def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 500
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
# Conditions
spec = model.spec
mods = spec.mods
freqs = spec.freqs
n_conditions = spec.n_conditions
n_trials = n_conditions * trials_per_condition
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(mods), len(freqs)))
context = {'mod': mods[k.pop(0)], 'freq': freqs[k.pop(0)]}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'psychometric':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
psychometric(trialsfile, plot)
plot.vline(config['model'].spec.boundary)
fig.save(path=config['figspath'], name='psychometric')
fig.close()
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, (config['figspath'], 'sorted'), network=network)
def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 500
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
# Conditions
spec = model.spec
mods = spec.mods
freqs = spec.freqs
n_conditions = spec.n_conditions
n_trials = n_conditions * trials_per_condition
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(mods), len(freqs)))
context = {'mod': mods[k.pop(0)], 'freq': freqs[k.pop(0)]}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'psychometric':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
psychometric(trialsfile, plot)
plot.vline(config['model'].spec.boundary)
fig.save(path=config['figspath'], name='psychometric')
fig.close()
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, (config['figspath'], 'sorted'), network=network)
def sort(trialsfile, plots, units=None, network='p', **kwargs):
"""
Sort trials.
"""
# Load trials
data = utils.load(trialsfile)
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
# Which network?
if network == 'p':
r = r_p
else:
r = r_v
# Number of units
N = r.shape[-1]
# Same for every trial
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Aligned to stimulus onset
#=====================================================================================
r_by_cond_stimulus = {}
n_r_by_cond_stimulus = {}
for n, trial in enumerate(trials):
if not perf.decisions[n]:
continue
if trial['mod'] == 'va':
continue
assert trial['mod'] == 'v' or trial['mod'] == 'a'
if not perf.corrects[n]:
continue
# Condition
mod = trial['mod']
choice = perf.choices[n]
cond = (mod, choice)
# Storage
r_by_cond_stimulus.setdefault(cond, np.zeros((Ntime_a, N)))
n_r_by_cond_stimulus.setdefault(cond, np.zeros((Ntime_a, N)))
# Firing rates
Mn = np.tile(M[:, n], (N, 1)).T
Rn = r[:, n] * Mn
# Align point
t0 = trial['epochs']['stimulus'][0] - 1
# Before
n_b = Rn[:t0].shape[0]
r_by_cond_stimulus[cond][Ntime - 1 - n_b:Ntime - 1] += Rn[:t0]
n_r_by_cond_stimulus[cond][Ntime - 1 - n_b:Ntime - 1] += Mn[:t0]
# After
n_a = Rn[t0:].shape[0]
r_by_cond_stimulus[cond][Ntime - 1:Ntime - 1 + n_a] += Rn[t0:]
n_r_by_cond_stimulus[cond][Ntime - 1:Ntime - 1 + n_a] += Mn[t0:]
for cond in r_by_cond_stimulus:
r_by_cond_stimulus[cond] = utils.div(r_by_cond_stimulus[cond],
n_r_by_cond_stimulus[cond])
#-------------------------------------------------------------------------------------
# Plot
#-------------------------------------------------------------------------------------
lw = kwargs.get('lw', 1.5)
dashes = kwargs.get('dashes', [3, 2])
vline_props = {'lw': kwargs.get('lw_vline', 0.5)}
if 'dashes_vline' in kwargs:
vline_props['linestyle'] = '--'
vline_props['dashes'] = dashes
colors_by_mod = {'v': Figure.colors('blue'), 'a': Figure.colors('green')}
linestyle_by_choice = {'L': '-', 'H': '--'}
lineprops = dict(lw=lw)
def plot_sorted(plot, unit, w, r_sorted):
t = time_a[w]
yall = [[1]]
for cond in [('v', 'H'), ('v', 'L'), ('a', 'H'), ('a', 'L')]:
mod, choice = cond
if mod == 'v':
label = 'Vis, '
elif mod == 'a':
label = 'Aud, '
else:
raise ValueError(mod)
if choice == 'H':
label += 'high'
elif choice == 'L':
label += 'low'
else:
raise ValueError(choice)
linestyle = linestyle_by_choice[choice]
if linestyle == '-':
lineprops = dict(linestyle=linestyle, lw=lw)
else:
lineprops = dict(linestyle=linestyle, lw=lw, dashes=dashes)
plot.plot(t,
r_sorted[cond][w, unit],
color=colors_by_mod[mod],
label=label,
**lineprops)
yall.append(r_sorted[cond][w, unit])
return t, yall
def on_stimulus(plot, unit):
w, = np.where((time_a >= -300) & (time_a <= 1000))
t, yall = plot_sorted(plot, unit, w, r_by_cond_stimulus)
plot.xlim(t[0], t[-1])
return yall
if units is not None:
for plot, unit in zip(plots, units):
on_stimulus(plot, unit)
else:
figspath, name = plots
for unit in xrange(N):
fig = Figure()
plot = fig.add()
#-----------------------------------------------------------------------------
yall = []
yall += on_stimulus(plot, unit)
plot.lim('y', yall, lower=0)
plot.vline(0)
plot.xlabel('Time (ms)')
plot.ylabel('Firing rate (a.u.)')
#-----------------------------------------------------------------------------
fig.save(path=figspath,
name=name + '_{}{:03d}'.format(network, unit))
fig.close()
def do(action, args, config):
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
wagers = spec.wagers
left_rights = spec.left_rights
cohs = spec.cohs
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(wagers), len(left_rights), len(cohs)))
context = {
'wager': wagers[k.pop(0)],
'left_right': left_rights[k.pop(0)],
'coh': cohs[k.pop(0)]
}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'sure_stimulus_duration':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
sure_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Probability sure target')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'correct_stimulus_duration':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
correct_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Probability correct')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'value_stimulus_duration':
trialsfile = runtools.activityfile(config['trialspath'])
fig = Figure()
plot = fig.add()
value_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Expected reward')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, os.path.join(config['figspath'], 'sorted'), network=network)
def sort(trialsfile, plots, unit=None, network='p', **kwargs):
# Load trials
data = utils.load(trialsfile)
if len(data) == 9:
trials, U, Z, A, P, M, perf, r_p, r_v = data
else:
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
if network == 'p':
print("Sorting policy network activity.")
r = r_p
else:
print("Sorting value network activity.")
