def make_weight_residual_plot(N, dt, s_infs_mcmc, s_trues, s_infs_map=None, resdir='.'):
# Scatter plot the average weights vs true weights
fig = plt.figure(figsize=(2.5,2.5))
ax = fig.gca()
for d,s_inf_mcmc in enumerate(s_infs_mcmc):
s_avg = average_list_of_dicts(s_inf_mcmc)
label = None if d >0 else "N-GLM"
scatter_plot_weight_residuals(dt, s_avg, s_trues[d], ax=ax, sz=20, color='r', label=label)
if s_infs_map is not None:
for d, s_inf_map in enumerate(s_infs_map):
label = None if d >0 else "L1-GLM"
scatter_plot_weight_residuals(dt, s_inf_map, s_trues[d], ax=ax, sz=15, color='b', label=label)
# Plot identity line
xlim = ax.get_xlim()
ax.plot(xlim, xlim, '-k', linewidth=0.5)
ax.legend(loc='upper left', prop={'size' : 8}, scatterpoints=1)
# Make room for the axis labels
plt.subplots_adjust(left=0.25, bottom=0.25)
# Save and close
fig.savefig(os.path.join(resdir, 'W_resid.pdf'))
plt.close(fig)
def make_weight_residual_plot(N,
dt,
s_infs_mcmc,
s_trues,
s_infs_map=None,
resdir='.'):
# Scatter plot the average weights vs true weights
fig = plt.figure(figsize=(2.5, 2.5))
ax = fig.gca()
for d, s_inf_mcmc in enumerate(s_infs_mcmc):
s_avg = average_list_of_dicts(s_inf_mcmc)
label = None if d > 0 else "N-GLM"
scatter_plot_weight_residuals(dt,
s_avg,
s_trues[d],
ax=ax,
sz=20,
color='r',
label=label)
if s_infs_map is not None:
for d, s_inf_map in enumerate(s_infs_map):
label = None if d > 0 else "L1-GLM"
scatter_plot_weight_residuals(dt,
s_inf_map,
s_trues[d],
ax=ax,
sz=15,
color='b',
label=label)
# Plot identity line
xlim = ax.get_xlim()
ax.plot(xlim, xlim, '-k', linewidth=0.5)
ax.legend(loc='upper left', prop={'size': 8}, scatterpoints=1)
# Make room for the axis labels
plt.subplots_adjust(left=0.25, bottom=0.25)
# Save and close
fig.savefig(os.path.join(resdir, 'W_resid.pdf'))
plt.close(fig)
def plot_results(population,
x_inf,
popn_true=None,
x_true=None,
resdir=None,
do_plot_connectivity=True,
do_plot_stim_resp=True,
do_plot_imp_responses=True,
do_plot_firing_rates=True,
do_plot_ks=True,
do_plot_logpr=True):
""" Plot the inferred stimulus tuning curves and impulse responses
"""
if not resdir:
resdir = '.'
