids, df[scov], df[sval])
# Add stats.
r, p = sp.stats.pearsonr(df[sval], df[scov])
pstr = util.format_pvalue(p)
txt = 'r = {:.2f}, {}'.format(r, pstr)
ax.text(0.02, 0.98, txt, va='top', ha='left',
transform=ax.transAxes)
# Set labels.
title = '{} {}'.format(task, stim)
xlab, ylab = [sn.replace('_', ' ') for sn in (scov, sval)]
putil.set_labels(ax, xlab, ylab, title)
# Set limits.
xlim = [0, 1]
putil.set_limits(ax, xlim, ylim)
# Save plot.
fst = util.format_to_fname(sup_title)
fname = '{}_cover_{}.png'.format(fst, vname)
ffig = util.join([fdir, vname, fname])
putil.save_fig(ffig, fig, sup_title)
def plot_ROC_mean(d_faroc,
t1=None,
t2=None,
ylim=None,
colors=None,
ylab='AROC',
ffig=None):
"""Plot mean ROC curves over given period."""
# Import results.
d_aroc = {}
for name, faroc in d_faroc.items():
aroc = util.read_objects(faroc, 'aroc')
d_aroc[name] = aroc.unstack().T
# Format results.
laroc = pd.DataFrame(pd.concat(d_aroc), columns=['aroc'])
laroc['task'] = laroc.index.get_level_values(0)
laroc['time'] = laroc.index.get_level_values(1)
laroc['unit'] = laroc.index.get_level_values(2)
laroc.index = np.arange(len(laroc.index))
# Init figure.
fig = putil.figure(figsize=(6, 6))
ax = sns.tsplot(laroc,
time='time',
value='aroc',
unit='unit',
condition='task',
color=colors)
# Highlight stimulus periods.
putil.plot_periods(ax=ax)
# Plot mean results.
[ax.lines[i].set_linewidth(3) for i in range(len(ax.lines))]
# Add chance level line.
putil.add_chance_level(ax=ax, alpha=0.8, color='k')
ax.lines[-1].set_linewidth(1.5)
# Format plot.
xlab = 'Time since S1 onset (ms)'
putil.set_labels(ax, xlab, ylab)
putil.set_limits(ax, [t1, t2], ylim)
putil.set_spines(ax, bottom=True, left=True, top=False, right=False)
putil.set_legend(ax, loc=0)
# Save plot.
putil.save_fig(ffig, fig, ytitle=1.05, w_pad=15)
def plot_scores(ax, Scores, Perm=None, Psdo=None, nvals=None, prds=None,
col='b', perm_col='grey', psdo_col='g', xlim=None,
ylim=ylim_scr, xlab=tlab, ylab=ylab_scr, title='',
ytitle=1.04):
"""Plot decoding accuracy results."""
lgn_patches = []
# Plot permuted results (if exist).
if not util.is_null(Perm) and not Perm.isnull().all().all():
x, pval = Perm.columns, Perm.loc['pval']
ymean, ystd = Perm.loc['mean'], Perm.loc['std']
plot_mean_std_sdiff(x, ymean, ystd, pval, pth=0.01, lw=6,
color=perm_col, ax=ax)
lgn_patches.append(putil.get_artist('permuted', perm_col))
# Plot population shuffled results (if exist).
if not util.is_null(Psdo) and not Psdo.isnull().all().all():
x, pval = Psdo.columns, Psdo.loc['pval']
ymean, ystd = Psdo.loc['mean'], Psdo.loc['std']
plot_mean_std_sdiff(x, ymean, ystd, pval, pth=0.01, lw=3,
color=psdo_col, ax=ax)
lgn_patches.append(putil.get_artist('pseudo-population', psdo_col))
# Plot scores.
plot_score_set(Scores, ax, color=col)
lgn_patches.append(putil.get_artist('synchronous', col))
# Add legend.
lgn_patches = lgn_patches[::-1]
putil.set_legend(ax, handles=lgn_patches)