r = r_v
# Number of units
N = r.shape[-1]
# Time
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Preferred targets
#=====================================================================================
preferred_targets = get_preferred_targets(trials, perf, r)
#=====================================================================================
# No-wager trials
#=====================================================================================
def get_no_wager(func_t0):
trials_by_cond = {}
for n, trial in enumerate(trials):
if trial['wager']:
continue
if trial['coh'] == 0:
continue
if perf.choices[n] is None:
continue
cond = trial['left_right']
m_n = np.tile(M[:,n], (N, 1)).T
r_n = r[:,n]*m_n
t0 = func_t0(trial['epochs'], perf.t_choices[n])
# Storage
trials_by_cond.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
# Average
for cond in trials_by_cond:
trials_by_cond[cond] = utils.div(trials_by_cond[cond]['r'],
trials_by_cond[cond]['n'])
return trials_by_cond
noTs_stimulus = get_no_wager(lambda epochs, t_choice: epochs['stimulus'][0] - 1)
noTs_choice = get_no_wager(lambda epochs, t_choice: t_choice)
#=====================================================================================
# Wager trials, aligned to stimulus onset
#=====================================================================================
def get_wager(func_t0):
trials_by_cond = {}
trials_by_cond_sure = {}
for n, trial in enumerate(trials):
if not trial['wager']:
continue
if perf.choices[n] is None:
continue
if trial['coh'] == 0:
continue
cond = trial['left_right']
m_n = np.tile(M[:,n], (N, 1)).T
r_n = r[:,n]*m_n
t0 = func_t0(trial['epochs'], perf.t_choices[n])
if perf.choices[n] == 'S':
# Storage
trials_by_cond_sure.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond_sure[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond_sure[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond_sure[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond_sure[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
else:
# Storage
trials_by_cond.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
# Average
for cond in trials_by_cond:
trials_by_cond[cond] = utils.div(trials_by_cond[cond]['r'],
trials_by_cond[cond]['n'])
# Average
for cond in trials_by_cond_sure:
trials_by_cond_sure[cond] = utils.div(trials_by_cond_sure[cond]['r'],
trials_by_cond_sure[cond]['n'])
return trials_by_cond, trials_by_cond_sure
Ts_stimulus, Ts_stimulus_sure = get_wager(lambda epochs, t_choice: epochs['stimulus'][0] - 1)
Ts_sure, Ts_sure_sure = get_wager(lambda epochs, t_choice: epochs['sure'][0] - 1)
Ts_choice, Ts_choice_sure = get_wager(lambda epochs, t_choice: t_choice)
#=====================================================================================
# Plot
#=====================================================================================
lw = kwargs.get('lw', 1.25)
dashes = kwargs.get('dashes', [3, 1.5])
in_opp_colors = {-1: '0.6', +1: 'k'}
def plot_noTs(noTs, plot, unit, tmin, tmax):
w, = np.where((tmin <= time_a) & (time_a <= tmax))
t = time_a[w]
yall = [[1]]
for lr in noTs:
color = in_opp_colors[lr*preferred_targets[unit]]
y = noTs[lr][w,unit]
plot.plot(t, y, color=color, lw=lw)
yall.append(y)
plot.xlim(tmin, tmax)
plot.xticks([0, tmax])
plot.lim('y', yall, lower=0)
return yall
def plot_Ts(Ts, Ts_sure, plot, unit, tmin, tmax):
w, = np.where((tmin <= time_a) & (time_a <= tmax))
t = time_a[w]
yall = [[1]]
for lr in Ts:
color = in_opp_colors[lr*preferred_targets[unit]]
y = Ts[lr][w,unit]
plot.plot(t, y, color=color, lw=lw)
yall.append(y)
for lr in Ts_sure:
color = in_opp_colors[lr*preferred_targets[unit]]
y = Ts_sure[lr][w,unit]
plot.plot(t, y, color=color, lw=lw, linestyle='--', dashes=dashes)
yall.append(y)
plot.xlim(tmin, tmax)
plot.xticks([0, tmax])
plot.lim('y', yall, lower=0)
return yall
if unit is not None:
y = []
tmin = kwargs.get('noTs-stimulus-tmin', -100)
tmax = kwargs.get('noTs-stimulus-tmax', 700)
y += plot_noTs(noTs_stimulus, plots['noTs-stimulus'], unit, tmin, tmax)
tmin = kwargs.get('noTs-choice-tmin', -500)
tmax = kwargs.get('noTs-choice-tmax', 0)
y += plot_noTs(noTs_choice, plots['noTs-choice'], unit, tmin, tmax)
tmin = kwargs.get('Ts-stimulus-tmin', -100)
tmax = kwargs.get('Ts-stimulus-tmax', 700)
y += plot_Ts(Ts_stimulus, Ts_stimulus_sure, plots['Ts-stimulus'], unit, tmin, tmax)
tmin = kwargs.get('Ts-sure-tmin', -200)
tmax = kwargs.get('Ts-sure-tmax', 700)
y += plot_Ts(Ts_sure, Ts_sure_sure, plots['Ts-sure'], unit, tmin, tmax)
tmin = kwargs.get('Ts-choice-tmin', -500)
tmax = kwargs.get('Ts-choice-tmax', 0)
y += plot_Ts(Ts_choice, Ts_choice_sure, plots['Ts-choice'], unit, tmin, tmax)
return y
else:
name = plots
for unit in xrange(N):
w = utils.mm_to_inch(174)
r = 0.35
fig = Figure(w=w, r=r)
x0 = 0.09
y0 = 0.15
w = 0.13
h = 0.75
dx = 0.05
DX = 0.08
fig.add('noTs-stimulus', [x0, y0, w, h])
fig.add('noTs-choice', [fig[-1].right+dx, y0, w, h])
fig.add('Ts-stimulus', [fig[-1].right+DX, y0, w, h])
fig.add('Ts-sure', [fig[-1].right+dx, y0, w, h])
fig.add('Ts-choice', [fig[-1].right+dx, y0, w, h])
#-----------------------------------------------------------------------------
y = []
plot = fig['noTs-stimulus']
y += plot_noTs(noTs_stimulus, plot, unit, -100, 700)
plot.vline(0)
plot = fig['noTs-choice']
y += plot_noTs(noTs_choice, plot, unit, -500, 200)
plot.vline(0)
plot = fig['Ts-stimulus']
y += plot_Ts(Ts_stimulus, Ts_stimulus_sure, plot, unit, -100, 700)
plot.vline(0)
plot = fig['Ts-sure']
y += plot_Ts(Ts_sure, Ts_sure_sure, plot, unit, -200, 700)
plot.vline(0)
plot = fig['Ts-choice']
y += plot_Ts(Ts_choice, Ts_choice_sure, plot, unit, -500, 200)
plot.vline(0)
for plot in fig.plots.values():
plot.lim('y', y, lower=0)
#-----------------------------------------------------------------------------
fig.save(name+'_{}{:03d}'.format(network, unit))
fig.close()
def sort(trialsfile, plots, units=None, network='p', **kwargs):
"""
Sort trials.