true_given = x_true is not None and popn_true is not None
# Make sure we have a list of x's
if not isinstance(x_inf, list):
x_inf = [x_inf]
# Evaluate the state for each of the parameter settings
N_samples = len(x_inf)
s_inf = []
for x in x_inf:
s_inf.append(population.eval_state(x))
s_true = None
if true_given:
s_true = popn_true.eval_state(x_true)
# Average the inferred states
s_avg = average_list_of_dicts(s_inf)
s_std = std_list_of_dicts(s_inf, s_avg)
N = population.N
# TODO Fix the averaging of W and A
# E[W] * E[A] != E[W*A]
# Plot the inferred connectivity matrix
if do_plot_connectivity:
print "Plotting connectivity matrix"
f = plt.figure()
plot_connectivity_matrix(s_inf, s_true)
f.savefig(os.path.join(resdir,'conn.pdf'))
plt.close(f)
if 'location_provider' in s_inf[0]['latent']:
f = plt.figure()
plot_locations(s_inf, color='r')
if true_given:
plot_locations([s_true], color='k')
f.savefig(os.path.join(resdir, 'locations.pdf'))
plt.close(f)
# Plot shared tuning curves
if 'sharedtuningcurve_provider' in s_inf[0]['latent']:
print "Plotting shared tuning curves"
for n in range(N):
f = plt.figure()
if true_given:
plot_spatiotemporal_tuning_curves(
s_avg,
s_true=s_true,
s_std=s_std,
color='r')
else:
plot_spatiotemporal_tuning_curves(
s_avg,
s_std=s_std,
color='k')
f.savefig(os.path.join(resdir,'tuning_curves.pdf'))
plt.close(f)
print "Plotting types"
f = plt.figure()
plot_latent_types(s_inf, s_true)
f.savefig(os.path.join(resdir, 'latent_types.pdf'))
plt.close(f)
# Plot stimulus response functions
if do_plot_stim_resp:
print "Plotting stimulus response functions"
for n in range(N):
f = plt.figure()
plot_stim_response(s_avg['glms'][n],
s_glm_std=s_std['glms'][n],
color='r')
if true_given:
plot_stim_response(s_true['glms'][n],
color='k')
f.savefig(os.path.join(resdir,'stim_resp_%d.pdf' % n))
plt.close(f)
# Plot the impulse responses
if do_plot_imp_responses:
print "Plotting impulse response functions"
f = plt.figure()
plot_imp_responses(s_avg,
s_std,
fig=f,
color='r',
use_bgcolor=True)
if true_given:
plot_imp_responses(s_true,
fig=f,
color='k',
linestyle='--',
use_bgcolor=False)
f.savefig(os.path.join(resdir,'imp_resp.pdf'))
plt.close(f)
# Plot the impulse response basis
if do_plot_imp_responses:
f = plt.figure()
plot_basis(s_avg)
f.savefig(os.path.join(resdir,'imp_basis.pdf'))
plt.close(f)
# Plot the firing rates
if do_plot_firing_rates:
print "Plotting firing rates"
T_lim = slice(0,2000)
for n in range(N):
f = plt.figure()
plot_firing_rate(s_avg['glms'][n],
s_std['glms'][n],
color='r',
T_lim=T_lim)
if true_given:
plot_firing_rate(s_true['glms'][n], color='k', T_lim=T_lim)
# Plot the spike times
St = np.nonzero(population.glm.S.get_value()[T_lim,n])[0]
plt.plot(St,s_avg['glms'][n]['lam'][T_lim][St],'ko')
plt.title('Firing rate %d' % n)
f.savefig(os.path.join(resdir,'firing_rate_%d.pdf' % n))
plt.close(f)
if do_plot_ks:
print "Plotting KS test results"
for n in range(N):
f = plt.figure()
St = np.nonzero(population.glm.S.get_value()[:,n])[0]
plot_ks(s_avg['glms'][n], St, population.glm.dt.get_value())
f.savefig(os.path.join(resdir, 'ks_%d.pdf' %n))
plt.close(f)
if do_plot_logpr:
print "Plotting log probability and log likelihood trace"
f = plt.figure()
plot_log_prob(s_inf, s_true=s_true, color='r')
f.savefig(os.path.join(resdir, 'log_prob.pdf'))
plt.close(f)
f = plt.figure()
plot_log_lkhd(s_inf, s_true=s_true, color='r')
f.savefig(os.path.join(resdir, 'log_lkhd.pdf'))
plt.close(f)
if 'logprior' in s_inf[0]:
f = plt.figure()
plot_log_prob(s_inf, key='logprior', s_true=s_true, color='r')
plt.ylabel('Log prior')
f.savefig(os.path.join(resdir, 'log_prior.pdf'))
plt.close(f)
if 'predll' in x_inf[0]:
f = plt.figure()
plot_log_prob(x_inf, key='predll', s_true=x_true, color='r')
plt.ylabel('Pred. Log Likelihood')
f.savefig(os.path.join(resdir, 'pred_ll.pdf'))
plt.close(f)
print "Plots can be found in directory: %s" % resdir