# Add chance level line.
# This currently plots all nvals combined across stimulus period!
if nvals is not None:
chance_lvl = 1.0 / nvals
putil.add_chance_level(ax=ax, ylevel=chance_lvl)
# Add stimulus periods.
if prds is not None:
putil.plot_periods(prds, ax=ax)
# Set axis limits.
xlim = xlim if xlim is not None else tlim
putil.set_limits(ax, xlim, ylim)
# Format plot.
putil.set_labels(ax, xlab, ylab, title, ytitle)
def sign_scatter(v1,
v2,
pvals=None,
pth=0.01,
scol='g',
nscol='k',
id_line=False,
fit_reg=False,
ax=None):
"""Plot scatter plot with significant points highlighted."""
# Init.
ax = putil.axes(ax)
s_pars = (True, scol, {'alpha': 1.0})
ns_pars = (False, nscol, {'alpha': 0.8})
# Binarize significance stats.
vsig = (pvals < pth if pvals is not None else pd.Series(True,
index=v1.index))
# Plot significant and non-significant points.
for b, c, a in [ns_pars, s_pars]:
if (vsig == b).any():
sns.regplot(v1.loc[vsig == b],
v2.loc[vsig == b],
fit_reg=fit_reg,
color=c,
scatter_kws=a,
ax=ax)
# Format plot.
sns.despine(ax=ax)
# Add identity line.
if id_line:
v_max = max(ax.get_xlim()[1], ax.get_ylim()[1])
putil.set_limits(ax, [0, v_max], [0, v_max])
putil.add_identity_line(ax=ax, equal_xy=True)
return ax
def rate(rate_list,
names=None,
prds=None,
evts=None,
cols=None,
baseline=None,
pval=0.05,
test='mann_whitney_u',
test_kws=None,
xlim=None,
ylim=None,
title=None,
xlab=None,
ylab=putil.FR_lbl,
add_lgn=True,
lgn_lbl='trs',
ffig=None,
ax=None):
"""Plot firing rate."""
# Init.
ax = putil.axes(ax)
if test_kws is None:
test_kws = dict()
# Plot periods and baseline first.
putil.plot_periods(prds, ax=ax)
if baseline is not None:
putil.add_baseline(baseline, ax=ax)
putil.set_limits(ax, xlim)
if not len(rate_list):
return ax
if cols is None:
cols = putil.get_colors(as_cycle=False)
if names is None:
names = len(rate_list) * ['']
# Iterate through list of rate arrays
xmin, xmax, ymax = None, None, None
for i, rts in enumerate(rate_list):
# Init.
name = names[i]
col = cols[i]
# Skip empty array (no trials).
if not rts.shape[0]:
continue
# Set line label. Convert to Numpy array to format floats nicely.
lbl = str(np.array(name)) if util.is_iterable(name) else str(name)
if lgn_lbl is not None:
lbl += ' ({} {})'.format(rts.shape[0], lgn_lbl)
# Plot mean +- SEM of rate vectors.
tvec, meanr, semr = rts.columns, rts.mean(), rts.sem()
ax.plot(tvec, meanr, label=lbl, color=col)
ax.fill_between(tvec,
meanr - semr,
meanr + semr,
alpha=0.2,
facecolor=col,
edgecolor=col)
# Update limits.
tmin, tmax, rmax = tvec.min(), tvec.max(), (meanr + semr).max()