"""
# Load trials
data = utils.load(trialsfile)
if len(data) == 9:
trials, U, Z, A, P, M, perf, r_p, r_v = data
else:
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
# Which network?
if network == 'p':
r = r_p
else:
r = r_v
# Data shape
Ntime = r.shape[0]
N = r.shape[-1]
# Same for every trial
time = trials[0]['time']
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Sort trials
#=====================================================================================
# Sort
trials_by_cond = {}
for n, trial in enumerate(trials):
if perf.choices[n] is None or not perf.corrects[n]:
continue
# Condition
gt_lt = trial['gt_lt']
fpair = trial['fpair']
if gt_lt == '>':
f1, f2 = fpair
else:
f2, f1 = fpair
cond = (f1, f2)
# Firing rates
Mn = np.tile(M[:, n], (N, 1)).T
Rn = r[:, n] * Mn
# Align point
t0 = trial['epochs']['f1'][0] - 1
# Storage
trials_by_cond.setdefault(cond, {
'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))
})
# Before
n_b = Rn[:t0].shape[0]
trials_by_cond[cond]['r'][Ntime - 1 - n_b:Ntime - 1] += Rn[:t0]
trials_by_cond[cond]['n'][Ntime - 1 - n_b:Ntime - 1] += Mn[:t0]
# After
n_a = Rn[t0:].shape[0]
trials_by_cond[cond]['r'][Ntime - 1:Ntime - 1 + n_a] += Rn[t0:]
trials_by_cond[cond]['n'][Ntime - 1:Ntime - 1 + n_a] += Mn[t0:]
# Average
for cond in trials_by_cond:
trials_by_cond[cond] = utils.div(trials_by_cond[cond]['r'],
trials_by_cond[cond]['n'])
#=====================================================================================
# Plot
#=====================================================================================
lw = kwargs.get('lw', 1.5)
w, = np.where((time_a >= -500) & (time_a <= 4000))
def plot_sorted(plot, unit):
t = 1e-3 * time_a[w]
yall = [[1]]
for (f1, f2), r in trials_by_cond.items():
plot.plot(t, r[w, unit], color=smap.to_rgba(f1), lw=lw)
yall.append(r[w, unit])
return t, yall
if units is not None:
for plot, unit in zip(plots, units):
plot_sorted(plot, unit)
else:
figspath, name = plots
for unit in xrange(N):
fig = Figure()
plot = fig.add()
#-----------------------------------------------------------------------------
t, yall = plot_sorted(plot, unit)
plot.xlim(t[0], t[-1])
plot.lim('y', yall, lower=0)
plot.highlight(0, 0.5)
plot.highlight(3.5, 4)
#-----------------------------------------------------------------------------
fig.save(path=figspath,
name=name + '_{}{:03d}'.format(network, unit))
fig.close()
def sort(trialsfile, plots, unit=None, network='p', **kwargs):
# Load trials
data = utils.load(trialsfile)
if len(data) == 9:
trials, U, Z, A, P, M, perf, r_p, r_v = data
else:
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
if network == 'p':
print("Sorting policy network activity.")
r = r_p
else:
print("Sorting value network activity.")
r = r_v
# Number of units
N = r.shape[-1]
# Time
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Preferred targets
#=====================================================================================
preferred_targets = get_preferred_targets(trials, perf, r)
#=====================================================================================
# No-wager trials
#=====================================================================================
def get_no_wager(func_t0):
trials_by_cond = {}
for n, trial in enumerate(trials):
if trial['wager']:
continue
if trial['coh'] == 0:
continue
if perf.choices[n] is None:
continue
cond = trial['left_right']
m_n = np.tile(M[:,n], (N, 1)).T
r_n = r[:,n]*m_n
t0 = func_t0(trial['epochs'], perf.t_choices[n])
# Storage
trials_by_cond.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
# Average
for cond in trials_by_cond:
trials_by_cond[cond] = utils.div(trials_by_cond[cond]['r'],
trials_by_cond[cond]['n'])
return trials_by_cond
noTs_stimulus = get_no_wager(lambda epochs, t_choice: epochs['stimulus'][0] - 1)
noTs_choice = get_no_wager(lambda epochs, t_choice: t_choice)
#=====================================================================================
# Wager trials, aligned to stimulus onset
#=====================================================================================
def get_wager(func_t0):
trials_by_cond = {}
trials_by_cond_sure = {}
for n, trial in enumerate(trials):
if not trial['wager']:
continue
if perf.choices[n] is None:
continue
if trial['coh'] == 0:
continue
cond = trial['left_right']
m_n = np.tile(M[:,n], (N, 1)).T
r_n = r[:,n]*m_n
t0 = func_t0(trial['epochs'], perf.t_choices[n])
if perf.choices[n] == 'S':
# Storage
trials_by_cond_sure.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond_sure[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond_sure[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond_sure[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond_sure[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
else:
# Storage
trials_by_cond.setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Before
n_b = r_n[:t0].shape[0]
trials_by_cond[cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
