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_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 plot_auc_over_time(auc,
tvec,
prds=None,
evts=None,
xlim=None,
ylim=None,
xlab='time',
ylab='AUC',
title=None,
ax=None):
"""Plot AROC values over time."""
# Init params.
ax = putil.axes(ax)
if xlim is None:
xlim = [min(tvec), max(tvec)]
# Plot periods first.
putil.plot_periods(prds, ax=ax)
# Plot AUC over time.
pplot.lines(tvec, auc, ylim, xlim, xlab, ylab, title, color='green', ax=ax)
# Add chance level line.
putil.add_chance_level(ax=ax)
# # Set minimum y axis scale.
# ymin, ymax = ax.get_ylim()
# ymin, ymax = min(ymin, 0.3), max(ymax, 0.7)
# ax.set_ylim([ymin, ymax])
# Set y tick labels.
if ylim is not None and ylim[0] == 0 and ylim[1] == 1:
tck_marks = np.linspace(0, 1, 5)
tck_lbls = np.array(tck_marks, dtype=str)
tck_lbls[1::2] = ''
putil.set_ytick_labels(ax, tck_marks, tck_lbls)
putil.set_max_n_ticks(ax, 5, 'y')
# Plot event markers.
putil.plot_event_markers(evts, ax=ax)
return 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 plot_scores_weights(recs, stims, res_dir, par_kws):
"""
Plot prediction scores and model weights for given recording and analysis.
"""
# 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 figures.
# For prediction scores.
fig_scr, _, axs_scr = putil.get_gs_subplots(nrow=len(recs),
ncol=len(tasks),
subw=8, subh=6,
create_axes=True)
# For unit weights (coefficients).
fig_wgt, _, axs_wgt = putil.get_gs_subplots(nrow=len(recs),
ncol=len(tasks),
subw=8, subh=6,
create_axes=True)
for irec, rec in enumerate(recs):
if verbose:
print('\n' + rec)
for itask, task in enumerate(tasks):
if verbose:
print(' ' + task)
# Init figures.
ax_scr = axs_scr[irec, itask]
ax_wgt = axs_wgt[irec, itask]
# Check if any result exists for rec-task combination.
if (((rec, task) not in rt_res.keys()) or
(not len(rt_res[(rec, task)].keys()))):
ax_scr.axis('off')
ax_wgt.axis('off')
continue
# Init data.
res = rt_res[(rec, task)]
vals = [v for v in res.keys() if not util.is_null(res[v])]
cols = sns.color_palette('hls', len(vals))
lnunits, lntrs, lncls, = [], [], []
for v, col in zip(vals, cols):
# Basic results.
vres = res[v]
Scores = vres['Scores']
Coefs = vres['Coefs']
Perm = vres['Perm']
Psdo = vres['Psdo']
# Decoding params.
lnunits.append(vres['nunits'])
lntrs.append(vres['ntrials'])
lncls.append(vres['nclasses'])
# Plot decoding accuracy.
plot_scores(ax_scr, Scores, Perm, Psdo, col=col)
# Add labels.
uni_lnunits = np.unique(np.array(lnunits).flatten())
if len(uni_lnunits) > 1 and verbose:
print('Different number of units found.')
nunits = uni_lnunits[0]
title = '{} {}, {} units'.format(' '.join(rec), task, nunits)
putil.set_labels(ax_scr, tlab, ylab_scr, title, ytitle=1.04)
# Add chance level line.
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)
# Plot stimulus periods.
prds = [[stim] + list(constants.fixed_tr_prds.loc[stim])
for stim in stims]
putil.plot_periods(prds, ax=ax_scr)
# Plot unit weights over time.
plot_weights(ax_wgt, Coefs, prds, tlim, tlab, title=title)
# Match axes across decoding plots.
# [putil.sync_axes(axs_scr[:, itask], sync_y=True)
# for itask in range(axs_scr.shape[1])]
# Save plots.
title = decutil.fig_title(res_dir, **par_kws)
fs_title = 'large'
w_pad, h_pad = 3, 3
# Performance.
ffig = decutil.fig_fname(res_dir, 'score', 'pdf', **par_kws)
putil.save_fig(ffig, fig_scr, title, fs_title, w_pad=w_pad, h_pad=h_pad)
# Weights.
ffig = decutil.fig_fname(res_dir, 'weight', 'pdf', **par_kws)
putil.save_fig(ffig, fig_wgt, title, fs_title, w_pad=w_pad, h_pad=h_pad)