x0 = model_output.x
f2, ax2 = plt.subplots(1, 2)
best = 0
for k in results.keys():
if results[k] is not None:
if results[k]['cc'].mean() > best:
best = results[k]['cc'].mean()
best_alpha = k
ax2[0].plot(k, results[k]['cc'].mean(), 'ko')
ax2[1].plot(k, results[k]['mse'].mean(), 'ro')
if best_alpha == 0:
best_alpha = alpha1[1]
ax2[0].set_ylabel('prediction correlation')
ax2[0].set_xlabel('pupil constraint')
ax2[0].set_aspect(cplt.get_square_asp(ax2[0]))
ax2[1].set_ylabel('NMSE')
ax2[1].set_xlabel('pupil constraint')
ax2[1].set_aspect(cplt.get_square_asp(ax2[1]))
f, ax = plt.subplots(4, 1)
ax[0].set_title('Simulated data')
ax[0].imshow(resp, aspect='auto')
ax[2].set_title('True state-variables')
ax[2].plot(lv1.T, color='purple', label='LV')
ax[2].plot(pupil.T, color='green', label='pupil')
ax[2].legend()
tar_resp = r['pop_psth'].extract_epoch(tar)[0, :, idx].squeeze()
tar_sem = r['pop_psth_sem'].extract_epoch(tar)[0, :, idx].squeeze()
ax[i].plot(time, tar_resp, label=tar)
ax[i].fill_between(time,
tar_resp - tar_sem,
tar_resp + tar_sem,
alpha=0.5,
lw=0)
except:
ax[i].plot(time, np.nan * np.ones(len(time)))
ax[i].fill_between(time,
np.nan * np.ones(len(time)),
np.nan * np.ones(len(time)),
alpha=0.5,
lw=0)
# figure out onset / offset bin
ax[i].axvline(onset, color='lightgrey', linestyle='--')
ax[i].axvline(offset, color='lightgrey', linestyle='--')
# add plot labels set lims
ax[i].legend(fontsize=8)
ax[i].set_xlabel('Time (s)', fontsize=8)
if i == 0:
ax[i].set_ylabel('Norm. Response')
ax[i].set_ylim((0, ylim))
ax[i].set_aspect(cplt.get_square_asp(ax[i]))
fig.tight_layout()
plt.show()
results['dc'] = d1
results['baseline'] = b
# create plots to evaluate outcome of fits
plt.figure(figsize=(6, 4))
ax2 = plt.subplot2grid((1, 2), (0, 0), rowspan=1, colspan=1)
ax3 = plt.subplot2grid((1, 2), (0, 1), rowspan=1, colspan=1)
# plot the model weights
ax2.scatter(results['gain'], results['dc'], s=25, color='k', edgecolor='white')
ax2.set_xlabel('gain weights', fontsize=8)
ax2.set_ylabel('DC weights', fontsize=8)
ax2.axhline(0, color='k', linestyle='--')
ax2.axvline(0, color='k', linestyle='--')
ax2.set_aspect(cplt.get_square_asp(ax2))
# plot the model performance
null_cc = glm.corrcoef_by_neuron(rec['resp']._data, rec['psth']._data)
ax3.scatter(null_cc, results['cc'], s=25, color='k', edgecolor='white')
ax3.plot([0, 1], [0, 1], 'k--')
ax3.axhline(0, color='k', linestyle='--')
ax3.axvline(0, color='k', linestyle='--')
ax3.set_xlabel('Null model (r0)', fontsize=8)
ax3.set_ylabel('Full model', fontsize=8)
ax3.set_aspect(cplt.get_square_asp(ax3))
plt.tight_layout()
# perform regression of different factors and determine how noise correlations change
r_true = rec['resp']._data
cmap='Purples',
c=rec['pupil']._data.squeeze())
ax[3].set_title('second-order pupil lv')
f.tight_layout()
# Look at model weights and prediction coef for each neuron
# and noise correlations before / after
f, ax = plt.subplots(2, 2)
ax[0, 0].scatter(w1, w2, s=25, color='k', edgecolor='white')
ax[0, 0].axhline(0, linestyle='--', color='k')
ax[0, 0].axvline(0, linestyle='--', color='k')
ax[0, 0].set_xlabel('first-order weights')
ax[0, 0].set_ylabel('second-order weights')
ax[0, 0].set_aspect(cplt.get_square_asp(ax[0, 0]))
null_cc = glm.corrcoef_by_neuron(rec['resp']._data, rec['psth']._data)
pred_cc = glm.corrcoef_by_neuron(rec['resp']._data, pred)
first_cc = glm.corrcoef_by_neuron(rec['resp']._data, pred2)
ax[0, 1].scatter(null_cc, pred_cc, s=25, color='k', edgecolor='white')
ax[0, 1].plot([0, 1], [0, 1], 'k--')
ax[0, 1].set_xlabel('null model (r0)')
ax[0, 1].set_ylabel('full model')
ax[0, 1].set_aspect(cplt.get_square_asp(ax[0, 1]))
# full model vs. first order only model
ax[1, 1].scatter(null_cc, first_cc, s=25, color='k', edgecolor='white')
ax[1, 1].plot([0, 1], [0, 1], 'k--')
ax[1, 1].set_xlabel('null model model')