trials_by_cond[cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
trials_by_cond[cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
trials_by_cond[cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
# Average
for cond in trials_by_cond:
trials_by_cond[cond] = utils.div(trials_by_cond[cond]['r'],
trials_by_cond[cond]['n'])
# Average
for cond in trials_by_cond_sure:
trials_by_cond_sure[cond] = utils.div(trials_by_cond_sure[cond]['r'],
trials_by_cond_sure[cond]['n'])
return trials_by_cond, trials_by_cond_sure
Ts_stimulus, Ts_stimulus_sure = get_wager(lambda epochs, t_choice: epochs['stimulus'][0] - 1)
Ts_sure, Ts_sure_sure = get_wager(lambda epochs, t_choice: epochs['sure'][0] - 1)
Ts_choice, Ts_choice_sure = get_wager(lambda epochs, t_choice: t_choice)
#=====================================================================================
# Plot
#=====================================================================================
lw = kwargs.get('lw', 1.25)
dashes = kwargs.get('dashes', [3, 1.5])
in_opp_colors = {-1: '0.6', +1: 'k'}
def plot_noTs(noTs, plot, unit, tmin, tmax):
w, = np.where((tmin <= time_a) & (time_a <= tmax))
t = time_a[w]
yall = [[1]]
for lr in noTs:
color = in_opp_colors[lr*preferred_targets[unit]]
y = noTs[lr][w,unit]
plot.plot(t, y, color=color, lw=lw)
yall.append(y)
plot.xlim(tmin, tmax)
plot.xticks([0, tmax])
plot.lim('y', yall, lower=0)
return yall
def plot_Ts(Ts, Ts_sure, plot, unit, tmin, tmax):
w, = np.where((tmin <= time_a) & (time_a <= tmax))
t = time_a[w]
yall = [[1]]
for lr in Ts:
color = in_opp_colors[lr*preferred_targets[unit]]
y = Ts[lr][w,unit]
plot.plot(t, y, color=color, lw=lw)
yall.append(y)
for lr in Ts_sure:
color = in_opp_colors[lr*preferred_targets[unit]]
y = Ts_sure[lr][w,unit]
plot.plot(t, y, color=color, lw=lw, linestyle='--', dashes=dashes)
yall.append(y)
plot.xlim(tmin, tmax)
plot.xticks([0, tmax])
plot.lim('y', yall, lower=0)
return yall
if unit is not None:
y = []
tmin = kwargs.get('noTs-stimulus-tmin', -100)
tmax = kwargs.get('noTs-stimulus-tmax', 700)
y += plot_noTs(noTs_stimulus, plots['noTs-stimulus'], unit, tmin, tmax)
tmin = kwargs.get('noTs-choice-tmin', -500)
tmax = kwargs.get('noTs-choice-tmax', 0)
y += plot_noTs(noTs_choice, plots['noTs-choice'], unit, tmin, tmax)
tmin = kwargs.get('Ts-stimulus-tmin', -100)
tmax = kwargs.get('Ts-stimulus-tmax', 700)
y += plot_Ts(Ts_stimulus, Ts_stimulus_sure, plots['Ts-stimulus'], unit, tmin, tmax)
tmin = kwargs.get('Ts-sure-tmin', -200)
tmax = kwargs.get('Ts-sure-tmax', 700)
y += plot_Ts(Ts_sure, Ts_sure_sure, plots['Ts-sure'], unit, tmin, tmax)
tmin = kwargs.get('Ts-choice-tmin', -500)
tmax = kwargs.get('Ts-choice-tmax', 0)
y += plot_Ts(Ts_choice, Ts_choice_sure, plots['Ts-choice'], unit, tmin, tmax)
return y
else:
name = plots
for unit in xrange(N):
w = utils.mm_to_inch(174)
r = 0.35
fig = Figure(w=w, r=r)
x0 = 0.09
y0 = 0.15
w = 0.13
h = 0.75
dx = 0.05
DX = 0.08
fig.add('noTs-stimulus', [x0, y0, w, h])
fig.add('noTs-choice', [fig[-1].right+dx, y0, w, h])
fig.add('Ts-stimulus', [fig[-1].right+DX, y0, w, h])
fig.add('Ts-sure', [fig[-1].right+dx, y0, w, h])
fig.add('Ts-choice', [fig[-1].right+dx, y0, w, h])
#-----------------------------------------------------------------------------
y = []
plot = fig['noTs-stimulus']
y += plot_noTs(noTs_stimulus, plot, unit, -100, 700)
plot.vline(0)
plot = fig['noTs-choice']
y += plot_noTs(noTs_choice, plot, unit, -500, 200)
plot.vline(0)
plot = fig['Ts-stimulus']
y += plot_Ts(Ts_stimulus, Ts_stimulus_sure, plot, unit, -100, 700)
plot.vline(0)
plot = fig['Ts-sure']
y += plot_Ts(Ts_sure, Ts_sure_sure, plot, unit, -200, 700)
plot.vline(0)
plot = fig['Ts-choice']
y += plot_Ts(Ts_choice, Ts_choice_sure, plot, unit, -500, 200)
plot.vline(0)
for plot in fig.plots.values():
plot.lim('y', y, lower=0)
#-----------------------------------------------------------------------------
fig.save(name+'_{}{:03d}'.format(network, unit))
fig.close()
kwargs = {'on-stimulus-tmin': -200, 'on-stimulus-tmax': 600,
'on-choice-tmin': -400, 'on-choice-tmax': 0,
'colors': 'kiani', 'dashes': [3.5, 2]}
rdm_analysis.sort_return(rdm_fixed_activity, fig.plots, **kwargs)
plot = fig['on-stimulus']
plot.xlim(-200, 600)
plot.xticks([-200, 0, 200, 400, 600])
plot.ylim(0.5, 1.2)
#plot.yticks([0.5, 1])
plot.xlabel('Time from stimulus (ms)')
plot.ylabel('Expected reward')
# Legend
props = {'prop': {'size': 8}, 'handlelength': 1.2,
'handletextpad': 1.1, 'labelspacing': 0.7}
plot.legend(bbox_to_anchor=(0.33, 1), **props)
plot = fig['on-choice']
plot.xlim(-400, 0)
plot.xticks([-400, -200, 0])
plot.ylim(0.5, 1.2)
#plot.yticks([0.5, 1])
plot.xlabel('Time from decision (ms)')
#=========================================================================================
fig.save()
def sort_epoch(behaviorfile,
activityfile,
epoch,
offers,
plots,
units=None,
network='p',
separate_by_choice=False,
**kwargs):
"""
Sort trials.