xmin = np.min([xmin, tmin]) if xmin is not None else tmin
xmax = np.max([xmax, tmax]) if xmax is not None else tmax
ymax = np.max([ymax, rmax]) if ymax is not None else rmax
# Set ticks, labels and axis limits.
if xlim is None:
if xmin == xmax: # avoid setting identical limits
xmax = None
xlim = (xmin, xmax)
if ylim is None:
ymax = 1.02 * ymax if (ymax is not None) and (ymax > 0) else None
ylim = (0, ymax)
if xlab is not None:
xlab = putil.t_lbl.format(xlab)
putil.format_plot(ax, xlim, ylim, xlab, ylab, title)
t1, t2 = ax.get_xlim() # in case it was set to None
tmarks, tlbls = putil.get_tick_marks_and_labels(t1, t2)
putil.set_xtick_labels(ax, tmarks, tlbls)
putil.set_max_n_ticks(ax, 7, 'y')
# Add legend.
if add_lgn and len(rate_list):
putil.set_legend(ax,
loc=1,
borderaxespad=0.0,
handletextpad=0.4,
handlelength=0.6)
# Add significance line to top of axes.
if (pval is not None) and (len(rate_list) == 2):
rates1, rates2 = rate_list
sign_prds = stats.sign_periods(rates1, rates2, pval, test, **test_kws)
putil.plot_signif_prds(sign_prds, color='m', linewidth=4.0, ax=ax)
# Plot event markers.
putil.plot_event_markers(evts, ax=ax)
# Save and return plot.
putil.save_fig(ffig)
return ax
def plot_group_violin(res,
x,
y,
groups=None,
npval=None,
pth=0.01,
color='grey',
ylim=None,
ylab=None,
ffig=None):
"""Plot group-wise results on violin plots."""
if groups is None:
groups = res['group'].unique()
# Test difference from zero in each groups.
ttest_res = {
group: sp.stats.ttest_1samp(gres[y], 0)
for group, gres in res.groupby(x)
}
ttest_res = pd.DataFrame.from_dict(ttest_res, 'index')
# Binarize significance test.
res['is_sign'] = res[npval] < pth if npval is not None else True
res['direction'] = np.sign(res[y])
# Set up figure and plot data.
fig = putil.figure()
ax = putil.axes()
putil.add_baseline(ax=ax)
sns.violinplot(x=x, y=y, data=res, inner=None, order=groups, ax=ax)
sns.swarmplot(x=x,
y=y,
hue='is_sign',
data=res,
color=color,
order=groups,
hue_order=[True, False],
ax=ax)
putil.set_labels(ax, xlab='', ylab=ylab)
putil.set_limits(ax, ylim=ylim)
putil.hide_legend(ax)
# Add annotations.
ymin, ymax = ax.get_ylim()
ylvl = ymax
for i, group in enumerate(groups):
gres = res.loc[res.group == group]
# Mean.
mean_str = 'Mean:\n' if i == 0 else '\n'
mean_str += '{:.2f}'.format(gres[y].mean())
# Non-zero test of distribution.
str_pval = util.format_pvalue(ttest_res.loc[group, 'pvalue'])
mean_str += '\n({})'.format(str_pval)
# Stats on difference from baseline.
nnonsign, ntot = (~gres.is_sign).sum(), len(gres)
npos, nneg = [
sum(gres.is_sign & (gres.direction == d)) for d in (1, -1)
]
sign_to_report = [('+', npos), ('=', nnonsign), ('-', nneg)]
nsign_str = ''
for symb, n in sign_to_report:
prc = str(int(round(100 * n / ntot)))
nsign_str += '{} {:>3} / {} ({:>2}%)\n'.format(
symb, int(n), ntot, prc)
lbl = '{}\n\n{}'.format(mean_str, nsign_str)
ax.text(i, ylvl, lbl, fontsize='smaller', va='bottom', ha='center')
# Save plot.
putil.save_fig(ffig, fig)
return fig, ax
def plot_combined_rec_mean(recs, stims, res_dir, par_kws,
list_n_most_DS, list_min_nunits,
n_boot=1e4, ci=95,
tasks=None, task_labels=None, add_title=True,
fig=None):
"""Test and plot results combined across sessions."""