"""
# Load trials
data = utils.load(activityfile)
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
if network == 'p':
print("POLICY NETWORK")
r = r_p
else:
print("VALUE NETWORK")
r = r_v
# Number of units
N = r.shape[-1]
# Same for every trial
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Sort trials
#=====================================================================================
# Epochs
events = ['offer', 'choice']
# Sort
events_by_cond = {e: {} for e in events}
n_by_cond = {}
n_nondecision = 0
for n, trial in enumerate(trials):
if perf.choices[n] is None:
n_nondecision += 1
continue
# Condition
offer = trial['offer']
choice = perf.choices[n]
if separate_by_choice:
cond = (offer, choice)
else:
cond = offer
n_by_cond.setdefault(cond, 0)
n_by_cond[cond] += 1
# Storage
for e in events_by_cond:
events_by_cond[e].setdefault(cond, {
'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))
})
# Firing rates
m_n = np.tile(M[:, n], (N, 1)).T
r_n = r[:, n] * m_n
for e in events_by_cond:
# Align point
if e == 'offer':
t0 = trial['epochs']['offer-on'][0]
elif e == 'choice':
t0 = perf.t_choices[n]
else:
raise ValueError(e)
# Before
n_b = r_n[:t0].shape[0]
events_by_cond[e][cond]['r'][Ntime - 1 - n_b:Ntime - 1] += r_n[:t0]
events_by_cond[e][cond]['n'][Ntime - 1 - n_b:Ntime - 1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
events_by_cond[e][cond]['r'][Ntime - 1:Ntime - 1 + n_a] += r_n[t0:]
events_by_cond[e][cond]['n'][Ntime - 1:Ntime - 1 + n_a] += m_n[t0:]
print("Non-decision trials: {}/{}".format(n_nondecision, len(trials)))
# Average trials
for e in events_by_cond:
for cond in events_by_cond[e]:
events_by_cond[e][cond] = utils.div(events_by_cond[e][cond]['r'],
events_by_cond[e][cond]['n'])
# Epochs
epochs = ['preoffer', 'postoffer', 'latedelay', 'prechoice']
# Average epochs
epochs_by_cond = {e: {} for e in epochs}
for e in epochs_by_cond:
if e == 'preoffer':
ev = 'offer'
w, = np.where((-500 <= time_a) & (time_a < 0))
elif e == 'postoffer':
ev = 'offer'
w, = np.where((0 <= time_a) & (time_a < 500))
elif e == 'latedelay':
ev = 'offer'
w, = np.where((500 <= time_a) & (time_a < 1000))
elif e == 'prechoice':
ev = 'choice'
w, = np.where((-500 <= time_a) & (time_a < 0))
else:
raise ValueError(e)
for cond in events_by_cond[ev]:
epochs_by_cond[e][cond] = np.mean(events_by_cond[ev][cond][w],
axis=0)
#=====================================================================================
# Classify units
#=====================================================================================
idpt = indifference_point(behaviorfile, offers)
unit_types = classify_units(trials, perf, r, idpt)
#unit_types = {}
numbers = {}
for v in unit_types.values():
numbers[v] = 0
for k, v in unit_types.items():
numbers[v] += 1
n_tot = np.sum(numbers.values())
for k, v in numbers.items():
print("{}: {}/{} = {}%".format(k, v, n_tot, 100 * v / n_tot))
#=====================================================================================
# Plot
#=====================================================================================
lw = kwargs.get('lw', 1.5)
ms = kwargs.get('ms', 6)
mew = kwargs.get('mew', 0.5)
rotation = kwargs.get('rotation', 60)
#min_trials = kwargs.get('min_trials', 100)
def plot_activity(plot, unit):
yall = [1]
min_trials = 20
# Pre-offer
epoch_by_cond = epochs_by_cond['preoffer']
color = '0.7'
if separate_by_choice:
for choice, marker in zip(['A', 'B'], ['d', 'o']):
x = []
y = []
for i, offer in enumerate(offers):
cond = (offer, choice)
if cond in n_by_cond and n_by_cond[cond] >= min_trials:
y_i = epoch_by_cond[cond][unit]
plot.plot(i,
y_i,
marker,
mfc=color,
mec=color,
ms=0.8 * ms,
mew=0.8 * mew,
zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers) - 1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=0.8 * lw, zorder=5)
else:
x = []
y = []
for i, offer in enumerate(offers):
y_i = epoch_by_cond[offer][unit]
plot.plot(i,
y_i,
'o',
mfc=color,
mec=color,
ms=0.8 * ms,
mew=0.8 * mew,
zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers) - 1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=0.8 * lw, zorder=5)
# Epoch
epoch_by_cond = epochs_by_cond[epoch]
if epoch == 'postoffer':
color = Figure.colors('darkblue')
elif epoch == 'latedelay':
color = Figure.colors('darkblue')
elif epoch == 'prechoice':
color = Figure.colors('darkblue')
else:
raise ValueError(epoch)
if separate_by_choice:
for choice, marker, color in zip(
['A', 'B'], ['d', 'o'],
[Figure.colors('red'),
Figure.colors('blue')]):
x = []
y = []
for i, offer in enumerate(offers):
cond = (offer, choice)
if cond in n_by_cond and n_by_cond[cond] >= min_trials:
y_i = epoch_by_cond[cond][unit]
yall.append(y_i)
plot.plot(i,
y_i,
marker,
mfc=color,
mec=color,
ms=ms,
mew=mew,
zorder=10)
if i != 0 and i != len(offers) - 1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=lw, zorder=5)
else:
x = []
y = []
for i, offer in enumerate(offers):
y_i = epoch_by_cond[offer][unit]
plot.plot(i,
y_i,
'o',
mfc=color,
mec=color,
ms=ms,
mew=mew,
zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers) - 1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=lw, zorder=5)
plot.xticks(range(len(offers)))
plot.xticklabels(['{}B:{}A'.format(*offer) for offer in offers],
rotation=rotation)
plot.xlim(0, len(offers) - 1)
plot.lim('y', yall, lower=0)
return yall
#-------------------------------------------------------------------------------------
if units is not None:
for plot, unit in zip(plots, units):
plot_activity(plot, unit)
else:
name = plots
for unit in xrange(N):
fig = Figure()
plot = fig.add()
plot_activity(plot, unit)
if separate_by_choice:
suffix = '_sbc'
else:
suffix = ''
if unit in unit_types:
plot.text_upper_right(unit_types[unit], fontsize=9)
fig.save(name +
'_{}{}_{}{:03d}'.format(epoch, suffix, network, unit))
fig.close()
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
plot.xlabel('$W$')
#for j in xrange(W.shape[1]):
# print(sum(1*(W[:,j] != 0)))
Wrec_gates = params['Wrec_gates']
if 'Wrec_gates' in masks:
Wrec_gates *= masks['Wrec_gates']
plot = fig['Wrec_lambda']
W = Wrec_gates[:,:N]
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}^\lambda$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
plot = fig['Wrec_gamma']
W = Wrec_gates[:,N:]
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}^\gamma$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
#'''
#=========================================================================================
fig.save(path=figspath, name='fig_weights_'+modelname)
plot.ylabel('Percent correct\n(decision trials)')
target_color = Figure.colors('red')
plot = fig['correct']
if modelname.startswith('rdm_fixed'):
target = 80
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'rdm_rt':
target = 80
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'mante':
target = 85
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'multisensory':
target = 82
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'romo':
target = 97
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'postdecisionwager':
target = 79
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'padoaschioppa2006':
target = 95
plot.hline(target, color=target_color, zorder=1)
#=========================================================================================
fig.save(path=figspath, name='fig_learning_' + modelname)
plot.ylabel('Percent correct\n(decision trials)')
target_color = Figure.colors('red')
plot = fig['correct']
if modelname.startswith('rdm_fixed'):
target = 80
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'rdm_rt':
target = 80
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'mante':
target = 85
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'multisensory':
target = 82
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'romo':
target = 97
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'postdecisionwager':
target = 79
plot.hline(target, color=target_color, zorder=1)
elif modelname == 'padoaschioppa2006':
target = 95
plot.hline(target, color=target_color, zorder=1)
#=========================================================================================
fig.save(path=figspath, name='fig_learning_'+modelname)
def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except IndexError:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
juices = spec.juices
offers = spec.offers
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(juices), len(offers)))
context = {
'juice': juices[k.pop(0)],
'offer': offers[k.pop(0)]
}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'choice_pattern':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
spec = config['model'].spec
choice_pattern(trialsfile, spec.offers, plot)
plot.xlabel('Offer (\#B : \#A)')
plot.ylabel('Percent choice B')
plot.text_upper_left('1A = {}B'.format(spec.A_to_B), fontsize=10)
fig.save(path=config['figspath'], name=action)
fig.close()
elif action == 'indifference_point':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
spec = config['model'].spec
indifference_point(trialsfile, spec.offers, plot)
plot.xlabel('$(n_B - n_A)/(n_B + n_A)$')
plot.ylabel('Percent choice B')
#plot.text_upper_left('1A = {}B'.format(spec.A_to_B), fontsize=10)
fig.save(path=config['figspath'], name=action)
fig.close()
#=====================================================================================
elif action == 'sort_epoch':
behaviorfile = runtools.behaviorfile(config['trialspath'])
activityfile = runtools.activityfile(config['trialspath'])
epoch = args[0]
if 'value' in args:
network = 'v'
else:
network = 'p'
separate_by_choice = ('separate-by-choice' in args)
sort_epoch(behaviorfile, activityfile, epoch, config['model'].spec.offers,
os.path.join(config['figspath'], 'sorted'),
network=network, separate_by_choice=separate_by_choice)
def sort_epoch(behaviorfile, activityfile, epoch, offers, plots, units=None, network='p',
separate_by_choice=False, **kwargs):
"""
Sort trials.
"""
# Load trials
data = utils.load(activityfile)
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
if network == 'p':
print("POLICY NETWORK")
r = r_p
else:
print("VALUE NETWORK")
r = r_v
# Number of units
N = r.shape[-1]
# Same for every trial
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Sort trials
#=====================================================================================
# Epochs
events = ['offer', 'choice']
# Sort
events_by_cond = {e: {} for e in events}
n_by_cond = {}
n_nondecision = 0
for n, trial in enumerate(trials):
if perf.choices[n] is None:
n_nondecision += 1
continue
# Condition
offer = trial['offer']
choice = perf.choices[n]
if separate_by_choice:
cond = (offer, choice)
else:
cond = offer
n_by_cond.setdefault(cond, 0)
n_by_cond[cond] += 1
# Storage
for e in events_by_cond:
events_by_cond[e].setdefault(cond, {'r': np.zeros((Ntime_a, N)),
'n': np.zeros((Ntime_a, N))})
# Firing rates
m_n = np.tile(M[:,n], (N,1)).T
r_n = r[:,n]*m_n
for e in events_by_cond:
# Align point
if e == 'offer':
t0 = trial['epochs']['offer-on'][0]
elif e == 'choice':
t0 = perf.t_choices[n]
else:
raise ValueError(e)
# Before
n_b = r_n[:t0].shape[0]
events_by_cond[e][cond]['r'][Ntime-1-n_b:Ntime-1] += r_n[:t0]
events_by_cond[e][cond]['n'][Ntime-1-n_b:Ntime-1] += m_n[:t0]
# After
n_a = r_n[t0:].shape[0]
events_by_cond[e][cond]['r'][Ntime-1:Ntime-1+n_a] += r_n[t0:]
events_by_cond[e][cond]['n'][Ntime-1:Ntime-1+n_a] += m_n[t0:]
print("Non-decision trials: {}/{}".format(n_nondecision, len(trials)))
# Average trials
for e in events_by_cond:
for cond in events_by_cond[e]:
events_by_cond[e][cond] = utils.div(events_by_cond[e][cond]['r'],
events_by_cond[e][cond]['n'])
# Epochs
epochs = ['preoffer', 'postoffer', 'latedelay', 'prechoice']
# Average epochs
epochs_by_cond = {e: {} for e in epochs}
for e in epochs_by_cond:
if e == 'preoffer':
ev = 'offer'
w, = np.where((-500 <= time_a) & (time_a < 0))
elif e == 'postoffer':
ev = 'offer'
w, = np.where((0 <= time_a) & (time_a < 500))
elif e == 'latedelay':
ev = 'offer'
w, = np.where((500 <= time_a) & (time_a < 1000))
elif e == 'prechoice':
ev = 'choice'
w, = np.where((-500 <= time_a) & (time_a < 0))
else:
raise ValueError(e)
for cond in events_by_cond[ev]:
epochs_by_cond[e][cond] = np.mean(events_by_cond[ev][cond][w], axis=0)
#=====================================================================================
# Classify units
#=====================================================================================
idpt = indifference_point(behaviorfile, offers)
unit_types = classify_units(trials, perf, r, idpt)
#unit_types = {}
numbers = {}
for v in unit_types.values():
numbers[v] = 0
for k, v in unit_types.items():
numbers[v] += 1
n_tot = np.sum(numbers.values())
for k, v in numbers.items():
print("{}: {}/{} = {}%".format(k, v, n_tot, 100*v/n_tot))
#=====================================================================================
# Plot
#=====================================================================================
lw = kwargs.get('lw', 1.5)
ms = kwargs.get('ms', 6)
mew = kwargs.get('mew', 0.5)
rotation = kwargs.get('rotation', 60)
#min_trials = kwargs.get('min_trials', 100)
def plot_activity(plot, unit):
yall = [1]
min_trials = 20
# Pre-offer
epoch_by_cond = epochs_by_cond['preoffer']
color = '0.7'
if separate_by_choice:
for choice, marker in zip(['A', 'B'], ['d', 'o']):
x = []
y = []
for i, offer in enumerate(offers):
cond = (offer, choice)
if cond in n_by_cond and n_by_cond[cond] >= min_trials:
y_i = epoch_by_cond[cond][unit]
plot.plot(i, y_i, marker, mfc=color, mec=color, ms=0.8*ms,
mew=0.8*mew, zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers)-1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=0.8*lw, zorder=5)
else:
x = []
y = []
for i, offer in enumerate(offers):
y_i = epoch_by_cond[offer][unit]
plot.plot(i, y_i, 'o', mfc=color, mec=color, ms=0.8*ms,
mew=0.8*mew, zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers)-1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=0.8*lw, zorder=5)
# Epoch
epoch_by_cond = epochs_by_cond[epoch]
if epoch == 'postoffer':
color = Figure.colors('darkblue')
elif epoch == 'latedelay':
color = Figure.colors('darkblue')
elif epoch == 'prechoice':
color = Figure.colors('darkblue')
else:
raise ValueError(epoch)
if separate_by_choice:
for choice, marker, color in zip(['A', 'B'], ['d', 'o'], [Figure.colors('red'), Figure.colors('blue')]):
x = []
y = []
for i, offer in enumerate(offers):
cond = (offer, choice)
if cond in n_by_cond and n_by_cond[cond] >= min_trials:
y_i = epoch_by_cond[cond][unit]
yall.append(y_i)
plot.plot(i, y_i, marker, mfc=color, mec=color, ms=ms, mew=mew, zorder=10)
if i != 0 and i != len(offers)-1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=lw, zorder=5)
else:
x = []
y = []
for i, offer in enumerate(offers):
y_i = epoch_by_cond[offer][unit]
plot.plot(i, y_i, 'o', mfc=color, mec=color, ms=ms, mew=mew, zorder=10)
yall.append(y_i)
if i != 0 and i != len(offers)-1:
x.append(i)
y.append(y_i)
plot.plot(x, y, '-', color=color, lw=lw, zorder=5)
plot.xticks(range(len(offers)))
plot.xticklabels(['{}B:{}A'.format(*offer) for offer in offers],
rotation=rotation)
plot.xlim(0, len(offers)-1)