# Init.
# putil.set_style('notebook', 'ticks')
vkey = 'all'
# This should be made more explicit!
prds = [[stim] + list(constants.fixed_tr_prds.loc[stim])
for stim in stims]
# Load all results to plot.
dict_rt_res = decutil.load_res(res_dir, list_n_most_DS, **par_kws)
# Create figures.
fig_scr, _, axs_scr = putil.get_gs_subplots(nrow=len(dict_rt_res),
ncol=len(list_min_nunits),
subw=8, subh=6, fig=fig,
create_axes=True)
# Query data.
allScores = {}
allnunits = {}
for n_most_DS, rt_res in dict_rt_res.items():
# Get accuracy scores.
dScores = {(rec, task): res[vkey]['Scores'].mean()
for (rec, task), res in rt_res.items()
if (vkey in res) and (res[vkey] is not None)}
allScores[n_most_DS] = pd.concat(dScores, axis=1).T
# Get number of units.
allnunits[n_most_DS] = {(rec, task): res[vkey]['nunits'].iloc[0]
for (rec, task), res in rt_res.items()
if (vkey in res) and (res[vkey] is not None)}
# Get # values (for baseline plotting.)
all_nvals = pd.Series({(rec, task): res[vkey]['nclasses'].iloc[0]
for (rec, task), res in rt_res.items()
if (vkey in res) and (res[vkey] is not None)})
un_nvals = all_nvals.unique()
if len(un_nvals) > 1 and verbose:
print('Found multiple # of classes to decode: {}'.format(un_nvals))
nvals = un_nvals[0]
allnunits = pd.DataFrame(allnunits)
# Plot mean performance across recordings and
# test significance by bootstrapping.
for inmost, n_most_DS in enumerate(list_n_most_DS):
Scores = allScores[n_most_DS]
nunits = allnunits[n_most_DS]
for iminu, min_nunits in enumerate(list_min_nunits):
ax_scr = axs_scr[inmost, iminu]
# Select only recordings with minimum number of units.
sel_rt = nunits.index[nunits >= min_nunits]
nScores = Scores.loc[sel_rt].copy()
# Nothing to plot.
if nScores.empty:
ax_scr.axis('off')
continue
# Prepare data.
if tasks is None:
tasks = nScores.index.get_level_values(1).unique() # in data
if task_labels is None:
task_labels = {task: task for task in tasks}
dScores = {task: pd.DataFrame(nScores.xs(task, level=1).unstack(),
columns=['accuracy'])
for task in tasks}
lScores = pd.concat(dScores, axis=0)
lScores['time'] = lScores.index.get_level_values(1)
lScores['task'] = lScores.index.get_level_values(0)
lScores['rec'] = lScores.index.get_level_values(2)
lScores.index = np.arange(len(lScores.index))
lScores.task.replace(task_labels, inplace=True)
# Add altered task names for legend plotting.
nrecs = {task_labels[task]: len(nScores.xs(task, level=1))
for task in tasks}
my_format = lambda x: '{} (n={})'.format(x, nrecs[x])
lScores['task_nrecs'] = lScores['task'].apply(my_format)
# Plot as time series.
sns.tsplot(lScores, time='time', value='accuracy', unit='rec',
condition='task_nrecs', ci=ci, n_boot=n_boot, ax=ax_scr)
# Add chance level line.
chance_lvl = 1.0 / nvals
putil.add_chance_level(ax=ax_scr, ylevel=chance_lvl)
# Add stimulus periods.
putil.plot_periods(prds, ax=ax_scr)
# Set axis limits.
putil.set_limits(ax_scr, tlim)
# Format plot.
title = ('{} most DS units'.format(n_most_DS)
if n_most_DS != 0 else 'all units')
title += (', recordings with at least {} units'.format(min_nunits)
if (min_nunits > 1 and len(list_min_nunits) > 1) else '')
ytitle = 1.0
putil.set_labels(ax_scr, tlab, ylab_scr, title, ytitle)
putil.hide_legend_title(ax_scr)