plot.lim('y', yall, lower=0)
return yall
#-------------------------------------------------------------------------------------
if units is not None:
for plot, unit in zip(plots, units):
plot_activity(plot, unit)
else:
name = plots
for unit in xrange(N):
fig = Figure()
plot = fig.add()
plot_activity(plot, unit)
if separate_by_choice:
suffix = '_sbc'
else:
suffix = ''
if unit in unit_types:
plot.text_upper_right(unit_types[unit], fontsize=9)
fig.save(name+'_{}{}_{}{:03d}'.format(epoch, suffix, network, unit))
fig.close()
def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except IndexError:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
juices = spec.juices
offers = spec.offers
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(juices), len(offers)))
context = {'juice': juices[k.pop(0)], 'offer': offers[k.pop(0)]}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'choice_pattern':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
spec = config['model'].spec
choice_pattern(trialsfile, spec.offers, plot)
plot.xlabel('Offer (\#B : \#A)')
plot.ylabel('Percent choice B')
plot.text_upper_left('1A = {}B'.format(spec.A_to_B), fontsize=10)
fig.save(path=config['figspath'], name=action)
fig.close()
elif action == 'indifference_point':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
spec = config['model'].spec
indifference_point(trialsfile, spec.offers, plot)
plot.xlabel('$(n_B - n_A)/(n_B + n_A)$')
plot.ylabel('Percent choice B')
#plot.text_upper_left('1A = {}B'.format(spec.A_to_B), fontsize=10)
fig.save(path=config['figspath'], name=action)
fig.close()
#=====================================================================================
elif action == 'sort_epoch':
behaviorfile = runtools.behaviorfile(config['trialspath'])
activityfile = runtools.activityfile(config['trialspath'])
epoch = args[0]
if 'value' in args:
network = 'v'
else:
network = 'p'
separate_by_choice = ('separate-by-choice' in args)
sort_epoch(behaviorfile,
activityfile,
epoch,
config['model'].spec.offers,
os.path.join(config['figspath'], 'sorted'),
network=network,
separate_by_choice=separate_by_choice)
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
plot.xlabel('$W$')
#for j in xrange(W.shape[1]):
# print(sum(1*(W[:,j] != 0)))
Wrec_gates = params['Wrec_gates']
if 'Wrec_gates' in masks:
Wrec_gates *= masks['Wrec_gates']
plot = fig['Wrec_lambda']
W = Wrec_gates[:, :N]
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}^\lambda$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
plot = fig['Wrec_gamma']
W = Wrec_gates[:, N:]
rho = matrixtools.spectral_radius(W)
plot_weights(plot, W)
plot.text_upper_left(r'$W_\text{rec}^\gamma$', fontsize=fontsize, dy=dy)
plot.text_upper_right(r'$\rho={:.3f}$'.format(rho), fontsize=fontsize, dy=dy)
#'''
#=========================================================================================
fig.save(path=figspath, name='fig_weights_' + modelname)
def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if action == 'plot_trial':
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 1000
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
juices = spec.juices
offers = spec.offers
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
fig = Figure(axislabelsize=10, ticklabelsize=9)
plot = fig.add()
plot_trial()
performance(config['savefile'], plot)
fig.save(path=config['figspath'], name='performance')
fig.close()
#=====================================================================================
elif 'trials' in action:
try:
trials_per_condition = int(args[0])
except IndexError:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
juices = spec.juices
offers = spec.offers
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(juices), len(offers)))
context = {'juice': juices[k.pop(0)], 'offer': offers[k.pop(0)]}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'choice_pattern':
trialsfile = runtools.behaviorfile(config['trialspath'])
# print trialsfile
# fig = Figure()
# plot = fig.add()
savefile = config['figspath']
spec = config['model'].spec
#print spec.offers
choice_pattern(trialsfile, spec.offers, savefile, action)
#plot.xlabel('Offer (\#B : \#A)')
#plot.ylabel('Percent choice B')
#plot.text_upper_left('1A = {}B'.format(spec.A_to_B), fontsize=10)
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, (config['figspath'], 'sorted'), network=network)
#=====================================================================================
elif action == 'statespace':
trialsfile = runtools.activityfile(config['trialspath'])
statespace(trialsfile, (config['figspath'], 'statespace'))
plot.xlim(-200, 600)
plot.xticks([-200, 0, 200, 400, 600])
plot.ylim(0.5, 1.2)
#plot.yticks([0.5, 1])
plot.xlabel('Time from stimulus (ms)')
plot.ylabel('Expected reward')
# Legend
props = {
'prop': {
'size': 8
},
'handlelength': 1.2,
'handletextpad': 1.1,
'labelspacing': 0.7
}
plot.legend(bbox_to_anchor=(0.33, 1), **props)
plot = fig['on-choice']
plot.xlim(-400, 0)
plot.xticks([-400, -200, 0])
plot.ylim(0.5, 1.2)
#plot.yticks([0.5, 1])
plot.xlabel('Time from decision (ms)')
#=========================================================================================
fig.save()
def do(action, args, config):
"""
Manage tasks.