# Match axes across decoding plots.
[putil.sync_axes(axs_scr[inmost, :], sync_y=True)
for inmost in range(axs_scr.shape[0])]
# Save plots.
list_n_most_DS_str = [str(i) if i != 0 else 'all' for i in list_n_most_DS]
par_kws['n_most_DS'] = ', '.join(list_n_most_DS_str)
title = ''
if add_title:
title = decutil.fig_title(res_dir, **par_kws)
title += '\n{}% CE with {} bootstrapped subsamples'.format(ci,
int(n_boot))
fs_title = 'large'
w_pad, h_pad = 3, 3
par_kws['n_most_DS'] = '_'.join(list_n_most_DS_str)
ffig = decutil.fig_fname(res_dir, 'combined_score', fformat, **par_kws)
putil.save_fig(ffig, fig_scr, title, fs_title, w_pad=w_pad, h_pad=h_pad)
return fig_scr, axs_scr, ffig
def plot_scores_across_nunits(recs, stims, res_dir, list_n_most_DS, par_kws):
"""
Plot prediction score results across different number of units included.
"""
# Init.
putil.set_style('notebook', 'ticks')
tasks = par_kws['tasks']
# Remove Passive if plotting Saccade or Correct.
if par_kws['feat'] in ['saccade', 'correct']:
tasks = tasks[~tasks.str.contains('Pas')]
# Load all results to plot.
dict_rt_res = decutil.load_res(res_dir, list_n_most_DS, **par_kws)
# Create figures.
fig_scr, _, axs_scr = putil.get_gs_subplots(nrow=len(recs),
ncol=len(tasks),
subw=8, subh=6,
create_axes=True)
# Do plotting per recording and task.
for irec, rec in enumerate(recs):
if verbose:
print('\n' + rec)
for itask, task in enumerate(tasks):
if verbose:
print(' ' + task)
ax_scr = axs_scr[irec, itask]
# Init data.
dict_lScores = {}
cols = sns.color_palette('hls', len(dict_rt_res.keys()))
lncls = []
for (n_most_DS, rt_res), col in zip(dict_rt_res.items(), cols):
# Check if results exist for rec-task combination.
if (((rec, task) not in rt_res.keys()) or
(not len(rt_res[(rec, task)].keys()))):
continue
res = rt_res[(rec, task)]
for v, col in zip(res.keys(), cols):
vres = res[v]
Scores = vres['Scores']
lncls.append(vres['nclasses'])
# Unstack dataframe with results.
lScores = pd.DataFrame(Scores.unstack(), columns=['score'])
lScores['time'] = lScores.index.get_level_values(0)
lScores['fold'] = lScores.index.get_level_values(1)
lScores.index = np.arange(len(lScores.index))
# Get number of units tested.
nunits = vres['nunits']
uni_nunits = nunits.unique()
if len(uni_nunits) > 1 and verbose:
print('Different number of units found.')
nunits = uni_nunits[0]
# Collect results.
dict_lScores[(nunits, v)] = lScores
# Skip rest if no data is available.
# Check if any result exists for rec-task combination.
if not len(dict_lScores):
ax_scr.axis('off')
continue
# Concatenate accuracy scores from every recording.
all_lScores = pd.concat(dict_lScores)
all_lScores['n_most_DS'] = all_lScores.index.get_level_values(0)
all_lScores.index = np.arange(len(all_lScores.index))
# Plot decoding results.
nnunits = len(all_lScores['n_most_DS'].unique())
title = '{} {}, {} sets of units'.format(' '.join(rec), task,
nnunits)
ytitle = 1.0
prds = [[stim] + list(constants.fixed_tr_prds.loc[stim])
for stim in stims]
# Plot time series.
palette = sns.color_palette('muted')
sns.tsplot(all_lScores, time='time', value='score', unit='fold',
condition='n_most_DS', color=palette, ax=ax_scr)
# Add chance level line.
# This currently plots a chance level line for every nvals,
# combined across stimulus period!
uni_ncls = np.unique(np.array(lncls).flatten())
if len(uni_ncls) > 1 and verbose:
print('Different number of classes found.')
for nvals in uni_ncls:
chance_lvl = 1.0 / nvals
putil.add_chance_level(ax=ax_scr, ylevel=chance_lvl)
# Add stimulus periods.
if prds is not None:
putil.plot_periods(prds, ax=ax_scr)
# Set axis limits.
putil.set_limits(ax_scr, tlim, ylim_scr)
# Format plot.
putil.set_labels(ax_scr, tlab, ylab_scr, title, ytitle)
# Match axes across decoding plots.
# [putil.sync_axes(axs_scr[:, itask], sync_y=True)
# for itask in range(axs_scr.shape[1])]
# Save plots.
list_n_most_DS_str = [str(i) if i != 0 else 'all' for i in list_n_most_DS]
par_kws['n_most_DS'] = ', '.join(list_n_most_DS_str)
title = decutil.fig_title(res_dir, **par_kws)
fs_title = 'large'
w_pad, h_pad = 3, 3
par_kws['n_most_DS'] = '_'.join(list_n_most_DS_str)
ffig = decutil.fig_fname(res_dir, 'score_nunits', fformat, **par_kws)
putil.save_fig(ffig, fig_scr, title, fs_title, w_pad=w_pad, h_pad=h_pad)
def plot_score_multi_rec(recs, stims, res_dir, par_kws):
"""Plot prediction scores for multiple recordings."""
# Init.
putil.set_style('notebook', 'ticks')
n_most_DS = par_kws['n_most_DS']
tasks = par_kws['tasks']
# Remove Passive if plotting Saccade or Correct.
if par_kws['feat'] in ['saccade', 'correct']:
tasks = tasks[~tasks.str.contains('Pas')]
# Load results.
rt_res = decutil.load_res(res_dir, **par_kws)[n_most_DS]
# Create figure.
ret = putil.get_gs_subplots(nrow=1, ncol=len(tasks),
subw=8, subh=6, create_axes=True)
fig_scr, _, axs_scr = ret
print('\nPlotting multi-recording results...')
for itask, task in enumerate(tasks):
if verbose:
print(' ' + task)
ax_scr = axs_scr[0, itask]
dict_lScores = {}
for irec, rec in enumerate(recs):
# Check if results exist for rec-task combination.
if (((rec, task) not in rt_res.keys()) or
(not len(rt_res[(rec, task)].keys()))):
continue
# Init data.
res = rt_res[(rec, task)]
cols = sns.color_palette('hls', len(res.keys()))
lncls = []
for v, col in zip(res.keys(), cols):
vres = res[v]
if vres is None:
continue
Scores = vres['Scores']
lncls.append(vres['nclasses'])
# Unstack dataframe with results.
lScores = pd.DataFrame(Scores.unstack(), columns=['score'])
lScores['time'] = lScores.index.get_level_values(0)
lScores['fold'] = lScores.index.get_level_values(1)
lScores.index = np.arange(len(lScores.index))
dict_lScores[(rec, v)] = lScores
if not len(dict_lScores):
ax_scr.axis('off')
continue
# Concatenate accuracy scores from every recording.
all_lScores = pd.concat(dict_lScores)
all_lScores['rec'] = all_lScores.index.get_level_values(0)
all_lScores['rec'] = all_lScores['rec'].str.join(' ') # format label
all_lScores.index = np.arange(len(all_lScores.index))
# Plot decoding results.
nrec = len(all_lScores['rec'].unique())
title = '{}, {} recordings'.format(task, nrec)
ytitle = 1.0
prds = [[stim] + list(constants.fixed_tr_prds.loc[stim])
for stim in stims]
# Plot time series.
palette = sns.color_palette('muted')
sns.tsplot(all_lScores, time='time', value='score', unit='fold',
condition='rec', color=palette, ax=ax_scr)
# Add chance level line.
# This currently plots a chance level line for every nvals,
# combined across stimulus period!
uni_ncls = np.unique(np.array(lncls).flatten())
if len(uni_ncls) > 1 and verbose:
print('Different number of classes found.')