"""
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
gt_lts = spec.gt_lts
fpairs = spec.fpairs
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(gt_lts), len(fpairs)))
context = {
'delay': 3000,
'gt_lt': gt_lts[k.pop(0)],
'fpair': fpairs[k.pop(0)]
}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'performance':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
performance(trialsfile, plot)
plot.xlabel('$f_1$ (Hz)')
plot.ylabel('$f_2$ (Hz)')
fig.save(os.path.join(config['figspath'], action))
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, (config['figspath'], 'sorted'), network=network)
def do(action, args, config):
print("ACTION*: " + str(action))
print("ARGS*: " + str(args))
#=====================================================================================
if 'trials' in action:
try:
trials_per_condition = int(args[0])
except:
trials_per_condition = 100
model = config['model']
pg = model.get_pg(config['savefile'], config['seed'], config['dt'])
spec = model.spec
wagers = spec.wagers
left_rights = spec.left_rights
cohs = spec.cohs
n_conditions = spec.n_conditions
n_trials = trials_per_condition * n_conditions
print("{} trials".format(n_trials))
task = model.Task()
trials = []
for n in xrange(n_trials):
k = tasktools.unravel_index(n, (len(wagers), len(left_rights), len(cohs)))
context = {
'wager': wagers[k.pop(0)],
'left_right': left_rights[k.pop(0)],
'coh': cohs[k.pop(0)]
}
trials.append(task.get_condition(pg.rng, pg.dt, context))
runtools.run(action, trials, pg, config['trialspath'])
#=====================================================================================
elif action == 'sure_stimulus_duration':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
sure_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Probability sure target')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'correct_stimulus_duration':
trialsfile = runtools.behaviorfile(config['trialspath'])
fig = Figure()
plot = fig.add()
correct_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Probability correct')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'value_stimulus_duration':
trialsfile = runtools.activityfile(config['trialspath'])
fig = Figure()
plot = fig.add()
value_stimulus_duration(trialsfile, plot)
plot.xlabel('Stimulus duration (ms)')
plot.ylabel('Expected reward')
fig.save(path=config['figspath'], name=action)
#=====================================================================================
elif action == 'sort':
if 'value' in args:
network = 'v'
else:
network = 'p'
trialsfile = runtools.activityfile(config['trialspath'])
sort(trialsfile, os.path.join(config['figspath'], 'sorted'), network=network)
def sort(trialsfile, plots, units=None, network='p', **kwargs):
"""
Sort trials.
"""
# Load trials
data = utils.load(trialsfile)
trials, U, Z, Z_b, A, P, M, perf, r_p, r_v = data
# Which network?
if network == 'p':
r = r_p
else:
r = r_v
# Number of units
N = r.shape[-1]
# Same for every trial
time = trials[0]['time']
Ntime = len(time)
# Aligned time
time_a = np.concatenate((-time[1:][::-1], time))
Ntime_a = len(time_a)
#=====================================================================================
# Aligned to stimulus onset
#=====================================================================================
r_by_cond_stimulus = {}
n_r_by_cond_stimulus = {}
for n, trial in enumerate(trials):
if not perf.decisions[n]:
continue
if trial['mod'] == 'va':
continue
assert trial['mod'] == 'v' or trial['mod'] == 'a'
if not perf.corrects[n]:
continue
# Condition
mod = trial['mod']
choice = perf.choices[n]
cond = (mod, choice)
# Storage
r_by_cond_stimulus.setdefault(cond, np.zeros((Ntime_a, N)))
n_r_by_cond_stimulus.setdefault(cond, np.zeros((Ntime_a, N)))
# Firing rates
Mn = np.tile(M[:,n], (N,1)).T
Rn = r[:,n]*Mn
# Align point
t0 = trial['epochs']['stimulus'][0] - 1
# Before
n_b = Rn[:t0].shape[0]
r_by_cond_stimulus[cond][Ntime-1-n_b:Ntime-1] += Rn[:t0]
n_r_by_cond_stimulus[cond][Ntime-1-n_b:Ntime-1] += Mn[:t0]
# After
n_a = Rn[t0:].shape[0]
r_by_cond_stimulus[cond][Ntime-1:Ntime-1+n_a] += Rn[t0:]
n_r_by_cond_stimulus[cond][Ntime-1:Ntime-1+n_a] += Mn[t0:]
for cond in r_by_cond_stimulus:
r_by_cond_stimulus[cond] = utils.div(r_by_cond_stimulus[cond],
n_r_by_cond_stimulus[cond])
#-------------------------------------------------------------------------------------
# Plot
#-------------------------------------------------------------------------------------
lw = kwargs.get('lw', 1.5)
dashes = kwargs.get('dashes', [3, 2])
vline_props = {'lw': kwargs.get('lw_vline', 0.5)}
if 'dashes_vline' in kwargs:
vline_props['linestyle'] = '--'
vline_props['dashes'] = dashes
colors_by_mod = {
'v': Figure.colors('blue'),
'a': Figure.colors('green')
}
linestyle_by_choice = {
'L': '-',
'H': '--'
}
lineprops = dict(lw=lw)
def plot_sorted(plot, unit, w, r_sorted):
t = time_a[w]
yall = [[1]]
for cond in [('v', 'H'), ('v', 'L'), ('a', 'H'), ('a', 'L')]:
mod, choice = cond
if mod == 'v':
label = 'Vis, '
elif mod == 'a':
label = 'Aud, '
else:
raise ValueError(mod)
if choice == 'H':
label += 'high'
elif choice == 'L':
label += 'low'
else:
raise ValueError(choice)
linestyle = linestyle_by_choice[choice]
if linestyle == '-':
lineprops = dict(linestyle=linestyle, lw=lw)
else:
lineprops = dict(linestyle=linestyle, lw=lw, dashes=dashes)
plot.plot(t, r_sorted[cond][w,unit],
color=colors_by_mod[mod],
label=label,
**lineprops)
yall.append(r_sorted[cond][w,unit])
return t, yall
def on_stimulus(plot, unit):
w, = np.where((time_a >= -300) & (time_a <= 1000))
t, yall = plot_sorted(plot, unit, w, r_by_cond_stimulus)
plot.xlim(t[0], t[-1])
return yall
if units is not None:
for plot, unit in zip(plots, units):
on_stimulus(plot, unit)
else:
figspath, name = plots
for unit in xrange(N):
fig = Figure()
plot = fig.add()
#-----------------------------------------------------------------------------
yall = []
yall += on_stimulus(plot, unit)
plot.lim('y', yall, lower=0)
plot.vline(0)
plot.xlabel('Time (ms)')
plot.ylabel('Firing rate (a.u.)')
#-----------------------------------------------------------------------------
fig.save(path=figspath, name=name+'_{}{:03d}'.format(network, unit))
fig.close()