for nvals in uni_ncls:
chance_lvl = 1.0 / nvals
putil.add_chance_level(ax=ax_scr, ylevel=chance_lvl)
# Add stimulus periods.
if prds is not None:
putil.plot_periods(prds, ax=ax_scr)
# Set axis limits.
putil.set_limits(ax_scr, tlim, ylim_scr)
# Format plot.
putil.set_labels(ax_scr, tlab, ylab_scr, title, ytitle)
# Save figure.
title = decutil.fig_title(res_dir, **par_kws)
fs_title = 'large'
w_pad, h_pad = 3, 3
ffig = decutil.fig_fname(res_dir, 'all_scores', fformat, **par_kws)
putil.save_fig(ffig, fig_scr, title, fs_title, w_pad=w_pad, h_pad=h_pad)
def rec_stability_test(UA, fname=None, periods=None):
"""Check stability of recording session across tasks."""
# Init.
if periods is None:
periods = ['whole trial', 'fixation']
# Init figure.
fig, gsp, axs = putil.get_gs_subplots(nrow=len(periods), ncol=1,
subw=10, subh=2.5, create_axes=True,
as_array=False)
for prd, ax in zip(periods, axs):
# Calculate and plot firing rate during given period in each trial
# across session for all units.
colors = putil.get_colors()
task_stats = pd.DataFrame(columns=['t_start', 't_stops', 'label'])
for task, color in zip(UA.tasks(), colors):
# Get activity of all units in task.
tr_rates = []
for u in UA.iter_thru([task]):
rates = u.get_prd_rates(prd, tr_time_idx=True)
tr_rates.append(util.remove_dim_from_series(rates))
tr_rates = pd.DataFrame(tr_rates)
# Not (non-empty and included) unit during task.
if not len(tr_rates.index):
continue
# Plot each rate in task.
tr_times = tr_rates.columns
pplot.lines(tr_times, tr_rates.T, zorder=1, alpha=0.5,
color=color, ax=ax)
# Plot mean +- sem rate.
tr_time = tr_rates.columns
mean_rate, sem_rate = tr_rates.mean(), tr_rates.std()
lower, upper = mean_rate-sem_rate, mean_rate+sem_rate
lower[lower < 0] = 0 # remove negative values
ax.fill_between(tr_time, lower, upper, zorder=2, alpha=.5,
facecolor='grey', edgecolor='grey')
pplot.lines(tr_time, mean_rate, lw=2, color='k', ax=ax)
# Add task stats.
task_lbl = '{}, {} units'.format(task, len(tr_rates.index))
# Add grand mean FR.
task_lbl += '\nFR: {:.1f} sp/s'.format(tr_rates.mean().mean())
# Calculate linear trend to test gradual drift.
slope, _, _, p_value, _ = sp.stats.linregress(tr_times, mean_rate)
slope = 3600*slope # convert to change in spike per hour
pval = util.format_pvalue(p_value, max_digit=3)
task_lbl += '\n$\delta$FR: {:.1f} sp/s/h'.format(slope)
task_lbl += '\n{}'.format(pval)
task_stats.loc[task] = (tr_times.min(), tr_times.max(), task_lbl)
# Set axes limits.
tmin, tmax = task_stats.t_start.min(), task_stats.t_stops.max()
putil.set_limits(ax, xlim=(tmin, tmax))
# Add task labels after all tasks have been plotted.
putil.plot_events(task_stats[['t_start', 'label']], y_lbl=0.75,
lbl_ha='left', lbl_rotation=0, ax=ax)
# Format plot.
xlab = 'Recording time (s)' if prd == periods[-1] else None
putil.set_labels(ax, xlab=xlab, ylab=prd)
putil.set_spines(ax, left=False)
# Save figure.
title = 'Recording stability of ' + UA.Name
putil.save_fig(fname, fig, title)
